The study used a publicly accessible dataset from the NCBI Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nhi.gov/geo/), identified by the GEO accession number GSE49175, the only dataset responding to the specific requirement of available transcriptome data for histologically normal tissue samples with measured mammographic density.

As described in the original article,14 the dataset consisted of 120 snap-frozen samples of normal breast tissue collected at the time of breast surgery from women of ages 20 to 74 years with newly diagnosed in situ or invasive breast carcinoma and associated with a mammographic density measurement of the unaffected breast taken previously. All the participants provided written informed consent under a protocol approved by the U.S. National Cancer Institute and local (Polish) Institutional Review Boards.

The percentage mammographic density was calculated by dividing the absolute dense area by the total breast area multiplied by 100. If the percentage value was less than 25, the breast was classified as non-dense, and if it was 25 or more, as dense.

The tissue composition of the samples, measured by digital image analysis, was expressed as the percentage of epithelium, fatty, and non-fatty stroma. The samples were then categorized in tertiles according to the following cutoff points: 7% and 16% for the epithelium, 11% and 34% for the non-fatty stroma, and 47% and 80% for the fatty stroma.15

The complete transcriptome of snap-frozen samples was obtained using the Agilent-014850 Whole Human Genome Microarray 4 × 44K G4112F (Feature Number version) platform (GEO accession GPL4133) and the Stratagene Universal Human Reference. The expression estimates, filtered and lowess-normalized, were uploaded in GEO database.

Seven genes were selected for the study, six of which, i.e., HMGCR (3-hydroxy-3- methylglutaryl-coenzyme A [HMG-CoA] reductase), FDPS (farnesyl diphosphate synthase), FDFT1 (farnesyl-diphosphate farnesyltransferase 1), GGPS1 (geranylgeranyl diphosphate synthase 1), SQLE (squalene epoxidase), and LSS (lanosterol synthase), code for the enzymes that play an essential role in cholesterol biosynthesis, and one (SREBF2, sterol regulatory element binding transcription factor 2) codes for the transcription factor that regulates the expression of HMGCR and SQLE (Suppl. Figure 1).

The Shapiro-Wilk test, used to check the normality of the distribution of the gene expression values, indicated that not all the genes were normally distributed, except for SREBF2. Therefore, the median value and the inter-quartile range (IQR) were used to describe the expression of the genes and non-parametric tests were applied. Accordingly, the differential expression of the genes between dense and non-dense subgroups was assessed using the unpaired two-sample Wilcoxon test, while the Kruskal-Wallis test was used to evaluate the differential expression of the genes as a function of the epithelium, fatty or non-fatty stroma content. Spearman’s correlation coefficient was calculated to assess the association between the genes. The analyses were performed using the open-source software R Core Team version 4.1.2 (http://www.R-project.org), and the P-value <0.05 was considered statistically significant.

According to the cutoff defined in the original study,14 56 (47%) breasts were classified as radiologically non-dense, and 64 (53%) as radiologically dense. However, since in a preliminary analysis, two samples (one in each subgroup) showed expression values considered as extreme outliers in most gene distribution, they were excluded from the subsequent statistical analyses.

The unpaired two-sample Wilcoxon test showed that the expression of HMGCR, FDPS, SQLE and SREBF2 was significantly higher in the dense than non-dense subgroup (respectively, -1.70 vs. -1.41, P=0.0028; -1.20 vs. -1.11, P=0.0501; -3.63 vs. -3.31 P=0.0003; -0.92 vs. -0.76, P=0.0271), and the correlation analysis indicated that the positive correlation between SREBF2 and HMGCR or SQLE found in non-dense breast subgroup (SREBF2*HMGCR: r=0.27, P=0.0495; SREBF2*SQLE: r=0.48, P=0.0002), substantially increased in the dense breast subgroup (SREBF2*HMGCR: r=0.44, P=0.0004; SREBF2*SQLE: r=0.74, P<0.0001) (Figure 1).

