Our results show that DCISpure has the most distinct gene expression profile from normal breast epithelium. We observed 2 differentially expressed genes between DCISpure and IDCpure and no differentially expressed genes between DCIScomp and IDCcomp. Six differentially expressed genes were found in DCISpure vs DCIScomp, of which 3 were also differentially expressed between control and DCIScomp, but not between control and DCISpure. Also, the same 3 genes (FGF2, GAS1, and SFRP1) showed distinct gene expression profiles between noninvasive and invasive groups. Thus, suggesting their involvement in DCIS progression, possibly by acquisition of invasive capacity after downregulation.
Other studies also have shown that in-situ and invasive stages of breast ductal carcinoma are similar in gene expression, suggesting that invasiveness necessary changes are already present in the in-situ lesions that will later progress to invasive ones. Therefore, DCIS should acquire enabling gene expression changes much before morphological alterations and invasiveness [5, 18, 19]. In this study, we found only 2 differentially expressed genes between IDCpure and DCISpure.
Interestingly, the in-situ stage (DCISpure) has more molecular differences with control than the invasive stage (IDCpure). However, considering that IDC is the most advanced stage in progression and morphology, we expected greater molecular changes in reference to non-neoplastic tissue.
Our result is probably due to early acquisition of tumor enabling features, which are later followed by minor ones . We believe that IDC loses some of the initial differentially expressed genes, therefore becoming more like normal tissue expression wise.
DCIScomp and IDCcomp of patients with DCIS-IDC do not have differentially expressed genes between them and are more expression wise distant from control than IDCpure. Initial gene expression changes may remain necessary in DCIS-IDC since acquisition of invasive potential has not yet been completed in all cells. Also, as suggested by Hu et al.  and Muggerud et al.  many processes involved in DCIS progression may be expression changes in the tumor microenvironment, and not necessarily in tumor cells .
We propose that the expression differences between DCISpure and DCIScomp would identify genes involved in DCIS progression. The 3 differentially expressed genes more likely involved in DCIS progression were FGF2, GAS1, and SFPR1, all downregulated in DCIScomp. This fact suggests that progression from DCISpure to DCIScomp may use silencing mechanisms more often than activating ones.
Our results were able to detect expression changes during the progression of DCISpure to DCIScomp and, finally, to IDC. Expression differences were larger when comparing DCIScomp vs control (104 differentially expressed genes), decreasing with DCIScomp vs DCISpure (6 differentially expressed genes) and DCIScomp vs IDCpure (2 differentially expressed genes).
When comparing differentially expressed genes between control and DCISpure, 31% are driver genes, whereas none of the genes that may be involved in DCIS progression or differentially expressed genes between DCISpure and IDCpure is driver genes, suggesting that major alterations occur at the beginning of carcinogenesis and not at the end.
To confirm gene involvement in invasion, we created two groups: a noninvasive group composed of control tissue and DCISpure and an invasive group composed of DCIS-IDC and IDCpure. In this analysis FGF2, GAS1, and SFPR1 were downregulated in the invasive group.
Epigenetic alterations may contribute to BC progression by transcriptionally silencing specific tumor suppressor genes [22, 23], which could explain the loss of expression that we observed. DNA hypermethylation of tumor suppressor genes is commonly observed during tumor progression [24, 25] and may be associated with DCIS progression . Studies have shown that a subset of genes is methylated during progression, reinforcing the role of the expression downregulation during BC initiation and/or progression [26-30]. In Conway et al.  study, FGF2 was among the relatively hypermethylated genes in hormone receptor-positive, luminal A, or p53 wild-type BCs and Bediaga et al.  showed that FGF2 displayed higher hypermethylation levels in the luminal B BC subtype. Lo et al.  related that SFRP1 gene was frequently hypermethylated in ductal carcinomas. Veeck et al.  found a tight correlation between promoter hypermethylation and SFRP1 downregulation in primary BC tissue. Studies about GAS1 hypermethylation are seen in lung, pancreatic and prostate cancers .
FGF2 gene plays an important role in angiogenesis . Its signaling pathway is influenced by crosstalk with integrins . Lower levels of FGF2 mRNA have been detected in BC when compared to normal tissues . The low expression also occurs in large tumors, late disease stages, and worse overall and disease-free survival [39, 40]. Pre-clinical studies showed that FGF2 overexpression could inhibit tumor growth [41, 42]. Also, many in vitro assays have demonstrated a potent inhibitory effect of FGF2 on BC cells, possibly involving the mitogen-activated protein kinase (MAPK) cascade and cell cycle G1/S transition [43-45]. Researchers have suggested that intermediate levels of adhesion between FGF2 and integrins are needed for optimal cell migration, which could explain the reduction in FGF2 expression during BC progression [46-48]. Enrichment analysis has shown statistically significant interactions between FGF2 and MAPK pathway genes and other components of the FGF family. UALCAN analysis has shown an upregulation of FGF2 in normal tissues, in comparison to primary BC and that FGF2 downregulation is associated with tumor progression. FGF2 involvement in BC progression is predominantly related to cell growth regulation .
GAS1 gene plays a role in growth suppression inhibiting DNA synthesis  and reducing cancer cell line proliferation . In BC cells, Jimenez et al.  proposed that this gene could inhibit growth by decreasing angiogenesis. GAs1 is more expressed in normal tissues, in comparison to primary breast tumor tissue, as seen in TCGA data. Hedgehog (Hh) signaling has been suggested as a critical determinant of tumor progression [52-55]. A progressive increase of Hh expression and Hh pathway activation has been observed from control, DCIS, DCIS with microinvasion and to IDC [56, 57]. GAS1 protein binds Sonic hedgehog (SHH), one of three Hh proteins, and may inhibit Hh signaling [58, 59]. The interaction of GAS1 with SHH was seen but was not statistically significant. The reduction of GAS1 expression during more advanced BC stages may occur to avoid Hh pathway inhibition.
SFRP1 gene is a negative regulator of the Wnt pathway, which is aberrantly activated in BC [34, 60, 61]. SFRPs are cell membrane receptors able to bind Wnt proteins in the extracellular compartment, inhibiting ligand-receptor interaction and signal transduction . Statistically significant interactions of SFRP1 with Wnt pathway genes were seen and enrichment analysis showed a negative regulation of canonical Wnt receptor signaling pathway. SFRP1 mRNA is strongly downregulated in BC, having a putative tumor suppressor role. Results of Gauger et al.  suggests that loss of SFRP1 expression allows non-malignant cells to acquire BC tumor features. We saw SFRP1 downregulation in primary BC in comparison to normal tissue. Lower SFRP1 expression was seen in invasive lesions.
Functional analyses of FGF2, GAs1 and SFRP1 suggests a role in DCIS progression, being negative regulators of cell cycle G1/S transition, Hh signaling, and the Wnt pathway, respectively. We propose that downregulation favors DCIS progression. Unfortunately, our samples could not be divided into high and low-grade DCIS, nor could we study samples according to cancer molecular subtypes. Studying these groups separately may reveal important events in the DCIS progression.