Molecular signatures have begun to elucidate the biological and molecular mechanisms underlying the phenotypic diversity of breast tumors. Breast tumors are characterized by five different molecular subtypes that are associated with distinct clinical outcomes in terms of prognosis, treatment response, and site of relapse. In particular, the basal-like and luminal B subtypes of tumors are more aggressive and have a higher tendency to metastasize to the lung than do the other subtypes. Given this difference in metastatic profiles of breast tumors, the six-gene signature (6GS) that we showed to be predictive of lung relapse was reexamined in the context of the tumor subtypes. This first analysis suggested that the 6GS is a surrogate for molecular subtype, discriminating basal-like tumors rather than tumors that metastasize to the lung. Here, we show that the 6GS discriminates the two overlapping features, the basal-like subtype and the tendency to metastasize to the lung. Nevertheless, the 6GS predicts lung metastases of breast tumors independent of the molecular subtypes. [Cancer Res 2009;69(24):9507–11]

Although a coherent picture has begun to emerge about the biological and molecular mechanisms of primary breast tumorigenesis, the processes that subsequently lead to invasion and metastasis have, until recently, remained relatively obscure. Direct monitoring of the metastatic process using clinical samples is barely feasible, and experimental models mimicking metastasis are limited with respect to several characteristics (including the lack of primary tumor heterogeneity and diversity) and thus likely represent a partial view of the process.

Nevertheless, several studies have addressed this crucial question leading to new insights in the field. Kang and colleagues first showed that metastasis to specific organs might require a set of functions beyond those underlying the emergence of the primary tumor. Indeed, their group identified genes specifically mediating metastasis to bone or lung using a model of MDA-MB-231–derived breast cancer cell lines with specific organ tropisms. Importantly, they characterized a “lung metastasis signature” (LMS) predictive of a high risk for developing lung metastases in human breast tumors (13).

With an alternative strategy, we used metastatic human samples rather than human cell clones selected in mice to identify genes involved in breast cancer metastasis to the lungs (4). We identified a “six-gene signature” (6GS) predicting lung relapse of breast cancer patients. Although these two lung metastasis signatures (LMS and 6GS) did not show any gene overlap, they presented highly concordant predictions, suggesting similar molecular pathways (4).

On the other hand, given the phenotypic diversity of breast tumors and the corresponding differences in prognosis and treatment response (57), Smid and colleagues investigated whether breast tumor molecular subtypes have clinical relevance with regard to the site of distant metastasis (8). They showed that subtypes of breast cancer show preferential sites of relapse. In particular, lung metastases are mainly generated by basal-like and luminal B tumors.

More recently, Culhane and colleagues suggested that our 6GS is mainly indicating basal-like tumors rather than identifying the site of distant metastasis independent of molecular subtype (9). In the present work, we attempted to determine whether the 6GS discriminates the tumors on the basis of their basal-like molecular portrait or on their outcome with regard to lung relapse, or both. Indeed, our results suggest that the 6GS is a valuable signature in predicting lung metastasis regardless of the subtype of the tumor of origin.

All of the samples analyzed in this report were studied in previously published reports (4, 811). Briefly, three series of primary breast tumors were examined: MSK (accession no. GSE2603), EMC (accession no. GSE2034 and GSE5327), and NKI cohorts (11).

The molecular profiles were assigned according to the expression of the intrinsic signature as previously reported by Perou and colleagues (5). The hierarchical clustering was done using either the list of all 427 genes (705 probes) of the primarily defined signature (6) or the top 66% variable probe sets (i.e., 451 probes) as described by Smid and colleagues (8).

Survival times with respect to lung metastasis were estimated by the Kaplan-Meier method. The log-rank test was used to determine the significance of differences in the 6GS expression in the three independent series. When testing the combined series, we tested a range of cutoff points (from 50th to 85th percentiles) to categorize patients into high-risk or low-risk group as previously described (4). The cutoff was set at the 70th percentile in the case of luminal B tumors and at the 85th percentile for basal tumors. Multivariate analyses were done using the Cox proportional hazards regression model.

