The molecular underpinning of lobular histological growth pattern: a genome-wide transcriptomic analysis of invasive lobular carcinomas and grade- and molecular subtype-matched invasive ductal carcinomas of no special type

Authors

  • Britta Weigelt,

    Corresponding author
    1. The Netherlands Cancer Institute, 1066 CX Amsterdam, The Netherlands
    Current affiliation:
    1. Cancer Research UK, London Research Institute, London, WC2A 3PX, UK.
    • Cancer Research UK, London Research Institute, Signal Transduction Laboratory, 44 Lincoln's Inn Fields, London, WC2A 3PX, UK.
    Search for more papers by this author
    • These authors contributed equally to this study.

  • Felipe C Geyer,

    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    Search for more papers by this author
    • These authors contributed equally to this study.

  • Rachael Natrajan,

    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    Search for more papers by this author
  • Maria A Lopez-Garcia,

    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    2. University Hospital Virgen del Rocio, Department of Pathology, Seville, Spain
    Search for more papers by this author
  • Amar S Ahmad,

    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    Search for more papers by this author
  • Kay Savage,

    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    Search for more papers by this author
  • Bas Kreike,

    1. The Netherlands Cancer Institute, 1066 CX Amsterdam, The Netherlands
    Search for more papers by this author
  • Jorge S Reis-Filho

    Corresponding author
    1. Molecular Pathology Team, The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, London, SW3 6JB, UK
    • The Breakthrough Breast Cancer Research Centre, Institute of Cancer Research, 237 Fulham Road, London, SW3 6JB, UK.
    Search for more papers by this author

  • No conflicts of interest were declared.

Abstract

Invasive lobular carcinoma (ILC) is the most frequent special type of breast cancer. The majority of these tumours are of low histological grade, express hormone receptors, and lack HER2 expression. The pleomorphic variant of ILCs (PLCs) is characterized by atypical cells with pleomorphic nuclei and is reported to have an aggressive clinical behaviour. Expression profiling studies have demonstrated that classic ILCs preferentially display a luminal phenotype, whereas PLCs may be of luminal, HER2 or molecular apocrine subtypes. The aims of this study were two-fold: to determine the transcriptomic characteristics of lobular carcinomas and to define the genome-wide transcriptomic differences between classic ILCs and PLCs. To define the transcriptomic characteristics of ILCs, minimizing the impact of histological grade and molecular subtype on the analysis, we subjected a series of grade- and molecular subtype-matched ILCs and invasive ductal carcinomas (IDCs) to genome-wide gene expression profiling using oligonucleotide microarrays. Hierarchical clustering analysis demonstrated that ILCs formed a separate cluster and a supervised analysis revealed that 5.8% of the transcriptionally regulated genes were significantly differentially expressed in ILCs compared to grade- and molecular subtype-matched IDCs. ILCs displayed down-regulation of E-cadherin and of genes related to actin cytoskeleton remodelling, protein ubiquitin, DNA repair, cell adhesion, TGF-beta signalling; and up-regulation of transcription factors/immediate early genes, lipid/prostaglandin biosynthesis genes, and cell migration-associated genes. Supervised analysis of classic ILCs and PLCs demonstrated that less than 0.1% of genes were significantly differentially expressed between these tumour subtypes. Our results demonstrate that ILCs differ from grade- and molecular subtype-matched IDCs in the expression of genes related to cell adhesion, cell-to-cell signalling, and actin cytoskeleton signalling. However, classic ILCs and PLCs are remarkably similar at the molecular level and should be considered as part of a spectrum of lesions. Copyright © 2009 Pathological Society of Great Britain and Ireland. Published by John Wiley & Sons, Ltd.

