Clinicopathological features of the SCAN-B study population
Clinicopathological features of all patients with ILC and NST in SCAN-B are detailed in Supplementary Table 1. Patients with ILC were diagnosed later than those with NST similar to the data from a large US multicenter study recently reported by us1, although the age of diagnosis was higher in SCAN-B: the median age in SCAN-B was 67 and 63 for patients diagnosed with ILC and NST, respectively, as compared to 61 (ILC) and 57 (NST). In SCAN-B, ILC were more frequently ER+ and HER2- than NST, (ILC: 98% ER+, 97% HER2- and NST: 84% ER+, 84% HER2- in NST in SCAN-B) as compared to 96% ER+, 90% HER2- in ILC and 77% ER+, 80% HER2- in NST in Oesterreich et al1). ILC had lower tumor grade than NST. Unlike prior studies1, rates of lymph node metastasis in ILC and NST were similar and quite low (<20%) in SCAN-B, and it is likely due to restriction of the study to early breast cancer.
Identifying DEGs between luminal A ILC and luminal A NST in SCAN-B
NST tumors had higher tumor purity than ILC (Figure 1A) in SCAN-B similar to other datasets9. There were 7,585 DEGs (FDR < 0.05) between luminal A ILC and luminal A NST after correcting for tumor purity, of which 3,188 genes had higher expression in ILC (Figure 1B,C, Supplementary Table 2). Figures 1D-E depict pathways enriched in DEGs in IPA and EnrichR/KEGG based pathway analysis. Of note, IPA identified several Gαs/cAMP/PKA/CREB signaling related pathways to be significantly enriched in DEGs with a positive z-score (signaling by Protein Kinase A, cAMP, CREB in neurons, Gαs, Cardiac β-adrenergic signaling) suggesting up-regulation in Luminal A ILC. KEGG pathways related to Gαs/cAMP/PKA/CREB signaling were also enriched in DEGs up-regulated in Luminal A ILC in the EnrichR pathway analysis (cAMP signaling, Adrenergic signaling).
Figure 1.
cAMP/PKA/CREB signaling is consistently up-regulated in luminal A ILC across multiple datasets and associated with survival in patients with ILC
To identify molecular alterations in ILC that are consistent across multiple datasets, we overlapped the DEGs from SCAN-B, TCGA (Supplementary Table 3) and METABRIC (Supplementary Table 4). We identified 530 and 429 genes with higher and lower expressions, respectively, in luminal A ILC vs luminal A NST tumors in all 3 datasets (Figure 2A,B, Supplementary Table 5). IPA-based pathway analysis of these 959 genes using logFC and FDR values from SCAN-B as input identified 115 pathways significantly enriched in the overlapping DEGs (log(P-value)>1.5). Hierarchical clustering of these pathways based on pairwise Jaccard similarity resulted in 5 distinct clusters (Figure 2C), including 1) Epithelial-mesenchymal transition/tumor microenvironment/wound healing, 2) Metabolism, 3) mTOR signaling, 4) EGFR and Integrins, and 5) cAMP/PKA signaling (Supplementary Table 6). Pathways from Clusters 1-4 have previously received some attention in the literature with respect to ILC biology4,9,26,27,28,29,30 whereas cluster 5 constituted of pathways related to Gαs/cAMP/PKA/CREB signaling (Figure 2D,E) that have not been discussed before in the context of ILC and hence we sought to focus on this discovery. Protein kinase A signaling, Cardiac β-adrenergic Signaling, Gαs Signaling, and cAMP-mediated signaling had positive z-scores suggesting up-regulation of Gαs/cAMP/PKA/CREB signaling in luminal A ILC. Table 1 describes the genes of Gαs/cAMP/PKA/CREB signaling that are significantly differentially expressed in all 3 datasets.
Figure 2.