Breast tumor subtypes and lung metastases clustered on the basis of the 6GS

The analyses were done on three independent series of primary breast tumors. The MSK series consists of locally advanced tumors, whereas the NKI and EMC series consist of early-stage tumors.

Figure 1 represents the distribution of the subtypes within the three tested series and the proportion of lung relapse among the subtypes. The MSK tumors differed from those in the other groups, as follows: The percentages of basal subtypes in the MSK, NKI, and EMC groups were 34%, 15%, and 25%, respectively. Among the basal subtypes in each cohort, the percentages of lung metastases were 43%, 15%, and 16%, respectively. In the MSK series, none of the luminal B tumors developed a lung relapse (82% of the lung relapses were of the basal-like subtype).

Figure 1.

Distribution of the five molecular subtypes within three series of primary breast tumors and the relative proportion of lung metastases among the subtypes. The MSK series (n = 82) show a different profile of subtypes and associated lung metastases as compared with the NKI (n = 295) and EMC (n = 344) series.

Figure 1.

Distribution of the five molecular subtypes within three series of primary breast tumors and the relative proportion of lung metastases among the subtypes. The MSK series (n = 82) show a different profile of subtypes and associated lung metastases as compared with the NKI (n = 295) and EMC (n = 344) series.

Close modal

To assess whether the 6GS provided additional information beyond the basal-like subtype of the tumor of origin, all three series were clustered using the 6GS (only five genes for NKI; Fig. 2). As previously shown (4), the hierarchical clustering of the MSK series yielded two major clusters. The C2 cluster included not only the majority (78%) of tumors relapsing to the lung but also the majority (78%) of the basal-like tumors (Fig. 2A). Because, in this series, most of the lung metastases are raised from basal tumors, it is not possible to assess the 6GS independently of the basal subtype.

Figure 2.

The hierarchical clustering of the MSK, NKI, and EMC series of breast tumors shows that the 6GS clustered the primary breast tumors both on their basal-like subtypes and on their lung metastatic abilities independently of the molecular subtype. Colored bars represent the different molecular subtypes of each breast tumor: red, basal; light blue, luminal B; dark blue, luminal A; green, erbb2; orange, normal-like. Triangles, lung metastasis colored as the subtype of the tumor of origin.

Figure 2.

The hierarchical clustering of the MSK, NKI, and EMC series of breast tumors shows that the 6GS clustered the primary breast tumors both on their basal-like subtypes and on their lung metastatic abilities independently of the molecular subtype. Colored bars represent the different molecular subtypes of each breast tumor: red, basal; light blue, luminal B; dark blue, luminal A; green, erbb2; orange, normal-like. Triangles, lung metastasis colored as the subtype of the tumor of origin.

Close modal

The hierarchical clustering of NKI (n = 295) also yielded two major clusters. The C2 cluster included the majority (87%) of the basal-like tumors together with the majority of tumors relapsing to the lung regardless of their molecular subtype (85%; Fig. 2B). This result suggests that the 6GS is characterized by two overlapping features: the basal subtype of tumors, which might be an intrinsic characteristic of the lung metastases, and the tendency to metastasize to the lung.

The hierarchical clustering of the EMC series showed a more complex pattern (Fig. 2C). There are two major clusters, with cluster C2 subdivided into six subgroups (C2a–C2f). Both the C1 and C2 clusters have comparable numbers of basal-like tumors (21% and 26%, respectively). However, clusters C1 and C2f at the opposite sides of the dendrogram contain the majority of the lung-relapsing tumors (42% and 29%, respectively) and a higher proportion of basal tumors (21% and 38%, respectively). Cluster C2f has a higher proportion of metastatic tumors of the basal subtype, whereas cluster C1, which shows the highest level of the 6GS, consists mainly of relapsing tumors of the luminal subtype.