Introduction

Invasive lobular carcinoma (ILC) is the most frequent special type of breast cancer, accounting for 5–15% of invasive breast cancers 1, and its incidence has been increasing over the last 20 years, mainly in women over 50 years of age. Invasive lobular cancers are characterized by rather discohesive cells individually dispersed or arranged in a single-cell-file pattern immersed in a fibrous stroma. These cells are characteristically round-to-ovoid and small, with eccentric and mildly to moderately atypical nuclei 1, 2. The vast majority of ILCs are of histological grade I/II, express oestrogen receptor, and lack HER2 expression/gene amplification, and fall into the ‘luminal’ molecular subgroup 3–5. In addition to the classic type of ILC, a pleomorphic variant characterized by cells with apocrine features and atypical and pleomorphic nuclei has been described. Pleomorphic lobular carcinomas (PLCs) occasionally lack hormone receptors and harbour amplification of oncogenes, including HER2. Microarray-based expression profiling analysis of pleomorphic lobular cancers demonstrates that these tumours may display luminal, molecular apocrine or HER2 transcriptomic profiles 6–12. The pleomorphic variant has also been suggested to have a more aggressive clinical behaviour 6.

Although the outcome of patients with ILCs does not appear to be significantly different from that of patients with invasive ductal carcinomas of no special type (IDCs-NST) 2, 4, 5, there is evidence to suggest that lobular carcinomas respond less often to neo-adjuvant and adjuvant chemotherapy 13–17. Furthermore, ILCs show a proclivity for metastatic dissemination to peculiar anatomical sites, such as the gastro-intestinal tract, uterus, meninges, ovary, and serosal cavities 1, 2, 4, 12.

There are several lines of evidence demonstrating that ILCs consistently lack the expression of E-cadherin, an adhesion molecule associated with cell survival and regarded by some as a tumour suppressor protein 18–20. Inactivation of E-cadherin through gene mutation, loss of heterozygosity, and/or gene promoter methylation has been reported in more than 95% of ILCs but is known to be altered in only a minority of IDCs-NST 2, 12, 18–20, in particular those of basal-like and triple-negative phenotype 21. Furthermore, other molecules downstream of E-cadherin, including β-catenin and p120 catenin, are known to be dysfunctional in ILCs 22–25.

Previous genome-wide, class-comparison micro- array-based expression profiling studies have determined the transcriptomic differences between ILCs and IDCs-NST 26–29. However, the results of these studies are confounded by the fact that ILCs more frequently express oestrogen receptor and are more often of histological grade I/II and of luminal molecular subtype than IDCs-NST 3. Given that oestrogen receptor expression, histological grade, and molecular subtype are associated with distinct genome-wide transcriptomic patterns 3, 30–33, the actual impact of lobular morphology on the expression profiles of breast cancer remains unclear.

The aims of this study were to determine the transcriptomic characteristics of ILCs as well as the basis of the classic and pleomorphic lobular histological growth pattern in breast cancer. To address this question, we analysed a series of 16 consecutive ILCs and 16 grade- and molecular subtype-matched IDCs-NST using genome-wide oligonucleotide microarrays. The results were subsequently validated in an independent cohort of 21 ILCs and 42 grade- and molecular subtype-matched IDCs-NST. As a second aim, we sought to define the genome-wide transcriptomic differences between classic and pleomorphic lobular carcinomas.

Materials and methods

Samples

Invasive lobular carcinomas (ILCs)

Samples of ILCs were selected from the frozen tissue bank of The Netherlands Cancer Institute/Antoni van Leeuwenhoek hospital (NKI/AVL). Before and after cutting tissue sections for RNA extraction, a representative section was stained with haematoxylin and eosin and semi-quantitatively assessed for the percentage of tumour areas over the total sample area. Only samples containing 50% tumour cells were selected for downstream analysis. Cases were reviewed according to the World Health Organisation criteria 1 and pleomorphic lobular carcinomas were classified according to the criteria of Eusebi et al6 and Weidner and Semple 7. The selection was carried out by independent review of the tumour sections by two pathologists (FCG and JRF) independently and only cases that fulfilled the diagnostic criteria for pure ILCs according to both observers were included (n = 11 classic variant; n = 9 pleomorphic variant). Previous analysis revealed that all cases included in this study were E-cadherin-negative (data not shown) 3. The transcriptomic profiles of the cases reported here are part of a previously published study 3 and are available at ArrayExpress (E-NCMF-3). Approval for the analysis of the samples included in the present study was granted by the NKI/AVL Research Ethics Committee.