Table 1. Genes relevant to cAMP/PKA/CREB signaling that are differentially expressed (FDR < 0.05) in SCAN-B, TCGA and METABRIC and their fold changes in the respective datasets
Gene
|
Description
|
Fold change in SCAN-B
|
Fold chnage in TCGA
|
Fold change in METABRIC
|
G alpha receptors
|
PTGER4
|
Prostaglandin E2 receptor
|
1.24
|
1.38
|
1.37
|
ADRB2
|
Beta adrenergic receptor
|
1.15
|
1.32
|
1.19
|
S1PR1
|
Sphingosine-1-phosphate receptor 1
|
1.33
|
1.37
|
1.24
|
G protein subunits
|
GNAS
|
G protein Subunit Alpha S
|
0.93
|
0.77
|
0.86
|
GNB1
|
G protein Subunit Beta 1
|
0.92
|
0.83
|
0.91
|
GNG7
|
G protein Subunit Gamma 7
|
1.19
|
1.45
|
1.15
|
GNG11
|
G protein Subunit Gamma 11
|
1.26
|
1.24
|
1.40
|
Adenylyl cyclase
|
ADCY4
|
Adenylate Cyclase 4
|
1.28
|
1.68
|
1.30
|
ADCY5
|
Adenylate Cyclase 5
|
1.62
|
1.53
|
1.07
|
Phosphodiesterases
|
PDE1A
|
Phosphodiesterase 1A
|
1.34
|
1.21
|
1.39
|
PDE2A
|
Phosphodiesterase 2A
|
1.47
|
1.44
|
1.11
|
PDE4B
|
Phosphodiesterase 4B
|
1.30
|
1.32
|
1.37
|
PDE9A
|
Phosphodiesterase 9A
|
1.20
|
1.34
|
1.11
|
A-kinase Anchoring Protein
|
AKAP7
|
A-kinase anchoring protein 7
|
1.17
|
1.17
|
1.19
|
AKAP8
|
A-kinase anchoring protein 8
|
1.04
|
1.22
|
1.08
|
Targets for PKA Phosphorylation
|
SRC
|
SRC Proto-oncogene
|
0.90
|
0.81
|
0.87
|
CAMK2B
|
Calcium/calmodulin dependent protein kinase II
|
0.57
|
0.68
|
0.75
|
PPARG
|
Peroxisome proliferator-activated receptor gamma
|
1.38
|
1.28
|
1.55
|
RYR3
|
Ryanodine receptor 3
|
1.76
|
1.45
|
1.09
|
PTGS2
|
Prostaglandin-endoperoxide synthase 2
|
1.23
|
1.39
|
1.30
|
LIPE
|
Hormone sensitive lipase
|
1.79
|
1.69
|
1.24
|
DUSP6
|
Dual specificity phosphatase 6
|
1.24
|
1.22
|
1.26
|
Other related genes
|
RGS2
|
Regulator of G Protein Signaling 2
|
1.16
|
1.48
|
1.25
|
RAPGEF3
|
Rap Guanine Nucleotide Exchange Factor 3
|
1.25
|
1.92
|
1.21
|
GSVA and GSEA provide alternative approaches to understanding gene expression. Indeed, pathways with a positive z-score in Jaccard cluster 5 (ie., Protein kinase A signaling, Cardiac β-adrenergic signaling, Gαs signaling, cAMP-mediated signaling) had higher GSVA scores in luminal A ILC than NST in all 3 datasets (Figure 2F). Additionally, GSVA was performed using genesets relevant to cAMP/PKA/CREB signaling downloaded from MSigDB. These genesets also had higher GSVA scores in luminal A ILC than luminal A NST in all 3 datasets (Figure 3A, Supplementary Table 7). Further, GSEA showed an up-regulation of these genesets in luminal A ILC in SCAN-B (Figure 3B).
In an RNA expression dataset from paired primary and metastatic breast cancer, primary ILC had higher GSVA scores for cAMP/PKA/CREB signaling pathways than primary NST (Supplementary figure 1A). When the GSVA scores were compared between paired primary breast tumors and metastatic lesions (ILC and NST combined), brain and ovarian metastases respectively had significantly higher and lower GSVA scores for some of the pathways related to cAMP/PKA/CREB signaling (Supplementary figure 1B).
There was a trend towards worse DSS in patients with luminal A ILC who had higher GSVA scores (>75th or >50th percentile) in three out of the eight cAMP/PKA/CREB signaling related pathways in METABRIC (Figure 3C-E; Supplementary Table 8). Logrank testing showed a trend towards worse DSS based on expression of GOBP regulation of cAMP (P=0.07, Figure 3C) and significantly worse DSS based on expression of KEGG cAMP mediated signaling (P=0.05, Figure 3D) and IPA’s Cardiac beta-adrenergic signaling (P=0.04, Figure 3E). A similar analysis did not reveal any significant association with DSS in patients with luminal A NST (Supplementary Table 8).