The 6GS prediction of lung metastases with respect to molecular subtypes

The 6GS was primarily developed on a series of breast cancer patients that had not received neoadjuvant or adjuvant therapy, meaning that the potential prognostic effect of the “lung metastasis classifier” would not be influenced by factors related to systemic treatment (4). The 6GS was then validated in data sets presenting different treatment status; in particular, all 344 patients of the EMC series did not receive any adjuvant systemic treatment, whereas patients from the NKI series had undergone several treatment categories (56% of patients received no adjuvant treatment, 7% received hormonal treatment only, 30.5% had chemotherapy only, and 7% had both). To asses that the treatment issue would not be involved in the lung metastasis–free survival of the breast cancer subtypes, we verified that the treatment status did not differ between the luminal B and the basal-like subtypes in the NKI cohort. We found that both subgroups had similar treatment distribution as the whole series.

Then, we evaluated the ability of the 6GS versus the basal-like subtype to predict lung metastasis, using the Kaplan-Meier method. The patients with tumors of the basal-like subtype had a shorter lung metastasis–free survival than those with tumors of the other subtypes in all three breast tumor series (P = 0.005, P = 0.06, and P < 10−3 for the NKI, EMC, and MSK series, respectively; Supplementary Fig. S1). However, in both the NKI and EMC series, the identification of patients with a higher risk of developing lung metastases is increased when evaluated on the basis of the 6GS expression (P = 0.001 and P = 0.002, respectively), whereas there was no difference for the MSK cohort, which was composed almost exclusively of lung-relapsing tumors of the basal-like phenotype (P < 10−3). The performance of the 6GS with regard to lung metastasis prediction is significantly higher than that of the basal subtype, suggesting that the 6GS contains additional information beyond the molecular subtype.

However, these analyses were done on heterogeneous series. A better analysis would involve evaluating the performance of the 6GS independently in the subgroups of basal-like and luminal B tumors, even though subdividing the samples might decrease the statistical significance of the analyses.

The 6GS still significantly discriminates the basal subgroup of tumors with respect to lung metastasis, even though its performance is relatively weak probably due to the overlapping of the 6GS with the basal type. In this subgroup, the breast cancer patients tend to have a shorter lung metastasis–free survival when expressing high levels of the 6GS (in the NKI and EMC cohorts; Fig. 3B). This tendency is strengthened when combining the three series (P = 0.046; Fig. 3B). When adjusted to other clinical parameters, the 6GS still tends to predict lung metastasis for basal tumors (P = 0.05; Table 1A).

Figure 3.

Performance of the 6GS in basal-like tumors. A, hierarchical clustering of the subgroup of basal-like tumors. B, Kaplan-Meier analysis in the combined cohort (n = 158). The 6GS better distinguished patients with a high risk of lung metastasis. Patients expressing higher levels of the 6GS had shorter lung metastasis–free survival (P = 0.046).

Figure 3.

Performance of the 6GS in basal-like tumors. A, hierarchical clustering of the subgroup of basal-like tumors. B, Kaplan-Meier analysis in the combined cohort (n = 158). The 6GS better distinguished patients with a high risk of lung metastasis. Patients expressing higher levels of the 6GS had shorter lung metastasis–free survival (P = 0.046).

Close modal
Table 1.