Control group (grade- and molecular subtype-matched IDCs-NST)

Gene expression data from 102 invasive breast carcinomas were retrieved that were part of an unrelated research project in our institute published by Kreike et al34, 35. This gene-expression data set differs only in tumour type analysis (ie predominantly IDCs-NST), but is similar with regard to experimental work-up. Of these 102 invasive breast carcinomas, 91 were IDCs-NST and were used as controls. Gene expression data are publicly available at ArrayExpress (E-NCMF-24). Detailed information on RNA extraction, amplification, labelling, hybridization, scanning, microarray platform, and analysis has been described previously 3, 34, 35.

Data analysis

A subset of the 34 580 probes was selected based on the following criteria: unambiguous mapping information for the probe; expression data available for at least 75% of all experiments; and expression level significantly different from the reference expression in at least 10% of experiments with a p value of less than 0.01 36. These criteria reduced the total transcript number to 7095 significantly regulated transcripts.

To define whether a tumour was of basal-like, luminal A, luminal B, HER2 or normal breast-like molecular subtype, we determined the Spearman's rank correlation of each case with the ‘Intrinsic/UNC’ class centroids by Hu et al37 and included only samples with a correlation of more than 0.1 to any centroid as previously described 3, 34, 36. Out of 306 unique ‘intrinsic genes’, 293 were identified in our data set.

For unsupervised clustering analysis, we performed complete- and average-linkage hierarchical clustering of a centred Pearson correlation similarity matrix with 7095 filtered genes using the Cluster software 38, and results were visualized with TreeView. Genes and arrays were median-centred as previously described 36.

To determine the genes significantly differentially expressed between ILCs and IDCs-NST, we used significance analysis of microarrays (SAM) software 39, performing 1000 permutations, as previously described 36.

Pathway analysis

Significantly regulated pathways and networks in the gene expression data were determined using (i) the Ingenuity Pathway Analysis (IPA) program (http://www.ingenuity.com) and (ii) the DAVID Gene Functional Classification Tool (http://david.abcc.ncifcrf.gov). For IPA analysis, the fold difference of differentially expressed transcripts identified by SAM were mapped to networks available in the Ingenuity database and ranked by score. The score indicates the likelihood of the genes in a network being found together due to chance. Using a 99% confidence level, scores of ≥ 3 are significant 3, 36, 40. In addition, Ensemble Gene IDs of the differentially expressed transcripts identified by SAM were uploaded to the ‘Functional Annotation Tool’ of the DAVID database, mapped to the KEGG-pathways, ranked by p value (EASE Score, modified Fisher's exact test) and corrected for multiple testing according to Benjamini–Hochberg.

Validation

To validate the findings from the microarray gene expression analysis, we retrieved consecutive classic ILCs, PLCs, and IDCs-NST from an independent series of invasive breast carcinomas from patients treated with therapeutic surgery followed by anthracycline therapy. This study was approved by the Royal Marsden Hospital Research Ethics Committee. The details of this cohort have been previously described 21, 41. The histological type was reviewed by two of the authors (JRF and FCG). The molecular subtype of each case was defined according to an immunohistochemical surrogate 42. Thirty consecutive ILCs (one grade I luminal, 14 grade II luminal, 12 grade III luminal, one grade II HER2, one grade III HER2, and one grade III basal-like) and 60 grade- and molecular subtype-matched IDCs-NST (two grade I luminal, 28 grade II luminal, 24 grade III luminal, two grade II HER2, two grade III HER2, and two grade III basal-like) were included (Supporting information, Supplementary Table 1).

