HER2 Support Group Forums - View Single Post

Lani · 10-18-2006, 10:22 AM

16q stands for part of chromosome 16 which contains genes important in survival of breast cancer

Her2ER- tumors are usually in the ERBB2 subtype
her2 ER+ tumors are usually in the luminal B subtype

Figure 5. Expression in 16q across the Sørlie et al., tumor subtypes indicates that the subtypes should be grouped into a high and a low expression group. The difference in average median-normalized expression in 16q was tested between all possible combinations (n = 10) of subtypes. To the left, thin lines indicate a significant difference in means between subtypes, thick lines indicate that there is no significant difference. The corresponding P values are listed to the right.

Table 1. Statistics by Tumor Subtype for 359 Classified Tumors
Basal ERBB2 Luminal A Luminal B Normal-like
N 57 40 118 53 91
ER-positive (%) 2 44 70 97 87 92
PgR-positive (%) 2 53 70 93 77 92
Grade 1 (%) 2 P < 0.001a 4 5 34 6 43
Grade 2 (%)a 22 38 55 38 52
Grade 3 (%)a 75 56 11 56 5
5 years RFS (%) 2 P < 0.001b 68 48 84 57 80
Mean expression in 16q (95% CI)c 0.207 0.209 0.024 0.199 0.023
ANOVAP < 0.001 (0.158-0.257) (0.157-0.261) (-0.011-0.059) (0.144-0.254) (-0.001-0.047)
Statistical tests and associated P-values are listed below the row names, in the case of Elston grade the Pearson 2 test (under Grade 1) refers to a test of all grades versus subtypes.
a Elston-Ellis grade.
b 5 years or more of recurrence-free survival
c For each patient, the average median-normalized expression (log2) for all genes in 16q was calculated. Mean with 95% confidence interval for subtypes, as indicated by column headers.
DISCUSSION

We have investigated the relationship between chromosomal positions and gene expression in breast cancer patients, specifically in relation to survival and the expression-based tumor classification described by Sørlie et al. ([2001], [2003]). Our major finding is a remarkable negative association between gene expression in 16q and survival. The Bonferroni corrected P values for overrepresentation of 16q in the 200 and 500 gene lists were both <2.2 × 10-16. This striking feature of the gene expression in 16q seems local in the sense that a majority of genes in 16q have a tendency for inverse association with survival, as is obvious when studying Figure 2, where the distribution of t-scores for 16q seems shifted to the left. A wider and possibly bimodal distribution for 16q suggests that there is a sub-distribution of genes (to the right) for which the negative association with survival is less pronounced. This sub-distribution, however, seems shifted to the left also; the rightmost part of the curve for 16q indicates lower density at given t-scores than what is indicated for average chromosomal arms. This was also found to be true when the height of the 16q curve was increased to compensate for the lower height of the right sub-distribution (not shown). The division of genes into groups corresponding to chromosomal arms is, at least to some extent, artificial; chromosomal aberrations cannot be expected to honor these boundaries. The overrepresentation approach could however be applied to the same gene lists to test for disproportionate contributions from chromosomal arms and bands, and check for consistency. Interestingly, for the negative association between 16q and survival, the result was quite consistent; 16q22, 16q24, 16q12, and 16q23-24 were deemed significant in the 200 and 500 gene lists. Only one chromosomal band, not located on 16q (20q11) had a significant P value for negative association with survival. There was less consistency in the gene lists positively associated with survival, e.g., the chromosomal bands 8p22-p21 were significantly overrepresented in the 200 and 500 gene lists, but the chromosomal arm 8p was not. A possible explanation for this could be that 16q loss often extends to the entire chromosomal arm, whereas 8p is often subjected to loss and gain/amplification with a breakpoint region in the center of the arm.

