Limits...
Identification of a novel luminal molecular subtype of breast cancer.

Dvorkin-Gheva A, Hassell JA - PLoS ONE (2014)

Bottom Line: All of the unclassifiable samples could be grouped into a new molecular subtype, which we termed "luminal-like".We found that patients harboring tumors of the luminal-like subtype have a better prognosis than those with basal-like breast cancer, a similar prognosis to those with ERBB2+, luminal B or claudin-low tumors, but a worse prognosis than patients with luminal A or normal-like breast tumors.Our findings suggest the occurrence of another molecular subtype of breast cancer that accounts for the vast majority of previously unclassifiable breast tumors.

View Article: PubMed Central - PubMed

Affiliation: Centre for Functional Genomics, Department of Biochemistry and Biomedical Sciences, McMaster University, Hamilton, Ontario, Canada.

ABSTRACT
The molecular classification of human breast tumors has afforded insights into subtype specific biological processes, patient prognosis and response to therapies. However, using current methods roughly one quarter of breast tumors cannot be classified into one or another molecular subtype. To explore the possibility that the unclassifiable samples might comprise one or more novel subtypes we employed a collection of publically available breast tumor datasets with accompanying clinical information to assemble 1,593 transcript profiles: 25% of these samples could not be assigned to one of the current molecular subtypes of breast cancer. All of the unclassifiable samples could be grouped into a new molecular subtype, which we termed "luminal-like". We also identified the luminal-like subtype in an independent collection of tumor samples (NKI295). We found that patients harboring tumors of the luminal-like subtype have a better prognosis than those with basal-like breast cancer, a similar prognosis to those with ERBB2+, luminal B or claudin-low tumors, but a worse prognosis than patients with luminal A or normal-like breast tumors. Our findings suggest the occurrence of another molecular subtype of breast cancer that accounts for the vast majority of previously unclassifiable breast tumors.

Show MeSH

Related in: MedlinePlus

Identification of the novel cluster.A. Distribution of Spearman Rank Correlation Coefficients. Distribution of correlation coefficients is shown in the form of a histogram. Gaussian distributions fitted by using the EM (Expectation-Maximalization) algorithm are shown in blue, the sum of these distributions is shown in green. Correlation coefficients lower than than 0.3 are marked in cyan; correlation coefficients higher than 0.3 are marked in red. B. Cophenetic coefficient obtained from NMF clustering. Optimal number of clusters established by this method is indicated by black arrow. C. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the subtype they were assigned to. D. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the dataset of their origin.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4116208&req=5

pone-0103514-g001: Identification of the novel cluster.A. Distribution of Spearman Rank Correlation Coefficients. Distribution of correlation coefficients is shown in the form of a histogram. Gaussian distributions fitted by using the EM (Expectation-Maximalization) algorithm are shown in blue, the sum of these distributions is shown in green. Correlation coefficients lower than than 0.3 are marked in cyan; correlation coefficients higher than 0.3 are marked in red. B. Cophenetic coefficient obtained from NMF clustering. Optimal number of clusters established by this method is indicated by black arrow. C. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the subtype they were assigned to. D. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the dataset of their origin.

Mentions: As reported previously, subtype assignment is performed by identifying a centroid representing a subtype to which a sample has the highest correlation [3]. The latter study did not assign subtypes to samples with a maximal correlation coefficient lower than 0.1, assuming that such low coefficients indicate that the samples are not similar enough to any of the examined subtypes and therefore cannot be assigned to any of them. However, because the cutoff was chosen based on a relatively small collection of samples and because the collection of samples we assembled for our study was much larger than that used initially by Perou and colleagues [1], [3], we attempted to re-define the cutoff. To the latter end we re-examined the distribution of correlation coefficients (Figure 1A) by using the EM (Expectation-Maximalization) algorithm [19] and found a statistically significant cutoff (pā€Š=ā€Š0.0018) of 0.3 for these coefficients. Hence the latter cutoff was used for subtype assignment.


Identification of a novel luminal molecular subtype of breast cancer.