Table 1. Histological grade (Bloom and Richardson) and molecular subtypes according to the ‘Intrinsic/UNC’ centroids (Hu et al37) of 16 invasive lobular carcinomas and 16 grade- and molecular subtype-matched invasive ductal carcinomas of no special type analysed by gene expression profiling, and of four excluded invasive lobular carcinomas
IDHistological typeHistological gradeVariantMolecular subtypeIDHistological typeHistological gradeMolecular subtype
3639ILCIIIPleomorphicBasal-like49IDC-NSTIIIBasal-like
3634ILCIIClassicHER2201IDC-NSTIIHER2
3620ILCIClassicLuminal A42IDC-NSTILuminal A
3627ILCIClassicLuminal A65IDC-NSTILuminal A
3618ILCIIClassicLuminal A43IDC-NSTIILuminal A
3622ILCIIClassicLuminal A54IDC-NSTIILuminal A
3625ILCIIClassicLuminal A68IDC-NSTIILuminal A
3628ILCIIClassicLuminal A69IDC-NSTIILuminal A
3629ILCIIClassicLuminal A87IDC-NSTIILuminal A
3637ILCIIClassicLuminal A104IDC-NSTIILuminal A
3638ILCIIPleomorphicLuminal A136IDC-NSTIILuminal A
3619ILCIIIPleomorphicLuminal A41IDC-NSTIIILuminal A
3624ILCIIIPleomorphicLuminal A52IDC-NSTIIILuminal A
3635ILCIIIPleomorphicLuminal A56IDC-NSTIIILuminal A
3640ILCIIIPleomorphicLuminal A63IDC-NSTIIILuminal A
3636ILCIIIPleomorphicNormal breast-like171IDC-NSTIIINormal breast-like
IDHistological typeHistological gradeVariantMolecular subtypeExclusion criteria
  1. IDC-NST = invasive ductal carcinoma of no special type; ILC = invasive lobular carcinoma; NC = non-classifiable.

3630ILCIIIPleomorphicNCMolecular subtype unclassifiable (correlation < 0.1)
3626ILCIIClassicNormal breast-likeNo grade- and molecular subtype-matched IDC-NST
3631ILCIIPleomorphicLuminal BNo grade- and molecular subtype-matched IDC-NST
3633ILCIIClassicLuminal BNo grade- and molecular subtype-matched IDC-NST

Quantitative real-time reverse transcriptase-PCR (qRT-PCR)

Total RNA was extracted from five whole-tissue 8 µm thick sections from the 30 ILCs and 60 IDCs-NST using the RNeasy FFPE RNA Isolation Kit (Qiagen) followed by an additional DNase treatment as previously described 41. RNA quantification was performed using the Ribogreen Quant-iT reagent (Invitrogen, UK) and reverse transcription was carried out as previously described 41 using Superscript III (Invitrogen, UK) and using 400 ng of RNA per reaction, with triplicate reactions performed for each sample. Out of these cases, 21 ILCs (12 grade II luminal, eight grade III luminal, and one grade III HER2; ten classic and 11 pleomorphic variants) and 42 grade- and molecular subtype-matched IDCs-NST (24 grade II luminal, 16 grade III luminal, and two grade III HER2) produced RNA of sufficient yield and quality to be subjected to quantitative real-time PCR analysis. Quantitative real-time PCR was performed for AFF1, ADAM9, ANKRD28, and HIF1A (Assay on demand IDs: AFF1_Hs00610559_m1, ADAM9-Hs00177638_m1, ANKRD28-Hs00389943_m1, HIF1A-Hs00936368_m1; Applied Biosystems) using TaqMan® chemistry on the ABI Prism 7900HT (Applied Biosystems), using the standard curve method 41. In addition, two reference genes (TFRC and MRPL19; Hs00174609_m1-TFRC, Hs00608522_ g1-MRPL19) were used, having been previously selected as effectively normalizing for degradation of RNA 41. Target gene expression levels were normalized to the geometric mean of the two reference genes and normalized to a calibrator (pool of FFPE tumour cDNAs from the same series). The parametricity of the data was tested using the Kolmogorov–Smirnov test and only HIF1A expression values were normally distributed. Therefore, comparison between the expression levels of ANKRD28, AFF1, and ADAM9 was carried out using the non-parametric Mann–Whitney U-test, whereas for HIF1A a two-tailed, homoscedastic t-test was employed.

Immunohistochemical staining of caveolin-1

Formalin-fixed, paraffin-embedded whole tissue sections were cut at 3 µm and mounted on silane-coated slides. Immunohistochemical analysis of caveolin-1 (CAV1) was performed as previously described 43 with the mouse monoclonal antibody 2297 (BD Transduction Laboratories, USA) at 1 : 150 dilution following heat-induced antigen retrieval [18 min, microwave oven, DAKO antigen retrieval solution (pH 6.0)]. Positive and negative controls (omission of the primary antibody and IgG-matched serum) were included for each immunohistochemical run. Moreover, normal endothelial cells present in all tissue sections were used as internal positive controls. The proportion and intensity of CAV1 expression in tumour and stromal cells were evaluated semi-quantitatively. The proportion was semi-quantitatively recorded as 0, no expression; 1, up to 1% of tumour cells; 2, > 1% to 10% of tumour cells; 3, > 10% to 33% of tumour cells; 4, > 33% to 66% of tumour cells; and 5, > 66% of tumour cells. Intensity of staining was recorded as 0, negative; 1, weak; 2, moderate (slightly weaker than that seen in endothelial cells); and 3, strong (of similar intensity to that seen in endothelial cells). Only membranous staining, with or without cytoplasmic staining, was considered specific 43, 44.