To investigate the relationship between gene expression in 16q and other tumor characteristics, we have used an expression estimate for the entire arm. It is reasonable to ask whether this estimate can be informative; are genes on this arm sufficiently co-expressed to make this summary valid? A heat map of the genes on this arm does give that impression (Supplementary Fig. 1; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). The tendency for genes to co-vary in the chromosomal arm is indicated by the SD of the estimate across patients. The SDs for all chromosomal arms (as a function of size) are plotted in Supplementary Figure 2 (Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). To assess the probability of the SD given a null-hypothesis of no tendency for coexpression, genes were randomly and repeatedly assigned to chromosomal arms, and SDs were calculated. Supplementary Table 4 (Supplementary material for this article can be found at http://www.interscience.wiley.com.la...5-2257/suppmat) shows the deviation from these null-distributions for each chromosomal arm. From this analysis we can draw two important conclusions: (i) it seems that for all chromosomal arms with >50 probed genes, a per patient summary of gene expression is in fact relevant since the spread in the estimate across patients is significantly higher than expected. This suggests that the added effects of chromosomal aberrations, other epigenetic regulation, and clustering of genes with similar transcriptional regulation significantly affects a majority of chromosomal arms; (ii) ranking of chromosomal arms according to deviation (z-score) from the corresponding null-distribution achieves an obvious similarity with a ranking based on frequency of chromosomal aberrations. Rennstam et al. ([2003]) report from a recent CGH experiment conducted on a cohort of 305 primary invasive breast cancers that gene copy number aberrations were observed in >90% of tumors, that all chromosomal arms were involved at various frequencies, and that the most common were: +1q (55%), +8q (41%), +16p (40%), +17q (28%), -13q (27%), -16q (22%), +20q (19%), -8p (18%), +11q (16%). The overlap in top ranking arms between Rennstam's et al. report and Supplementary Table 4 is consistent with previous reports that chromosomal aberrations are a major reason for the apparent genomic clustering of similarly expressed genes in breast and several other cancers (Phillips et al., [2005]; Fujii et al., [2002]; Aggarwal et al., [2005]; Reyal et al., [2005]). It is therefore reasonable to consider known chromosomal aberrations in 16q. Loss of heterozygosity in this arm is one of the most frequent events in breast cancer. E-cadherin is likely to be the tumor suppressor gene explaining recurrent loss of 16q in lobular breast cancers (Berx et al., [1995]). In search for the target gene in ductal carcinomas (comprising the majority of breast cancers) the smallest region of overlap (SRO) with regards to copy number aberrations has been exhaustively searched for. Several SROs have been suggested, but no clear consensus has been reached (Rakha et al., [2006]). Several putative genes have been screened and thus far none has fulfilled the criteria for target genes. Although the culprit gene seems elusive, 16q has been implicated in relation to important tumor characteristics. Loss of genomic material in 16q has been proposed an early event in a low grade-good prognosis pathway of breast cancer progression. In a CGH experiment, where invasive ductal Grade I and III breast carcinomas were contrasted, Roylance et al. ([1999]) found that 65% of Grade I tumors had lost the long arm of chromosome 16 compared with only 16% of Grade III tumors. This pattern of loss led the investigators to conclude that the majority of Grade I tumors do not progress to Grade III tumors, since it would be necessary for the Grade I tumors to regain lost material in 16q in order for this to happen. That at least two different genetic pathways for tumor progression exist for invasive ductal carcinomas (comprising the majority of breast carcinomas), and that loss of 16q is a key cytogenetic event, is further supported by Buerger et al., ([2001]). In Table 1, the same relationship between grade and 16q expression can be observed: tumors with high expression in 16q are more frequently high grade tumors and the inverse is true for tumors with low expression in 16q. Stratification of the 16q expression estimate across patient cohorts and histological grade demonstrated that 16q expression is generally higher in the Uppsala patient cohort, despite a smaller percentage of Grade 3 tumors (Supplementary Table 3; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). This systematic difference is unlikely to be biological in origin; when the cohorts are considered independently the same relationship between grade and 16q expression is obvious. The Grade 3 tumors have the highest and most distinguished 16q expression in both cohorts. As the mean expression in 16q has overlapping confidence intervals between Grade 1 and 2 tumors, it is reasonable to test the ability for 16q gene expression to discriminate between Grade 3 tumors on one hand, and Grade 1 and 2 tumors on the other (Supplementary Fig. 3; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). The ROC of this analysis is 0.72, which is slightly worse than the ability to discriminate between Sørlie subtypes. This is interesting when considering the studies of Roylance et al., ([1999]) and Buerger et al., ([2001]), where grade seems to be the most important feature with regards to the tumor profile of chromosomal gains and losses. Our analysis suggests that the expression-based classification of Sørlie et al. ([2001], [2003]) agrees better with 16q expression than stratification according to grade does. Obviously, it is important to keep in mind that we have used different methods; Roylance et al. ([1999]) and Buerger et al. ([2001]) have assessed DNA in relation to chromosomal position, we have assessed RNA.