Figure 3.
cAMP/PKA/CREB signaling and forskolin mediated phosphorylation of CREB are higher in ILC cell lines
Given the bioinformatic findings showing enhanced cAMP/PKA/CREB signaling in ILC tumors, we investigated these findings using in vitro models. We first compared the expression of cAMP/PKA/CREB related pathways in ILC (HCC2185, MDAMB134, MDAMB330, SUM44PE) or ILC-like (MPE600, CAMA1, HCC2218, MDAMB453, OCUBM, SKBR5, ZR7530) vs NST (BT474, BT483, EFM19, HCC1008, HCC1419, HCC1500, LY2, MCF7, MDAMB157, MDAMB175, T47D, UACC812, ZR75-1, ZR75B) cell lines using publicly available gene expression data17. Of the four pathways from Cluster 5 of Jaccard clustering with positive z-score (PKA signaling; Cardiac β-adrenergic Signaling; Gαs Signaling, cAMP-mediated signaling) and the four MSigDB pathways with higher GSEA/GSVA scores in luminal A ILC than luminal A NST in SCAN-B (KEGG cAMP mediated signaling; REACTOME PKA mediated phosphorylation of CREB; GOBP cellular response to cAMP; GOBP regulation of cAMP mediated signaling), GSVA scores for GOBP cellular response to cAMP, GOBP regulation of cAMP mediated signaling and KEGG cAMP mediated signaling (Figure 4A) were higher in ILC and ILC-like cell lines as compared to NST cell lines (statistically significant for GOBP regulation of cAMP mediated signaling, P = 0.03). The GSVA scores for the other 5 pathways tested were similar in ILC/ILC-like and NST cell lines (Supplementary Figure 2)
We next used an activator of cAMP/PKA/CREB pathway, forskolin, which stimulates adenylate cyclase and thereby increases intracellular cAMP levels31 for in vitro studies. Treatment of ILC/ILC-like and NST cell lines with forskolin for 20 minutes increased CREB phosphorylation at Ser133 (Figure 4B) and did not alter the expression levels of CREB or PKA-Cα compared with vehicle treatment (Figure 4C). The forskolin induced phosphorylation of CREB was 2-fold higher in ILC/ILC-like cell lines as compared to NST. ILC/ILC-like cell lines incubated overnight in low serum media also demonstrated enhanced CREB phosphorylation than NST (Supplementary Figure 3) indicating that ILC/ILC-like cell lines have higher cAMP/PKA/CREB signaling irrespective of the growth media. Baseline level of p-CREB was higher in ILC/ILC-like cell lines than NST in low serum media (Supplementary Figure 4).
We next studied the effect of forskolin treatment on the levels of CREB, p-CREB and PKA-Cα in NST and ILC PDOs. The histology of the PDOs was validated by presence or absence of E-cadherin (Supplementary Figure 5). Akin to cell lines, forskolin treatment resulted in higher levels of p-CREB in ILC than NST PDOs, although this was not statistically significant (Figure 4D, Supplementary Figure 5). CREB levels were also higher with forskolin treatment in ILC PDOs. Together, these data suggest increased activity of cAMP/PKA/CREB signaling in ILC in vitro models.
Given that activated PKA-Cα localizes to the nucleus to phosphorylate CREB, we performed co-immunofluorescence of PKA-Cα and p-CREB on representative NST and ILC/ILC-like cell lines. Following forskolin treatment, PKA-Cα was more evident in the nucleus and the staining of nuclear p-CREB was enhanced across the tested cell lines (Supplementary Figure 6) with no obvious differences among them.
Loss of E-cadherin in MCF7 and T47D with CRISPR-mediated deletion of CDH1 did not increase phosphorylation of CREB with Forskolin (Supplementary Figure 7), suggesting that E-cadherin loss is not sufficient for increased cAMP/PKA/CREB signaling in ILC.
Figure 4.
Forskolin inhibits in vitro cell growth
To understand the effect of increased cAMP/PKA/CREB signaling on proliferation of breast cancer cells, we performed growth assays in 6 ILC cell lines (SUM44PE, MM134, MM330, WCRC25, MM453, HCC2185) and 4 NST cell lines (MCF7, T47D, MM415, BT474) treated with forskolin. Forskolin has anti-cancer and anti-proliferative effects in multiple cancer types32 and augments the cytotoxicity of chemotherapy agents33. Increasing forskolin doses inhibited cell growth in all breast cancer cell lines (Figure 5A), with a lower IC50 in the ILC cell lines (median IC50 27µM) compared to the NST cell lines (median IC50 78.5 µM, P = 0.07, Figure 5B).
Figure 5.