Results

ILCs are molecularly distinct from histological grade- and molecular subtype-matched IDCs-NST

We selected 20 invasive ILCs according to the criteria of the WHO 1, 11 of which were classic and nine were pleomorphic variants (Figure 1 and Table 1) 6, 7. To define the basal-like, luminal A, luminal B, HER2, and normal breast-like molecular subtype class, we determined the correlation between the expression profile of each tumour with the ‘Intrinsic/UNC’ class centroids described by Hu et al37 as previously described 3, 36, 40. Of the 20 ILCs included in this study, 13 displayed a luminal A, two a luminal B, one a basal-like, one a HER2, and two a normal breast-like expression profile. One ILC could not be assigned to a molecular subtype class (ie correlation with any class centroid < 0.1) and was excluded from further analysis (Table 1).

Figure 1.

Representative photomicrographs of (A) classic invasive lobular and (B) pleomorphic invasive lobular carcinoma (original magnification 200×)

To assess whether the histological differences bet- ween ILCs and IDCs-NST are apparent at the transcriptomic level, we selected from a control group of 102 invasive breast carcinomas 16 IDCs-NST that were histological grade- and molecular subtype-matched with 16 of the 19 included ILCs (Table 1); no IDCs-NST of the same histological grade and molecular subtype were available for two ILCs of luminal B and one of normal breast-like phenotype, which were excluded from further analysis. The final data set comprised 16 ILCs (two grade I, eight grade II, and six grade III; 13 luminal A, one HER2, one normal breast-like, and one basal-like) and 16 grade- and molecular subtype-matched IDCs-NST (Table 1).

Unsupervised hierarchical clustering using 7095 significantly regulated transcripts revealed, regardless of the clustering algorithm employed (ie complete- versus average-linkage), two main clusters, one of which was significantly enriched for IDCs-NST (Fisher's exact test p = 0.000068, Figure 2; Fisher's exact test p = 0.0113, Supporting information, Supplementary Figure 1), whereas ILCs and the remaining grade- and molecular subtype-matched IDCs-NST were admixed in the other cluster. The clustering appeared to be driven by the histological type of the tumours rather than their histological grade or molecular subtype class (Figure 2 and Supporting information, Supplementary Figure 1). These results provide evidence to suggest that ILCs, as a group, are distinct from histological grade- and molecular subtype-matched IDCs-NST at the transcriptomic level.

Figure 2.

Unsupervised hierarchical complete-linkage clustering of 16 invasive lobular carcinomas (ILC) and 16 histological grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDC-NST), using 7095 significantly regulated transcripts

Given the separation observed between ILCs and histological grade- and molecular subtype-matched IDCs-NST in the hierarchical clustering, we sought to define their transcriptomic differences. SAM analysis using the 7095 significantly regulated transcripts revealed 414 transcripts (5.84%) differentially expressed between ILCs and grade- and molecular subtype-matched IDCs-NST [61 up- and 353 down-regulated transcripts in ILCs; false discovery rate (FDR) < 1%]. As expected 26–29, and in support of our sample selection, CDH1 was identified as the top down-regulated gene in ILCs compared with grade- and molecular subtype-matched IDCs-NST (Supporting information, Supplementary Table 2). Of note, genes involved in actin cytoskeleton remodelling (eg ROCK1, ROCK2), in ubiquitin conjugation (eg USP16, USP24, SUMO1), DNA repair (eg RAD21, RAD50, MSH2), cell adhesion (eg ITGB1, LAMA3, LAMC1), TGF-beta signalling (eg SMAD4, EP300, RPS6KB1), and CCND1 were down-regulated, whereas transcription factors and immediate early genes (eg HOXB6, ELF4, EGR1, IER2), lipid synthesis/prostaglandin biosynthesis genes (eg PTGDS, PTGIS) but also WNT6, the cell migration associated gene ANKRD28 as well as AFF1 were up-regulated in ILCs compared with grade- and molecular subtype-matched IDCs-NST.