The idea that the unique position held by 16q in relation to survival is a reflection of recurrent loss of 16q in a favorable prognosis pathway of tumor progression is intriguing, when expression in 16q across the Sørlie et al. breast cancer subtypes is considered. Expression in 16q across the Sørlie et al. tumor subtypes, as reproduced in the current data set, indicate that they should be grouped into a high and a low expression group. The high expression group consists of basal, ERBB2, and luminal B subtypes, the low expression group of luminal A and normal-like subtypes. It is interesting that separating the patients in two groups, based on expression in 16q, produces a better agreement with the Sørlie et al. breast cancer classification than with the division of patients into poor and favorable prognosis categories. A possible explanation for this would be that the 16q and Sørlie et al. classifications are more direct descriptions of the biology of the tumor. Survival has a more complex relationship to tumor biology, and is affected by potentially tumor independent factors, such as treatment. Given the proposed model of tumor progression, expression in 16q across the Sørlie et al. subtypes would suggest that the ERBB2, basal, and luminal B tumors progress along the high grade-poor prognosis path, while luminal A and normal-like tumors progress along the low grade-good prognosis path. The identification of a possibly pivotal gene expression difference in 16q between luminal type A and luminal type B tumors is unexpected. The luminal type B designation does, however, seem to identify a group of tumors with poor prognosis and, in contrast to what is expected for the poor prognosis pathway, a quite high frequency of ER receptor expression. Recently, Roylance et al. ([2006]) have published a paper, describing results of an array-CGH experiment focusing on 16q. They now interpret their data as being more consistent with a significant number of ductal carcinomas progressing through grades, with subsequent accumulation of segmental gains on 16q in the higher grade lesions. Our findings do not contradict this, but a no less interesting questions is raised: have ERBB2, basal, and luminal B subtypes often progressed further along this proposed common path, explaining why they frequently have high expression in 16q, or is increasing 16q expression in fact accompanied by a phenotype shift from luminal A and normal-like subtypes to the others? Additional experiments are of course necessary to elucidate this.

The fundamental aim of this study was to test if specific chromosomal positions - implicated by a possible influence on gene expression - could be connected to useful clinical endpoints, and to investigate if the result would reflect previous findings regarding chromosomal aberrations in breast cancer. Regarding 16q, our data on gene expression are consistent with previous findings on copy number aberrations in this region, as discussed. Corrected P values were significant for a few other chromosomal arms also: 20q, 1p, 13q, and 9p. In 20q multiple regions - among them 20q11 - have been found to be recurrently amplified in breast cancer (Kallioniemi et al., [1994]; Hodgson et al., [2003]), and amplification in at least one region has been associated with a significantly shorter disease-free survival (Tanner et al., [1995]). Loss of heterozygosity in 1p has been reported in a large number of human cancers, including breast cancer, where association with reduced patient survival has been demonstrated (Ragnarsson et al., [1999]). The long arm of chromosome 13 harbors two well-known tumor suppressor genes, RB1 (13q14) and BRCA2 (13q12-13), the latter involved in hereditary breast cancer. In sporadic breast cancer LOH in several 13q regions - 13q12-13, 13q14, 13q21-22, and 13q31-q34 - is recurrent (Eiriksdottir et al., [1998]; Tong et al., [2004]), and for 13q12-13 it has been associated with a 3- to 4-fold increased risk of relapse and death (Eiriksdottiret al., [1998]). In 9p, loss of heterozygosity is frequent and has been associated with more rapid cell division and aneuploidy, although no significant association with survival was apparent (Eiriksdottir et al., [1995]). The expected effects on gene expression of these described gene dosage alterations are all consistent with our findings.

