Introduction

Approximately 80% of postmenopausal breast cancer are estrogen receptor (ER)-positive and, for this reason, usually treated with endocrine therapy. Tamoxifen has been the conventional long-term treatment for several years. Then, to overcome the emergence of tumor resistance, it was first flanked and then replaced by aromatase inhibitors (AIs) both in operable and advanced ER-positive breast cancer achieving excellent results as progression-free and survival time [1–3].

Despite the recognized superiority of AIs over tamoxifen, the response rate in the neoadjuvant setting is only 50% to 70% because of an acquired or de novo resistance to estrogen deprivation therapy. Several putative mechanisms of resistance, mainly based on in vitro experiments, have been proposed. They include intrinsic resistance of tumors to estrogen, aromatase-independent estrogenic hormones, signal transduction by non-endocrine pathways, and selection of hormone-insensitive clones during AI therapy [4, 5], none of which are entirely convincing.

Cholesterol is an essential component of cell membranes because required for membrane composition and fluidity. In addition to this activity at the membrane level, cholesterol is the precursor in several biochemical pathways, including the synthesis of steroid hormones. Generally supplied by the bloodstream, the de novo cholesterol biosynthesis can be activated in response to the increased cell requirement during the physiological tissue remodeling and the proliferation of cancer cells, especially if steroid hormone-dependent. In these latter, it concurs in promoting and sustaining cell growth as the precursor of steroid hormones and may contribute to the failure of the treatment with AIs.

Recent findings [6] indicated that the mevalonate pathway, the core of the biosynthetic process leading to cholesterol, may interplay with the Hippo signaling pathway through the geranylgeranyl pyrophosphate (GG-PP) that plays an essential role in protein isoprenylation, a process which allows some proteins, including small guanosine triphosphatases (GTPases), such as cell division cycle 42 (Cdc42), ras-related C3 botulinum toxin substrate 1 (Rac1), and ras homolog family member A (RhoA), to be anchored at the plasma membrane and enzymatically activated.

Cumulating evidence indicated that the augmented production of isoprenoids, due to the overexpression of the gene coding for the 3-hydroxy-3-methylglutaryl-coenzyme A (HMG-CoA) reductase and the following increase in mevalonic acid level, results in the isoprenylation and activation of the small GTPases that interplay with the Hippo signaling pathway by controlling the phosphorylation of Yes-associated protein (YAP)/transcriptional co-activator with PDZ-binding motif (TAZ), the main downstream effectors [7]. Once anchored to the plasma membrane by isoprenylation and activated, GTPases inhibit YAP/TAZ phosphorylation allowing the translocation into the nucleus, the binding to transcriptional enhancer factor (TEA) domain family member 4 (TEAD) transcription factors, and the expression of a wide variety of genes involved in essential physiological processes, such as stem cell functions, cell proliferation and differentiation, organ growth, tissue regeneration, and embryonic development [8–11].

Therefore, the experimental findings indicating that the abnormal accumulation of YAP/TAZ into the nucleus may promote the neoplastic transformation by stimulating cell proliferation, inhibiting apoptosis, inducing the acquisition of additional stem cell properties, and supporting cancer cell dissemination by reducing cell contact inhibition, and triggering epithelial to mesenchymal transition, are expected [12–15]. Likewise, it is unsurprising that several solid tumors, including breast cancer (especially those poorly differentiated), show a nuclear accumulation of YAP/TAZ [16, 17].

The studies also showed that, in tumor cells, the inhibition of the HMG-CoA reductase activity by statins, causing isoprenoid depletion and the unsuccessful activation of small GTPases, promoted the phosphorylation of YAP/TAZ, which is sequestered in cytoplasm and degraded, thus blocking gene transcription [18, 19].

The aim of this in silico study was to compare the expression of the genes involved in the biosynthesis of cholesterol in responsive and nonresponsive tumors from patients treated with a neoadjuvant third-generation AI (letrozole) and explore their association with the genes involved in estrogens production, Hippo signaling pathway, and cell cycle control to verify the potentiality of statins as an additional treatment to overcome the resistance to AIs.