Multivariate analysis of the 6GS performance in basal-like and luminal B tumors

VariableHR (95% CI)P
(A) Basal-like tumors 
    6GS (pos. vs neg.) 2.21 (0.99–4.96) 0.05 
    Lymph node positive (yes vs no) 1.27 (0.57–2.85) 0.56 
    ER negative (yes vs no) 1.21 (0.42–3.46) 0.72 
 
(B) Luminal B tumors 
    6GS (pos. vs neg.) 3.34 (1.34–8.36) 0.01 
    Lymph node positive (yes vs no) 0.74 (0.27–2.08) 0.57 
    ER negative (yes vs no) 0.71 (0.09–5.39) 0.74 
VariableHR (95% CI)P
(A) Basal-like tumors 
    6GS (pos. vs neg.) 2.21 (0.99–4.96) 0.05 
    Lymph node positive (yes vs no) 1.27 (0.57–2.85) 0.56 
    ER negative (yes vs no) 1.21 (0.42–3.46) 0.72 
 
(B) Luminal B tumors 
    6GS (pos. vs neg.) 3.34 (1.34–8.36) 0.01 
    Lymph node positive (yes vs no) 0.74 (0.27–2.08) 0.57 
    ER negative (yes vs no) 0.71 (0.09–5.39) 0.74 

Abbreviations: HR, hazard ratio; 95% CI, 95% confidence interval; ER, estrogen receptor.

The performance of the 6GS is highly significant within the luminal B subgroup. In the NKI series, the 6GS effectively clusters the tumors that metastasize to the lung (Fig. 4A). More importantly, in both the NKI and EMC series, the 6GS is associated with the propensity of luminal B tumors to metastasize to the lung (Fig. 4B). This correlation is strengthened in the combined series of luminal B tumors. The 6GS shows a significant correlation with lung relapse (P = 0.005; Fig. 4B). The specificity and sensitivity of the 6GS to predict lung metastasis in the subgroup of luminal B tumors were estimated at 74% and 58%, respectively. To ensure that the 6GS improved risk stratification independently of the standard clinical variables, we performed a multivariate analysis on the combined cohort (Table 1B). The Cox model showed that the 6GS is an independent predictor of lung metastasis (P = 0.01).

Figure 4.

Performance of the 6GS in luminal B tumors. A, hierarchical clustering of the subgroup of luminal B tumors. In the NKI series, the 6GS clustered together the majority of primary breast tumors that metastasized to the lung. B, Kaplan-Meier analysis in the combined cohort (n = 165). The 6GS better distinguished patients with a high risk of lung metastasis. Patients expressing higher levels of the 6GS had shorter lung metastasis–free survival (P = 0.005).

Figure 4.

Performance of the 6GS in luminal B tumors. A, hierarchical clustering of the subgroup of luminal B tumors. In the NKI series, the 6GS clustered together the majority of primary breast tumors that metastasized to the lung. B, Kaplan-Meier analysis in the combined cohort (n = 165). The 6GS better distinguished patients with a high risk of lung metastasis. Patients expressing higher levels of the 6GS had shorter lung metastasis–free survival (P = 0.005).

Close modal

We have shown that the basal-like signature and the lung metastasis signature exhibit both overlapping and distinct clinical and molecular features.

These results are in agreement with those reported by Smid and colleagues showing that the site of distant relapse was not randomly distributed across subtypes (8). Lung metastases were more frequently observed among the basal-like and the luminal B tumors. Importantly, they verified that the site of relapse was not associated with prognosis, suggesting that pairing of a site of relapse to a subtype is not based on common prognostic outcome (8). Our results showed that the 6GS is a good predictor of lung metastasis but is not an indicator of overall prognosis (4). In contrast, the basal subtype signature is associated with a poor prognosis, whereas its performance in predicting lung metastasis is relatively weak.

In addition, Smid and colleagues described molecular similarities between the site of relapse and the subtype. In particular, they showed that 88% of the differentially expressed genes in patients with lung relapses were also found in the basal subtype. In our study, two genes in the 6GS were also included in the intrinsic signature; DSC2 is a marker of the basal-like subtype, and ANP32E is a luminal B–associated gene.

Finally, Smid and colleagues identified a significant correlation between “focal adhesion” pathways and tumors from patients with lung metastases. These results also accord with our data; the 6GS consists of an integrin (ITGB8), a desmosomal protein (DSC2), a focal adhesion molecule (FERMT1), and a membrane-bound glycoprotein (UGT8).