So, we may conclude that the chromosomal positions overrepresented in relation to survival largely reflect published data on recurrent chromosomal aberrations. For researchers dealing with expression data in cancer this is an important issue since physical genomic clustering of genes with similar transcription is not confined to tumor tissues; it has previously been described in yeast and normal tissues of mice and human (Cohen et al., [2000]; Hughes et al., [2000]; Su et al., [2004]). In cells or tissues, where euploidy can be expected, epigenetic mechanisms of gene regulation and clustering of functionally related genes are believed to be the cause. Although direct examinations of the effect of gene copy number aberrations on gene expression has been described as substantial (Hyman et al., [2002]; Pollack et al., [2002]; Linn et al., [2003]; Wolf et al., [2004]; Aggarwal et al., [2005]; Bea et al., [2005]; Grade et al., [2006]), the inverse is not necessarily true; clustering of functionally related genes or epigenetic mechanisms of gene regulation could be the major cause for disproportionate contributions from certain genomic regions to differential gene expression in cancer also.

In conclusion, our results indicate that important biological information can be extracted from gene expression data in breast cancer by studying non-random connections between chromosomal positions and gene expression. We feel that investigation of a possible influence of specific chromosomal positions on gene expression in tumors should be attempted more often. When Sørlie and coworkers first classified breast tumors based on pervasive gene expression differences (Perou et al., [2000]), a group of breast cancers characterized by few chromosomal aberrations but recurrent loss of 16q had been described a decade earlier (Dutrillaux et al., [1990]). Our data strongly indicate that this described group of tumors significantly overlaps with the luminal A and normal-like subtypes, illustrating how novel gene expression-based descriptions of tumor biology can find additional support by using the extensive amount of published data regarding chromosomal aberrations available in cancer. It is likely that additional chromosomal positions can be implicated in relation to other clinical variables, or in relation to results from any of the many bioinformatics tools available for analysis of gene expression data. This holds promise of pinpointing chromosomal regions - or even individual genes - pivotal in activating or suppressing pathways, transcriptional networks and other gene expression themes important for various aspects of tumor biology.

10-18-2006, 10:22 AM	#32
Lani Senior Member Join Date: Mar 2006 Posts: 4,782	Posting this hot-off-the press important data 2 avoid its transfer2 "articles" 16q stands for part of chromosome 16 which contains genes important in survival of breast cancer Her2ER- tumors are usually in the ERBB2 subtype her2 ER+ tumors are usually in the luminal B subtype Figure 5. Expression in 16q across the Sørlie et al., tumor subtypes indicates that the subtypes should be grouped into a high and a low expression group. The difference in average median-normalized expression in 16q was tested between all possible combinations (n = 10) of subtypes. To the left, thin lines indicate a significant difference in means between subtypes, thick lines indicate that there is no significant difference. The corresponding P values are listed to the right. Table 1. Statistics by Tumor Subtype for 359 Classified Tumors Basal ERBB2 Luminal A Luminal B Normal-like N 57 40 118 53 91 ER-positive (%) 2 44 70 97 87 92 PgR-positive (%) 2 53 70 93 77 92 Grade 1 (%) 2 P < 0.001a 4 5 34 6 43 Grade 2 (%)a 22 38 55 38 52 Grade 3 (%)a 75 56 11 56 5 5 years RFS (%) 2 P < 0.001b 68 48 84 57 80 Mean expression in 16q (95% CI)c 0.207 0.209 0.024 0.199 0.023 ANOVAP < 0.001 (0.158-0.257) (0.157-0.261) (-0.011-0.059) (0.144-0.254) (-0.001-0.047) Statistical tests and associated P-values are listed below the row names, in the case of Elston grade the Pearson 2 test (under Grade 1) refers to a test of all grades versus subtypes. a Elston-Ellis grade. b 5 years or more of recurrence-free survival c For each patient, the average median-normalized expression (log2) for all genes in 16q was calculated. Mean with 95% confidence interval for subtypes, as indicated by column headers. DISCUSSION We have investigated the relationship between chromosomal positions and gene expression in breast cancer patients, specifically in relation to survival and the expression-based tumor classification described by Sørlie et al. ([2001], [2003]). Our major finding is a remarkable negative association between gene expression in 16q and survival. The Bonferroni corrected P values for overrepresentation of 16q in the 200 and 500 gene lists were both <2.2 × 10-16. This striking feature of the gene expression in 16q seems local in the sense that a majority of genes