Materials and methods

The study was carried out using a publicly accessible dataset retrieved from the NCBI Gene Expression Omnibus (GEO) database (https://www.ncbi.nlm.nih.gov/geo/), identified by the GEO accession number GSE5462, and that responded to the following selection criteria: an adequate number of sequential biopsies from the same tumor before and after a short treatment with letrozole; availability of information about the clinical response; gene expression profile obtained using a commercial and therefore reproducible platform.

As described in the original articles [20, 21], postmenopausal patients with large operable or locally advanced breast cancer were recruited prospectively into a neoadjuvant study with the AI letrozole (2.5 mg/day, oral). Women with ER-positive breast cancer had tumor biopsies taken before and after 10 to 14 days of treatment and were assessed for clinical response (based on changes in tumor volume) at three months. Fifty patients had available paired tumor samples: 36 responsive and 14 nonresponsive tumors. All patients consented to provide excess tissues for research purposes, and the study was approved by the ethics committee of the University of Edinburgh, Edinburgh, UK (2001/W/BU/09 and 2001/W/BU/10).

Gene expression data, obtained using Affymetrix Human Genome U133A Array (GEO accession GPL96), were provided as filtered and log₂ transformed expression estimates.

Gene set selection

According to the aim of the study, a panel of 37 genes was chosen that included: 13 genes coding for the enzymes involved in the biosynthesis of cholesterol (Figure 1), 9 involved in the synthesis of estrogens from cholesterol (Figure S1), 10 coding the essential elements of the Hippo signaling pathway and some representative YAP/TAZ target genes [namely, baculoviral inhibitor of apoptosis (IAP) repeat containing 5 (BIRC5), cellular communication network factor 1 (CCN1), and CCN2] [22, 23] (Figure S2), and 5 [cyclin D1 (CCND1), cyclin dependent kinase 4 (CDK4), CDK6, CDK inhibitor 1A (CDKN1A), and CDKN1B] involved in cell cycle control. The latter genes were chosen on the notion that the estrogen-driven activation of the cyclin D-CDK4/CDK6 complex is essential for the physiologic mammary development and proliferation of ER-positive breast cancers [24, 25], and CDKN1A (P21, Cip1) and CDKN1B (P27, Kip1) are pivotal players in the control of the G₁-phase of the cell cycle (details on the selected genes in Table S1).

Display full size

Figure 1. 

Schematic diagram depicting the enzymes involved in cholesterol biosynthesis (in parenthesis, the corresponding gene). Cholesterol biosynthesis is regulated mainly by the SREBP2, which controls the expression of several genes involved in the biosynthetic pathway, including HMGCR, FDFT1, and SQLE. ACAT2: acetyl-coenzyme A (CoA) acetyltransferase 2; HMGCS1: HMG-CoA synthase 1; HMGCR: HMG-CoA reductase; MVK: mevalonate kinase; PMVK: phosphomevalonate kinase; MVD: mevalonate PP decarboxylase; FDPS: farnesyl diphosphate synthase; FDFT1: farnesyl-diphosphate farnesyltransferase 1; SQLE: squalene epoxidase; LSS: lanosterol synthase; GGPS1: GG diphosphate synthase 1; SREBP2: sterol regulatory element-binding protein-2; SREBF2: sterol regulatory element-binding factor-2; NSDHL: nicotinamide adenine dinucleotide (phosphate) dependent steroid dehydrogenase-like

Results

Expression of the genes involved in cholesterol biosynthesis in responsive and nonresponsive tumors

Wilcoxon paired test showed a significant decrease in the expression level of several genes involved in cholesterol biosynthesis in responsive tumors. In particular, a decreased expression level was observed in ACAT2, which encodes the first enzyme of the biosynthetic chain, HMGCS1 and HMGCR, which encode the two enzymes that trigger the mevalonate pathway, FDPS, which codes for the enzyme that catalyzes the synthesis of farnesyl PP (FPP), the precursor for the synthesis of isoprenoids, and SQLE, which encodes the key enzyme for cholesterol commitment (Figure 2). Conversely, no decrease was observed in nonresponsive tumors.

Display full size

Figure 2. 