Table 2. Quantitative real-time RT-PCR analysis of the expression of ANKRD28 and AFF1 in 21 invasive lobular carcinomas and 42 grade- and molecular subtype-matched invasive ductal carcinomas of no special type
 TypeNMeanStandard error meanp value*
  • *

    Mann–Whitney U-test.

  • Kolmogorov-Smirnov test for normality demonstrated non-parametric distribution for AFF1 and ANKRD28.

ANKRD28 0.0398
 IDC421.88610.1534 
 ILC212.43170.2340 
AFF1 0.012659
 IDC421.68330.0910 
 ILC212.15340.1608 

These differences were further explored by Ingenuity Pathway Analysis (IPA) of the 414 differentially regulated transcripts identified by SAM, which revealed that among the ten most significant networks four played a role in ‘Cellular Assembly and Organization’ and ‘Cell-To-Cell Signalling and Interaction’ (scores 36, 34, 30, and 21, respectively) and were predominantly down-regulated in ILCs compared with grade- and molecular subtype-matched IDCs-NST (Figure 3A and Supporting information, Supplementary Table 3). In addition, down-regulation of genes related to a ‘DNA Replication, Recombination, and Repair, Cell Cycle, Lipid Metabolism’ network (score 36) (eg RAD21, NBN, PRKDC, SPAST) was observed in ILCs versus IDCs-NST of the same histological grade and molecular phenotype (Figure 3B and Supporting information, Supplementary Table 3).

Figure 3.

Ingenuity Pathway Analysis. (A) Genes of the ‘Cellular Assembly and Organization, Cell-to-Cell Signalling and Interaction, Cellular Function and Maintenance’ network (score 34) and (B) ‘DNA Replication, Recombination, and Repair, Cell Cycle, Lipid Metabolism’ network (score 36) are down-regulated in ILCs versus grade- and molecular subtype-matched IDCs-NST. (C) Differentially expressed transcripts between ILCs and grade- and molecular subtype-matched IDCs-NST are significantly enriched for genes of the canonical ‘Regulation of Actin-based Motility by Rho’ (p = 0.00225) and (D) ‘Actin Cytoskeleton Signalling’ (p = 0.00666) pathways. Green: down-regulation; red: up-regulation. IDC-NST = invasive ductal carcinomas of no special type; ILC = invasive lobular carcinoma

Table 3. Quantitative real-time RT-PCR analysis of the expression of HIF1A and ADAM9 in ten classic and 11 pleomorphic invasive lobular carcinomas
 SubtypeNMeanStandard error meanp value
  • *

    Kolmogorov-Smirnov test for normality demonstrated normal distribution for HIF1A and non-parametric distribution for ADAM9.

  • Two-tailed, homoscedastic t-test.

  • Mann–Whitney U-test.

HIF1A* 0.04954
 Classic101.028690.159363 
 Pleomorphic111.7903580.313856 
ADAM9* 0.7045
 Classic101.5144770.217398 
 Pleomorphic111.9286120.448318 

Pathway analysis demonstrated that the 414 differentially expressed transcripts between ILCs and grade- and molecular subtype-matched IDCs-NST were significantly enriched for genes of the canonical ‘Protein Ubiquitination’ (p = 0.00285), ‘p53 Signalling’ (p = 0.00388), and ‘Role of BRCA1 in DNA Damage Response’ (p = 0.00568), as well as ‘Regulation of Actin-based Motility by Rho’ (p = 0.00225) and ‘Actin Cytoskeleton Signalling’ (p = 0.00666) pathways (Figures 3C and 3D and Supporting information, Supplementary Table 3). In support of this finding, pathway analysis using DAVID identified the ‘Focal adhesion’ KEGG-pathway as being significantly enriched (p = 0.00011; Benjamini–Hochberg 0.0221; Supporting information, Supplementary Table 4), as well as the ‘TGF-beta signalling pathway’ and ‘Regulation of actin cytoskeleton’ KEGG-pathways (p = 0.00178 and p = 0.00866, respectively), which were not statistically significant after Benjamini–Hochberg correction (Supporting information, Supplementary Table 4).