in 16q have a tendency for inverse association with survival, as is obvious when studying Figure 2, where the distribution of t-scores for 16q seems shifted to the left. A wider and possibly bimodal distribution for 16q suggests that there is a sub-distribution of genes (to the right) for which the negative association with survival is less pronounced. This sub-distribution, however, seems shifted to the left also; the rightmost part of the curve for 16q indicates lower density at given t-scores than what is indicated for average chromosomal arms. This was also found to be true when the height of the 16q curve was increased to compensate for the lower height of the right sub-distribution (not shown). The division of genes into groups corresponding to chromosomal arms is, at least to some extent, artificial; chromosomal aberrations cannot be expected to honor these boundaries. The overrepresentation approach could however be applied to the same gene lists to test for disproportionate contributions from chromosomal arms and bands, and check for consistency. Interestingly, for the negative association between 16q and survival, the result was quite consistent; 16q22, 16q24, 16q12, and 16q23-24 were deemed significant in the 200 and 500 gene lists. Only one chromosomal band, not located on 16q (20q11) had a significant P value for negative association with survival. There was less consistency in the gene lists positively associated with survival, e.g., the chromosomal bands 8p22-p21 were significantly overrepresented in the 200 and 500 gene lists, but the chromosomal arm 8p was not. A possible explanation for this could be that 16q loss often extends to the entire chromosomal arm, whereas 8p is often subjected to loss and gain/amplification with a breakpoint region in the center of the arm. To investigate the relationship between gene expression in 16q and other tumor characteristics, we have used an expression estimate for the entire arm. It is reasonable to ask whether this estimate can be informative; are genes on this arm sufficiently co-expressed to make this summary valid? A heat map of the genes on this arm does give that impression (Supplementary Fig. 1; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). The tendency for genes to co-vary in the chromosomal arm is indicated by the SD of the estimate across patients. The SDs for all chromosomal arms (as a function of size) are plotted in Supplementary Figure 2 (Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). To assess the probability of the SD given a null-hypothesis of no tendency for coexpression, genes were randomly and repeatedly assigned to chromosomal arms, and SDs were calculated. Supplementary Table 4 (Supplementary material for this article can be found at http://www.interscience.wiley.com.la...5-2257/suppmat) shows the deviation from these null-distributions for each chromosomal arm. From this analysis we can draw two important conclusions: (i) it seems that for all chromosomal arms with >50 probed genes, a per patient summary of gene expression is in fact relevant since the spread in the estimate across patients is significantly higher than expected. This suggests that the added effects of chromosomal aberrations, other epigenetic regulation, and clustering of genes with similar transcriptional regulation significantly affects a majority of chromosomal arms; (ii) ranking of chromosomal arms according to deviation (z-score) from the corresponding null-distribution achieves an obvious similarity with a ranking based on frequency of chromosomal aberrations. Rennstam et al. ([2003]) report from a recent CGH experiment conducted on a cohort of 305 primary invasive breast cancers that gene copy number aberrations were observed in >90% of tumors, that all chromosomal arms were involved at various frequencies, and that the most common were: +1q (55%), +8q (41%), +16p (40%), +17q (28%), -13q (27%), -16q (22%), +20q (19%), -8p (18%), +11q (16%). The overlap in top ranking arms between Rennstam's et al. report and Supplementary Table 4 is consistent with previous reports that chromosomal aberrations are a major reason for the apparent genomic clustering of similarly expressed genes in breast and several other cancers (Phillips et al., [2005]; Fujii et al., [2002]; Aggarwal et al., [2005]; Reyal et al., [2005]). It is therefore reasonable to consider known chromosomal aberrations in 16q. Loss of heterozygosity in this arm is one of the most frequent events in breast cancer. E-cadherin is likely to be the tumor suppressor gene explaining recurrent loss of 16q in lobular breast cancers (Berx et al., [1995]). In search for the target gene in ductal carcinomas (comprising the majority of breast cancers) the smallest region of overlap (SRO) with regards to copy number aberrations has been exhaustively searched for. Several SROs have been suggested, but no clear consensus has been reached (Rakha et al., [2006]). Several putative genes