Differential expression of ACAT2, HMGCS1, HMGCR, FDPS, and SQLE gene before and after a 2-week treatment with letrozole in responsive and nonresponsive tumors. Wilcoxon signed-rank test for paired samples was applied. N.S.: non-significant

Basal association among the five genes modulated by letrozole and the other genes involved in cholesterol biosynthesis in responsive and nonresponsive tumors

PCA indicated that the basal expression of the five genes modulated by letrozole showed a different pattern of expression in responsive and nonresponsive tumors (Figure 3).

Display full size

Figure 3. 

Biplots visualizing the relationship among the genes downregulated by the letrozole treatment (“downregulated”) and the other genes involved in the biosynthesis of the cholesterol (“unaffected”) in responsive and nonresponsive tumors. Correlation between variables is represented by arrows (vectors) within the correlation circle of radius 1. The length of each vector approximates the quality of the representation of the variable: the longer the vector better the variable representation. The position of vectors represents their pairwise correlation: Vectors close to one another are positively correlated, whereas negatively correlated variables are positioned on the opposite side. If vectors are orthogonal, the variables have a small or null correlation

Indeed, in responsive tumors, HMGCR was associated positively with HMGCS1 (r = 0.56, P = 0.0004), and both with MVK (r = 0.54, P = 0.0007, and r = 0.59, P = 0.0002, respectively) and MVD (r = 0.35, P = 0.0386 and r = 0.49, P = 0.0023, respectively); besides, HMGCR and HMGCS1 were associated negatively with NSDHL (r = –0.58, P = 0.0002 and r = –0.56, P = 0.0003, respectively). ACAT2 was associated positively with SQLE (r = 0.41, P = 0.0123), while FDPS was positively associated with PMVK (r = 0.74, P < 0.0001), GGPS1 (r = 0.39, P = 0.0186) and NSDHL (r = 0.35, P = 0.0329).

Conversely, in nonresponsive tumors, ACAT2, HMGCR, and FDPS were associated positively with each other (specifically, ACAT2 vs. HMGCR: r = 0.53, P = 0.0567; ACAT2 vs. FDPS: r = 0.46, P = 0.0972; HMGCR vs. FDPS: r = 0.52, P = 0.0591), and with SQLE (r = 0.43, P = 0.1239; r = 0.34, P = 0.2334; r = 0.44, P = 0.1198, respectively). Besides, ACAT2 and HMGCR were associated negatively with FDFT1 (r = –0.50, P = 0.0704 and r = –0.44, P = 0.1154, respectively). Regarding HMGCS1, it was associated positively with FDFT1 (r = 0.59, P = 0.0271) and MVK (r = 0.39, P = 0.1625) and negatively with PMVK (r = –0.36, P = 0.2129), while FDPS, in addition to the association with ACAT2 and HMGCR, was associated positively with PMVK (r = 0.44, P = 0.1195). SQLE, in addition to the association with ACAT2, HMGCR, and FDPS, was associated positively with NSDHL (r = 0.60, P = 0.0249) and negatively with MVK (r = –0.46, P = 0.1017) and FDFT1 (r = –0.40, P = 0.1510).

Basal relationship between ACAT2, HMGCS1, HMGCR, FDPS, or SQLE and the genes involved in estrogens biosynthesis in responsive and nonresponsive tumors

The basal relationship between the genes downregulated in responsive tumors and those involved in the production of estrogens differed in nonresponsive tumors compared to responsive ones (Figure 4). In responsive tumors, HMGCS1 and HMGCR were associated positively with hydroxy-delta-5-steroid dehydrogenase, 3 beta- and steroid delta-isomerase 2 (HSD3B2, r = 0.58, P = 0.0002 and r = 0.47, P = 0.0036, respectively) and steroid sulfatase (STS, r = 0.55, P = 0.0006 and r = 0.46; P = 0.0045, respectively); HMGCS1 was also associated positively with cytochrome P450 family 11 subfamily A member 1 (CYP11A1, r = 0.49, P = 0.0023) and CYP17A1 (r = 0.38, P = 0.0218); HMGCR was associated negatively, respectively, with sulfotransferase family 1A member 1 (SULT1A1, r = –0.61, P < 0.0001) and FDPS was associated negatively, with STS (r = –0.58, P = 0.0002).