Dvorkin-Gheva A, Hassell JA - PLoS ONE (2014)

Identification of the novel cluster.A. Distribution of Spearman Rank Correlation Coefficients. Distribution of correlation coefficients is shown in the form of a histogram. Gaussian distributions fitted by using the EM (Expectation-Maximalization) algorithm are shown in blue, the sum of these distributions is shown in green. Correlation coefficients lower than than 0.3 are marked in cyan; correlation coefficients higher than 0.3 are marked in red. B. Cophenetic coefficient obtained from NMF clustering. Optimal number of clusters established by this method is indicated by black arrow. C. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the subtype they were assigned to. D. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the dataset of their origin.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4116208&req=5

pone-0103514-g001: Identification of the novel cluster.A. Distribution of Spearman Rank Correlation Coefficients. Distribution of correlation coefficients is shown in the form of a histogram. Gaussian distributions fitted by using the EM (Expectation-Maximalization) algorithm are shown in blue, the sum of these distributions is shown in green. Correlation coefficients lower than than 0.3 are marked in cyan; correlation coefficients higher than 0.3 are marked in red. B. Cophenetic coefficient obtained from NMF clustering. Optimal number of clusters established by this method is indicated by black arrow. C. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the subtype they were assigned to. D. Principal Component Analysis performed on 1,593 tumor samples by using 710 genes (see Table S1). Samples are colored by the dataset of their origin.
Mentions: As reported previously, subtype assignment is performed by identifying a centroid representing a subtype to which a sample has the highest correlation [3]. The latter study did not assign subtypes to samples with a maximal correlation coefficient lower than 0.1, assuming that such low coefficients indicate that the samples are not similar enough to any of the examined subtypes and therefore cannot be assigned to any of them. However, because the cutoff was chosen based on a relatively small collection of samples and because the collection of samples we assembled for our study was much larger than that used initially by Perou and colleagues [1], [3], we attempted to re-define the cutoff. To the latter end we re-examined the distribution of correlation coefficients (Figure 1A) by using the EM (Expectation-Maximalization) algorithm [19] and found a statistically significant cutoff (pā€Š=ā€Š0.0018) of 0.3 for these coefficients. Hence the latter cutoff was used for subtype assignment.

Bottom Line: All of the unclassifiable samples could be grouped into a new molecular subtype, which we termed "luminal-like".We found that patients harboring tumors of the luminal-like subtype have a better prognosis than those with basal-like breast cancer, a similar prognosis to those with ERBB2+, luminal B or claudin-low tumors, but a worse prognosis than patients with luminal A or normal-like breast tumors.Our findings suggest the occurrence of another molecular subtype of breast cancer that accounts for the vast majority of previously unclassifiable breast tumors.

View Article: PubMed Central - PubMed

Affiliation: Centre for Functional Genomics, Department of Biochemistry and Biomedical Sciences, McMaster University, Hamilton, Ontario, Canada.

ABSTRACT
The molecular classification of human breast tumors has afforded insights into subtype specific biological processes, patient prognosis and response to therapies. However, using current methods roughly one quarter of breast tumors cannot be classified into one or another molecular subtype. To explore the possibility that the unclassifiable samples might comprise one or more novel subtypes we employed a collection of publically available breast tumor datasets with accompanying clinical information to assemble 1,593 transcript profiles: 25% of these samples could not be assigned to one of the current molecular subtypes of breast cancer. All of the unclassifiable samples could be grouped into a new molecular subtype, which we termed "luminal-like". We also identified the luminal-like subtype in an independent collection of tumor samples (NKI295). We found that patients harboring tumors of the luminal-like subtype have a better prognosis than those with basal-like breast cancer, a similar prognosis to those with ERBB2+, luminal B or claudin-low tumors, but a worse prognosis than patients with luminal A or normal-like breast tumors. Our findings suggest the occurrence of another molecular subtype of breast cancer that accounts for the vast majority of previously unclassifiable breast tumors.

Show MeSH
Related in: MedlinePlus