It is likely that the basal-like and the lung metastatic phenotypes share common biological pathways. Our results provide evidence that the 6GS might reflect a unifying signaling pathway that mediates both phenotypes.

Molecular signatures have been intensely studied in efforts to explain the phenotypic diversity of breast tumors. The large number of intrinsic genes that discriminate the five molecular subtypes suggests that the underlying biology is indeed complex and very different among breast tumors.

Recent studies have shown that the metastatic proclivity might be attributable, in part, to lineage-specific factors. In two different models of carcinogenesis (adenocarcinoma of the breast and melanoma), the differentiation program of the normal cell of origin exerts a strong influence on eventual metastasis (12, 13).

The timing of acquisition of the metastatic ability and the organ specificity remain unclear. A key to understanding these critical issues requires refining the molecular portraits of breast tumors and their metastases so that they reflect all biological and clinical characteristics.

No potential conflicts of interest were disclosed.

Grant support: The Breast Cancer Research Foundation.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.

1
Kang
Y
,
Siegel
PM
,
Shu
W
, et al
. 
A multigenic program mediating breast cancer metastasis to bone
.
Cancer Cell
2003
;
3
:
537
49
.
2
Minn
AJ
,
Gupta
GP
,
Siegel
PM
, et al
. 
Genes that mediate breast cancer metastasis to lung
.
Nature
2005
;
436
:
518
24
.
3
Minn
AJ
,
Gupta
GP
,
Padua
D
, et al
. 
Lung metastasis genes couple breast tumor size and metastatic spread
.
Proc Natl Acad Sci U S A
2007
;
104
:
6740
5
.
4
Landemaine
T
,
Jackson
A
,
Bellahcene
A
, et al
. 
A six-gene signature predicting breast cancer lung metastasis
.
Cancer Res
2008
;
68
:
6092
9
.
5
Perou
CM
,
Sorlie
T
,
Eisen
MB
, et al
. 
Molecular portraits of human breast tumors
.
Nature
2000
;
406
:
747
52
.
6
Sorlie
T
,
Perou
CM
,
Tibshirani
R
, et al
. 
Gene expression patterns of breast carcinomas distinguish tumor subclasses with clinical implications
.
Proc Natl Acad Sci U S A
2001
;
98
:
10869
74
.
7
Rouzier
R
,
Perou
CM
,
Symmans
WF
, et al
. 
Breast cancer molecular subtypes respond differently to preoperative chemotherapy
.
Clin Cancer Res
2005
;
11
:
5678
85
.
8
Smid
M
,
Wang
Y
,
Zhang
Y
, et al
. 
Subtypes of breast cancer show preferential site of relapse
.
Cancer Res
2008
;
68
:
3108
14
.
9
Culhane
AC
,
Quackenbush
J
. 
Confounding effects in “a six-gene signature predicting breast cancer lung metastasis”
.
Cancer Res
2009
;
69
:
7480
5
.
10
van't Veer
LJ
,
Dai
H
,
van de Vijver
MJ
, et al
. 
Gene expression profiling predicts clinical outcome of breast cancer
.
Nature
2002
;
415
:
530
6
.
11
van de Vijver
MJ
,
He
YD
,
van't Veer
LJ
, et al
. 
A gene expression signature as a predictor of survival in breast cancer
.
N Engl J Med
2002
;
347
:
1999
2009
.
12
Ince
TA
,
Richardson
AL
,
Bell
GW
, et al
. 
Transformation of different human breast epithelial cell types leads to distinct tumor phenotypes
.
Cancer Cell
2007
;
12
:
160
70
.
13
Gupta
PB
,
Kuperwasser
C
,
Brunet
JP
, et al
. 
The melanocyte differentiation program predisposes to metastasis after neoplastic transformation
.
Nat Genet
2005
;
37
:
1047
54
.

Supplementary data