Display full size

Figure 4. 

Biplots visualize the relationship among the genes downregulated by the letrozole treatment (“cholesterol”) and the genes involved in the synthesis of estrogens (“estrogens”) in responsive and nonresponsive tumors. Correlation between variables is represented by arrows (vectors) within the correlation circle of radius 1. The length of each vector approximates the quality of the representation of the variable: the longer the vector better the variable representation. The position of vectors represents their pairwise correlation: Vectors close to one another are positively correlated, whereas negatively correlated variables are positioned on the opposite side. If vectors are orthogonal, the variables have a small or null correlation. HSD17B2: hydroxysteroid 17-beta dehydrogenase 2

Conversely, in nonresponsive tumors, ACAT2 and HMGCR were associated positively with CYP19A1 (r = 0.53, P = 0.0522 and r = 0.49, P = 0.0809, respectively) which, conversely, was associated negatively with HMGCS1 (r = –0.59, P = 0.0249). Besides, while HMGCS1 was associated positively with CYP11A1 (r = 0.69, P = 0.0064) and CYP17A1 (r = 0.61, P = 0.0199) similarly to responsive tumors, FDPS and SQLE were associated negatively with CYP17A1 (r = –0.43, P = 0.1239 and r = –0.50, P = 0.0694, respectively) and SQLE was associated negatively also with CYP11A1 (r = –0.48, P = 0.0840). Furthermore, HMGCR and FDPS were associated negatively with HSD17B2 (r = –0.39, P = 0.1656, and r = –0.55, P = 0.0423, respectively), and FDPS was associated negatively also with STS (r = –0.43, P = 0.1281).

Basal relationship between ACAT2, HMGCS1, HMGCR, FDPS, or SQLE and the genes involved in the Hippo signaling pathway in responsive and nonresponsive tumors

PCA showed that, in the nonresponsive tumors, the expression of the genes downregulated in responsive tumors had a different association profile with the genes involved in the Hippo pathway compared to responsive tumors (Figure 5).

Display full size

Figure 5. 

Biplots visualizing the relationship among the genes downregulated by the letrozole treatment (“cholesterol”) and the genes involved in the Hippo pathway (“hippo”) in responsive and nonresponsive tumors. Correlation between variables is represented by arrows (vectors) within the correlation circle of radius 1. The length of each vector approximates the quality of the representation of the variable: the longer the vector better the variable representation. The position of vectors represents their pairwise correlation: Vectors close to one another are positively correlated, whereas negatively correlated variables are positioned on the opposite side. If vectors are orthogonal, the variables have a small or null correlation. LATS1: large tumor suppressor kinase 1; MOB1A: Mps one binder (MOB) kinase activator 1A; SAV1: salvador family WW domain containing protein 1; STK3: serine/threonine kinase 3; WWTR1: WW domain containing transcription regulator 1; YAP1: Yes1 associated transcriptional regulator

In particular, in nonresponsive tumors, HMGCR and FDPS were associated positively with YAP1 (r = 0.38, P = 0.1767 and r = 0.43, P = 0.1237, respectively), and HMGCR showed a positive association also with BIRC5 (r = 0.49, P = 0.0764), one of most representative genes regulated by the YAP/TAZ complex. Besides, ACAT2 and FDPS were associated positively with STK3 (r = 0.70, P = 0.0058 and r = 0.43, P = 0.1216, respectively), and FDPS and SQLE were associated negatively with STK4 (r = –0.46, P = 0.0945 and r = –0.39, P = 0.1635, respectively).

Conversely, in responsive tumors, HMGCS1 and HMGCR were associated positively with STK3 (r = 0.62, P < 0.0001 and r = 0.38, P = 0.0210, respectively), and LATS1 (r = 0.61, P < 0.0001 and r = 0.38, P = 0.0211, respectively).

