Phenotype definition is critical in linkage analysis. The extent of MD can be measured by the percent of the breast that is dense or by the total area that is dense. DA is highly correlated with PMD, but has a skewed distribution. Nevertheless, after transformation and adjustment for covariates, repeated reads of the same mammogram showed very high correlations for DA and for PMD. Therefore, we examined the evidence for linkage to both these phenotypes, and found evidence for suggestive linkage at multiple locations. Furthermore, since MD is known to vary through life, genes that appear to influence MD in mid-life could reflect early influences on MD at the time the breast forms in adolescence, or through subsequent changes or rates of change with increasing age, parity and the menopause, or through influences on both formation and age change in MD [30
]. Our study was designed so that there was rarely more than five years difference between the ages at mammogram for sister pairs in a family; hence the study was partially matched within families for the large effect of age on MD. In addition, MD measures were adjusted for age prior to linkage analysis.
Given the number of study participants and available pedigree information, we had estimated that we would have 80% power to obtain a LOD score in the range of 3 with a locus that could explain 25% of the heritability of PMD. Despite the large collection of families and careful consideration of mammographic density parameters, our primary linkage analysis did not yield LOD scores that exceeded desired thresholds set by genome-wide gene-dropping simulations. With different modeling conditions, three loci on chromosomes 7p14.1-p13, 12p11.22-q13.11, and 7q32.3-q34, showed LOD scores that approach or exceed 3. However, given that these results were obtained after multiple analyses, these signals should be considered as 'suggestive'.
With the development of a model for covariate effects that was inspired by Pike and colleagues [26
], evidence for linkage increased on chromosome 7p, with a maximum LOD score of 3.29. This increase associated with a more careful phenotype definition is suggestive that there may be loci influencing aspects of MD or MD changes, but also that the ideal parameterization for MD is still unknown. The 1-LOD drop interval bounded by rs1949880
corresponds to a 7.2 Mb region containing 72 genes, including small clusters of snoRNA and piRNA genes toward its proximal boundary. Of interest at the proximal boundary, are the insulin-like growth factor binding protein genes, IGFBP1
, both of which have been hypothesized to be involved in mammographic density (and in breast cancer and other cancers) and have been considered in previous association studies examining MD phenotypes [31
]. Of these, two studies included more than 1,000 unrelated women [32
] with investigation of the common genetic variation in these genes, but they did not reveal consistent evidence of association with MD phenotypes. Also, a recent meta-analysis of 4,877 women did not identify association with MD in this region [29
]. As linkage analysis may be less sensitive to allele frequency issues, these genes remain as interesting candidates. Another gene of potential interest within this region includes the ras-related gene, v-ral simian leukemia viral oncogene homolog A (RALA
) with its implicated role in signalling and growth.
When we examined families containing younger sisters (two sisters with mammograms under age 50 years), the peak on chromosome 7q became stronger. This peak with a 1-LOD drop interval bounded by rs4728251
, corresponds to 9.6 Mb containing 69 genes. Phenotypes at younger ages are expected to display a stronger genetic component [36
] but there was no evidence that PMD h2
varied by age in our previous twin studies [30
], and thus age should continue to be considered in future studies. A gene of potential interest within this region is a member of the RAS
oncogene family, RAB19
Chromosome 12: On chromosome 12, a maximum LOD score of 3.3 was seen in families where the sisters would be expected to have dissimilar DA, after adjustment for factors affecting sex hormone levels. Such predicted differences could be due to: differences in the numbers of pregnancies; ages at menopause or menarche; weight; or height. Linkage in this context would identify families where phenotypes are more similar than expected, and this could imply a larger potential genetic effect. The 1-LOD interval of our linkage signal, bounded by rs1909160 and rs1978161 corresponds to a large physical distance of 16.7 Mb encompassing the centromere and contains 80 genes. Genes of potential interest within this region include the vitamin D receptor (VDR) and collagen type II (COL2A1) given their association to breast biology.
We did not detect any significant evidence for linkage on chromosome 5 as was reported in a previous study of 89 multi-generation pedigrees [25
], which, like our study, included mostly Caucasian women; however, there may have been differences in the family ancestries or characteristics of the source populations that are not obvious.
That high mammographic density is associated with risk for breast cancer motivated our study, with anticipation that some gene determinants of MD may be candidates for involvement or development of malignancy [29
]. Many groups have pursued genome-wide linkage studies for breast cancer using non-BRCA1/2
high risk families, more recently with families of confined ancestries (see [38
], and within) after family sets of diverse origins did not yield strong linkage signals [39
]. Suggestive evidence for a chromosome 7 locus was identified in one study [39
], however, this region is more proximal occurring at 7q21.11-q21.3, and does not overlap with the suggestive MD locus detected in our younger women set. Large scale association studies have also been carried out, several with discovery or first-stage study sizes that exceed 1,000 cases of breast cancer [41
] leading to loci for consideration, with considerable family risk remaining to be explained [45
]. We note that none of our suggestive linkage peaks observed coincide with the candidate genes and their local SNPs that reached genome-wide significance in these large studies.