have been screened and thus far none has fulfilled the criteria for target genes. Although the culprit gene seems elusive, 16q has been implicated in relation to important tumor characteristics. Loss of genomic material in 16q has been proposed an early event in a low grade-good prognosis pathway of breast cancer progression. In a CGH experiment, where invasive ductal Grade I and III breast carcinomas were contrasted, Roylance et al. ([1999]) found that 65% of Grade I tumors had lost the long arm of chromosome 16 compared with only 16% of Grade III tumors. This pattern of loss led the investigators to conclude that the majority of Grade I tumors do not progress to Grade III tumors, since it would be necessary for the Grade I tumors to regain lost material in 16q in order for this to happen. That at least two different genetic pathways for tumor progression exist for invasive ductal carcinomas (comprising the majority of breast carcinomas), and that loss of 16q is a key cytogenetic event, is further supported by Buerger et al., ([2001]). In Table 1, the same relationship between grade and 16q expression can be observed: tumors with high expression in 16q are more frequently high grade tumors and the inverse is true for tumors with low expression in 16q. Stratification of the 16q expression estimate across patient cohorts and histological grade demonstrated that 16q expression is generally higher in the Uppsala patient cohort, despite a smaller percentage of Grade 3 tumors (Supplementary Table 3; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). This systematic difference is unlikely to be biological in origin; when the cohorts are considered independently the same relationship between grade and 16q expression is obvious. The Grade 3 tumors have the highest and most distinguished 16q expression in both cohorts. As the mean expression in 16q has overlapping confidence intervals between Grade 1 and 2 tumors, it is reasonable to test the ability for 16q gene expression to discriminate between Grade 3 tumors on one hand, and Grade 1 and 2 tumors on the other (Supplementary Fig. 3; Supplementary material for this article can be found at http://www.interscience.wiley.com.la...-2257/suppmat). The ROC of this analysis is 0.72, which is slightly worse than the ability to discriminate between Sørlie subtypes. This is interesting when considering the studies of Roylance et al., ([1999]) and Buerger et al., ([2001]), where grade seems to be the most important feature with regards to the tumor profile of chromosomal gains and losses. Our analysis suggests that the expression-based classification of Sørlie et al. ([2001], [2003]) agrees better with 16q expression than stratification according to grade does. Obviously, it is important to keep in mind that we have used different methods; Roylance et al. ([1999]) and Buerger et al. ([2001]) have assessed DNA in relation to chromosomal position, we have assessed RNA. The idea that the unique position held by 16q in relation to survival is a reflection of recurrent loss of 16q in a favorable prognosis pathway of tumor progression is intriguing, when expression in 16q across the Sørlie et al. breast cancer subtypes is considered. Expression in 16q across the Sørlie et al. tumor subtypes, as reproduced in the current data set, indicate that they should be grouped into a high and a low expression group. The high expression group consists of basal, ERBB2, and luminal B subtypes, the low expression group of luminal A and normal-like subtypes. It is interesting that separating the patients in two groups, based on expression in 16q, produces a better agreement with the Sørlie et al. breast cancer classification than with the division of patients into poor and favorable prognosis categories. A possible explanation for this would be that the 16q and Sørlie et al. classifications are more direct descriptions of the biology of the tumor. Survival has a more complex relationship to tumor biology, and is affected by potentially tumor independent factors, such as treatment. Given the proposed model of tumor progression, expression in 16q across the Sørlie et al. subtypes would suggest that the ERBB2, basal, and luminal B tumors progress along the high grade-poor prognosis path, while luminal A and normal-like tumors progress along the low grade-good prognosis path. The identification of a possibly pivotal gene expression difference in 16q between luminal type A and luminal type B tumors is unexpected. The luminal type B designation does, however, seem to identify a group of tumors with poor prognosis and, in contrast to what is expected for the poor prognosis pathway, a quite high frequency of ER receptor expression. Recently, Roylance et al. ([2006]) have published a paper, describing results of an array-CGH experiment focusing on 16q. They now interpret their data as being more consistent with a significant number of ductal carcinomas progressing through grades, with subsequent