Basal relationship between ACAT2, HMGCS1, HMGCR, FDPS, or SQLE and the genes involved in cell cycle control in responsive and nonresponsive tumors

As shown in Figure 6, in nonresponsive tumors, all but the HMGCS1 gene were associated negatively with CDK6 (specifically, ACAT2: r = –0.56, P = 0.0377; HMGCR: r = –0.52, P = 0.0546; FDPS: r = –0.54, P = 0.0477; SQLE: r = –0.62, P = 0.0184). Besides, in addition to the positive association with CDK6 (r = 0.47, P = 0.0918), HMGCS1 was associated negatively with CCND1 (r = –0.49, P = 0.0756) and CDK4 (r = –0.44, P = 0.1129). In responsive tumors, FDPS was associated positively with CDK4 (r = 0.40, P = 0.0145).

Display full size

Figure 6. 

Biplots visualizing the relationship among the genes downregulated by the letrozole treatment (“cholesterol”) and the genes involved in cell cycle control (“proliferation”) in responsive and nonresponsive tumors. Correlation between variables is represented by arrows (vectors) within the correlation circle of radius 1. The length of each vector approximates the quality of the representation of the variable: the longer the vector better the variable representation. The position of vectors represents their pairwise correlation: Vectors close to one another are positively correlated, whereas negatively correlated variables are positioned on the opposite side. If vectors are orthogonal, the variables have a small or null correlation

Discussion

Present results indicated that in responsive tumors, the treatment with letrozole was associated with a statistically significant decrease in the expression level of five genes that play a crucial role in the biosynthesis of cholesterol: ACAT2, which encodes the first enzyme of the biosynthetic chain; HMGCS1 and HMGCR, which encode the two enzymes that initiate the mevalonate pathway; FDPS, which codes for the enzyme that catalyzes the two-steps condensation of dimethylallyl PP with two molecules of isopentenyl PP (IPP) to form FPP, the precursor of GG-PP; SQLE, which encodes the enzyme that commits cholesterol biosynthesis irreversibly.

This finding could appear unexpected given the specific mechanism of action of letrozole, which inhibits the aromatase by competitively binding to the heme of the cytochrome P450 of the enzyme. However, because some of these genes (especially HMGCR and FDPS) [27, 28] are regulated by estrogen through a functional estrogen response element (ERE) in the promoter region, the decrease in the expression level observed in responsive tumors can be ascribed to the inhibition of the aromatase activity that reduced the estradiol availability. This hypothesis agrees with the finding that, in nonresponsive tumors, the five genes had no evidence of modulation but, taken as a whole, they were associated with several genes involved in estrogen biosynthesis. Indeed, they associated positively with CYP19A1, the gene coding for the aromatase, and negatively with HSD17B2 and SULT1A, which code, respectively, the enzyme that converts the estradiol to the less active estrone and the enzyme that neutralizes the estradiol by converting it into estradiol-sulfate. That is in agreement with the functional cooperation of cholesterol biosynthesis and estradiol production in the promotion and support of cancer cell proliferation: Estradiol acts as a potent growth factor for ER-positive cancer cells, and cholesterol plays the double role of an essential component of cell membranes and precursor for estrogen biosynthesis. Therefore, it is not surprising that, in nonresponsive tumors, the expression of the five genes was associated with that of the genes involved in cell cycle control, and in particular, with CDK4 and CDK6, the genes coding the two CDKs that drive the G₁-phase of the cell cycle. Indeed, all but HMGCS1 genes were associated positively with CDK4 and negatively with CDK6, recently recognized as a novel downregulated estrogen-responsive gene that prevents cell proliferation and regulates cell cycle exit during cell differentiation [29].

In addition to the association with the genes involved in cell proliferation, HMGCR, and FDPS were associated positively with YAP1, which codes for the main transcriptional coactivator regulated by the Hippo signaling pathway (Figure 7), and BIRC5, one of the most representative YAP/TAZ-regulated gene that codes for survivin, an anti-apoptosis protein, overexpressed in most tumors including breast cancer [30].

Display full size

Figure 7. 