As shown in Table 1 and Figure 2, when the samples from non-dense and dense breasts were categorized based on their fatty component, a progressive decrease in the expression of HMGCR, SQLE, and SREBF2 was found in both density subgroups following an increase in the fatty content. Specifically, with the increase in fatty content, the median value of HMGCR decreased from -1.31 (0.53) to -1.94 (0.41) (P= 0.0057) in the non-dense breast and from -1.25 (0.47) to -1.93 (0.60) (P=0.0036), the median value of SQLE decreased from -3.33 (0.47) to -4.30 (0.76) (P=0.0001) in the non-dense breast and from -3.20 (0.97) to -4.02 (0.80) (P=0.0042), and the median value of SREBF2 decreased from -0.68 (0.24) to -0.98 (0.26) (P=0.0059) in the non-dense breast and from -0.72 (0.31) to -0.87 (0.40) (P=0.0069) in the dense breast.

Conversely, the expression of LSS significantly increased in the samples with the highest fatty content, with the median value increasing from -0.47 (0.60) to 0.09 (0.85) (P=0.0002) in the non-dense breast and from -0.57 (0.32) to 0.17 (0.53) (P=0.0042).

An opposite trend was found when the samples were categorized based on the amount of the stroma component: the expression of HMGCR, SQLE, and SREBF2 progressively increased with an increase in the stroma content, while the expression of LSS decreased (Figure 3). Specifically, with an increase in the stromal component, the median value of HMGCR increased from -1.95 (0.36) to -1.28 (0.46) (P=0.0014) in the non-dense breast and from -1.89 (0.51) to -1.27 (0.45) (P=0.0114), the median value of SQLE increased from -4.29 (0.68) to -3.14 (0.97) (P=0.0001) in the non-dense breast and from -3.92 (0.90) to -3.20 (0.89) (P=0.0262), and the median value of SREBF2 increased from -0.98 (0.29) to -0.56 (0.23) (P=0.0107) in the non-dense breast and from -0.84 (0.43) to -0.76 (0.34) (P=0.0595) in the dense breast. Conversely, the median value of LSS decreased from 0.31 (0.74) to -0.64 (0.41) (P=0.0002) in the non-dense breast and from 0.06 (0.72) to -0.53 (0.54) (P=0.0250) in the dense breast.

No significant change was observed when the samples were categorized as a function of their epithelium content except for the progressive decrease in the expression of GGPS1 found in samples from dense breasts.

According to the radiological criterion, a breast is classified as dense when the epithelial plus non-fatty stroma compartment is ≥ 25% of the total area. Otherwise, the sample is classified as non-dense. In line with this assumption, the samples were reclassified based on the percentage of the epithelial, fatty and non-fatty stroma components evaluated by digital image analysis. Consequently, samples with a non-fatty stroma content < 11% (I tertiles) and an epithelium content < 16% (I and II tertiles) were reclassified as “highly fatty”, whereas samples with a non-fatty stroma content > 11% (II and III tertiles) and an epithelial content > 7% (II and III tertiles) were reclassified as “highly non-fatty”. According to this new criterion, 35% of the samples were highly fatty, and 65% were highly non-fatty.

All but one (97%) of the highly fatty samples had a fatty content > 80%, but only 67% had been classified as radiologically non-dense. Similarly, only 57% of highly non-fatty samples had been classified as radiologically dense despite a fat content < 80% in 95% of cases (Figure 4).

Statistical analysis showed that five genes were differentially expressed when highly non-fatty and highly fatty subgroups were compared. HMGCR, SQLE, and SREBF2 were more expressed in the highly non-fatty subgroup (respectively, -1.48 vs. -1.94, P<0.0001; -3.39 vs. -4.18, P<0.0001; -0.77 vs. -0.94, P=0.0103), while FDFT1 and LSS were more expressed in highly fatty one (0.28 vs. -0.60, P<0.0001 and -1.05 vs. -1.32, P=0.0459, respectively). Furthermore, the correlation analysis indicated that while in the highly non-fatty subgroup, SREBF2 was positively associated with both HMGCR (r=0.53, P<0.0001) and SQLE (r=0.73, P<0.0001), in the highly fatty subgroup, these positive correlations disappeared (SREBF2*HMGCR: r=-0.19, P=0.3026) or substantially decreased (SREBF2*SQLE: r=0.41, P=0.0173) (Figure 5).