accumulation of segmental gains on 16q in the higher grade lesions. Our findings do not contradict this, but a no less interesting questions is raised: have ERBB2, basal, and luminal B subtypes often progressed further along this proposed common path, explaining why they frequently have high expression in 16q, or is increasing 16q expression in fact accompanied by a phenotype shift from luminal A and normal-like subtypes to the others? Additional experiments are of course necessary to elucidate this. The fundamental aim of this study was to test if specific chromosomal positions - implicated by a possible influence on gene expression - could be connected to useful clinical endpoints, and to investigate if the result would reflect previous findings regarding chromosomal aberrations in breast cancer. Regarding 16q, our data on gene expression are consistent with previous findings on copy number aberrations in this region, as discussed. Corrected P values were significant for a few other chromosomal arms also: 20q, 1p, 13q, and 9p. In 20q multiple regions - among them 20q11 - have been found to be recurrently amplified in breast cancer (Kallioniemi et al., [1994]; Hodgson et al., [2003]), and amplification in at least one region has been associated with a significantly shorter disease-free survival (Tanner et al., [1995]). Loss of heterozygosity in 1p has been reported in a large number of human cancers, including breast cancer, where association with reduced patient survival has been demonstrated (Ragnarsson et al., [1999]). The long arm of chromosome 13 harbors two well-known tumor suppressor genes, RB1 (13q14) and BRCA2 (13q12-13), the latter involved in hereditary breast cancer. In sporadic breast cancer LOH in several 13q regions - 13q12-13, 13q14, 13q21-22, and 13q31-q34 - is recurrent (Eiriksdottir et al., [1998]; Tong et al., [2004]), and for 13q12-13 it has been associated with a 3- to 4-fold increased risk of relapse and death (Eiriksdottiret al., [1998]). In 9p, loss of heterozygosity is frequent and has been associated with more rapid cell division and aneuploidy, although no significant association with survival was apparent (Eiriksdottir et al., [1995]). The expected effects on gene expression of these described gene dosage alterations are all consistent with our findings. So, we may conclude that the chromosomal positions overrepresented in relation to survival largely reflect published data on recurrent chromosomal aberrations. For researchers dealing with expression data in cancer this is an important issue since physical genomic clustering of genes with similar transcription is not confined to tumor tissues; it has previously been described in yeast and normal tissues of mice and human (Cohen et al., [2000]; Hughes et al., [2000]; Su et al., [2004]). In cells or tissues, where euploidy can be expected, epigenetic mechanisms of gene regulation and clustering of functionally related genes are believed to be the cause. Although direct examinations of the effect of gene copy number aberrations on gene expression has been described as substantial (Hyman et al., [2002]; Pollack et al., [2002]; Linn et al., [2003]; Wolf et al., [2004]; Aggarwal et al., [2005]; Bea et al., [2005]; Grade et al., [2006]), the inverse is not necessarily true; clustering of functionally related genes or epigenetic mechanisms of gene regulation could be the major cause for disproportionate contributions from certain genomic regions to differential gene expression in cancer also. In conclusion, our results indicate that important biological information can be extracted from gene expression data in breast cancer by studying non-random connections between chromosomal positions and gene expression. We feel that investigation of a possible influence of specific chromosomal positions on gene expression in tumors should be attempted more often. When Sørlie and coworkers first classified breast tumors based on pervasive gene expression differences (Perou et al., [2000]), a group of breast cancers characterized by few chromosomal aberrations but recurrent loss of 16q had been described a decade earlier (Dutrillaux et al., [1990]). Our data strongly indicate that this described group of tumors significantly overlaps with the luminal A and normal-like subtypes, illustrating how novel gene expression-based descriptions of tumor biology can find additional support by using the extensive amount of published data regarding chromosomal aberrations available in cancer. It is likely that additional chromosomal positions can be implicated in relation to other clinical variables, or in relation to results from any of the many bioinformatics tools available for analysis of gene expression data. This holds promise of pinpointing chromosomal regions - or even individual genes - pivotal in activating or suppressing pathways, transcriptional networks and other gene expression themes important for various aspects of tumor biology.