Correlation between HMGCR or FDPS and YAP1, BIRC5, STK4, or LATS1 in nonresponsive tumors (evaluated by using the nonparametric Spearman’s correlation coefficient test)

The positive association between HMGCR or FDPS and YAP1 is of particular interest because, on one side, confirms the functional role of the mevalonate pathway on the activity of the YAP coactivator, through the production of isoprenoids that, anchoring to the membrane and activating the small GTPases, inhibits the phosphorylation of YAP and the following sequestration in the cytoplasm; on the other side, they suggest a relationship between mevalonate and Hippo pathways also at the genomic level. Indeed, the expression of FDPS, which is the regulatory node in the production of isoprenoids, other than associated positively with that of YAP1, was associated negatively with that of STK4 and LATS1, which code for the two STKs that concur in the inactivation of YAP by phosphorylation and can be inhibited by prenylated GTPases (Figure 7). This result is in agreement with the evidence that the expression levels of the kinases involved in the Hippo signaling pathway are downregulated in human breast cancer tissues compared with normal breast tissues [31], and high expression levels of STKs are associated with longer disease-free and overall survival [32]. Besides, some recent studies [33, 34] showed that LATS1 kinase controls human breast cell fate by promoting the ER ubiquitination and ubiquitin-mediated proteasomal degradation and that the YAP/TEAD complex can activate the expression of estrogen-regulated genes by binding to the enhancer region of ERα, thus provide solid pieces of evidence of a close interrelationship among cholesterol biosynthesis, estrogens production, and the Hippo signaling pathway.

Remarkable is also the finding that, in nonresponsive tumors, the HMGCS1 gene, which codes the enzyme turning acetoacetyl-CoA into HMG-CoA (the direct precursor of the mevalonic acid), lost the positive association with HMGCR while retaining the positive association with CDK6 and MOB1A and the negative association with CCND1 and CDK4. That corroborates the notion that the dysregulation of the gene coding for the rate-limiting enzyme HMG-CoA reductase causes, on one side, the abnormal production of bioactive estrogens and, on the other side, the synthesis of the isoprenoids necessary for the isoprenylation of the small GTPases and the subsequent inhibition of YAP inactivation, provides a convincing explanation for the tumor resistance to AIs and supports the use of statins to contrast this metabolic dysfunction. In agreement with this idea, a recent population-based cohort study [35] showed improved patient survival and reduced risk for contralateral breast cancer in postmenopausal women assuming statins for more than five years with conventional AI therapy.

Several treatments have been investigated to overcome tumor resistance to AIs, and the CDK4/CDK6 inhibitors provided the most encouraging results [36–39]. Accordingly, specific CDK4/CDK6 inhibitors such as palbociclib and abemaciclib have become the standard of care in either the adjuvant or second-line setting for patients with ER-positive breast cancer in combination with AI therapy [40–42]. However, despite the substantial improvement in disease-free and overall survival, most tumors fail to respond, and patients ultimately experience disease progression because of acquired resistance [43]. Currently, excluding chemotherapy, no alternative strategies are available to overcome the resistance to CDK4/CDK6 inhibitors.

The cumulating evidence shows that the inhibition of the mevalonate pathway by statins caused the accumulation of the YAP coactivator in the cytoplasm hampering its nuclear translocation and the subsequent transcription of the YAP/TAZ-dependent genes [6, 44, 45] and that statins were effective in preventing and treating several tumors, including breast cancer [46–49], increased the interest for this class of molecules, initially developed as cholesterol-lowering drugs, for the treatment of ER-positive breast cancer with the double advantage of reducing the production of bioactive estrogens in an AI-independent manner and inhibiting the YAP/TAZ-dependent transcriptional activity.

In conclusion, although obtained on a limited number of patients, present results indicated that in nonresponsive tumors, the expression of ACAT2, HMGCS1, HMGCR, FDPS, and SQLE genes, which play a crucial role in the biosynthesis of cholesterol and isoprenoids production was unaffected by the treatment with letrozole and associated with that of some genes involved in estrogens production (CYP19A1, HSD17B2, and SULT1A), cell cycle control (CDK4 and CDK6), and Hippo signaling pathway (YAP1 and BIRC5), corroborating the notion that the dysregulation of the mevalonate pathway contributes to the resistance to letrozole and supporting the clinical use of statins to contrast this metabolic dysfunction.