Taken together, our results provide evidence to suggest that ILCs are transcriptionally distinct from histological grade- and molecular subtype-matched IDCs-NST. Furthermore, discrete molecular pathways/networks were found to be activated in ILCs compared with grade- and molecular subtype-matched IDCs-NST. In accordance with the distinctive invasion pattern and discohesiveness of cells from ILCs 1, 2, these cancers displayed down-regulation of actin cytoskeleton-, cell adhesion, and cell-to-cell signalling genes compared with grade- and molecular subtype-matched IDCs-NST.

Low-grade ILCs are molecularly distinct from molecular subtype-matched low-grade IDCs-NST

Given that invasive breast cancers of low histological grade have been reported to have similar clinical presentations and immunohistochemical profiles 45, 46, and that histological grade is associated with distinct genome-wide transcriptomic patterns 30–33, we sought to assess the differences at the molecular level between low-grade ILCs and grade- and molecular subtype-matched IDCs-NST separately.

From the initial series of 16 grade- and molecular subtype-matched ILCs and IDCs-NST, we retrieved ten grade I/II ILCs and ten grade- and molecular subtype-matched IDCs-NST (Table 1). Unsupervised hierarchical clustering analysis of ten ILCs of histological grades I and II and ten grade- and molecular subtype-matched IDCs-NST revealed, similar to the clustering of the whole cohort, independent of the cluster algorithm used, two main clusters, of which one was significantly enriched for IDCs-NST (Fisher's exact test p = 0.00071; Supporting information, Supplementary Figure 2). These results provide evidence to suggest that also low-grade ILCs and IDCs of the same molecular subtype are distinct at the molecular level.

To determine the differences between grade I and II ILCs and IDCs-NST of the same molecular subtype, we performed SAM using the 7095 significantly regulated transcripts. This analysis revealed 270 transcripts (3.81%) differentially expressed between grade I and II ILCs and molecular subtype-matched IDCs-NST of the same grade (77 up- and 193 transcripts down-regulated in ILCs; FDR < 1.05%; Supporting information, Supplementary Table 5). Similar to the analysis of the whole cohort, CDH1 was identified as the top down-regulated gene in ILCs of grades I and II compared with grade- and molecular subtype-matched IDCs-NST, as were genes involved in ubiquitin conjugation (eg USP1, USP24, DUB, HECTD1), actin cytoskeleton remodelling (eg ROCK1, ROCK2), cell adhesion (eg ITGB1, LAMA3, ADAM10, ADAM9), and cell cycle (eg STAG2, RB1), which were part of the networks ‘Cancer, Cellular Growth and Proliferation, Endocrine Systems Disorder’ (score 37) and ‘Cell-to-Cell Signalling and Interaction, Cancer, Reproductive System Disease’ (score 36) (Supporting information, Supplementary Figures 3A and 3B and Supplementary Table 6). In contrast, genes involved in lipid metabolism (eg LPL, LEP, LIPE, FABP4, PTGDS, PTGIS), as seen in the network ‘Lipid Metabolism, Molecular Transport, Small Molecule Biochemistry’ (score 37; Supporting information, Supplementary Figure 3C and Supplementary Table 6), as well as ESR2 and AQP7, but also transcription factors (eg HOXB6, ELF4) and WNT6, ANKRD28 and AFF1, were up-regulated in ILCs compared with grade- and molecular subtype-matched IDCs-NST. These findings suggest that although low-grade ILCs and IDCs-NST share similar clinical characteristics and immunohistochemical profiles and may form a ‘low-grade breast neoplasia family’ 45, 46, these entities are not necessarily identical at the molecular level. In fact, our results demonstrate that grade I and II ILCs have activation of distinct molecular networks and may have a more overt luminal phenotype compared with molecular subtype-matched IDCs-NST of grades I and II.

Transcriptomic differences between high-grade ILCs and molecular subtype-matched high-grade IDCs-NST

For completeness, we also compared grade III ILCs and molecular subtype-matched grade III IDCs-NST at the molecular level. It should be noted, however, that the groups included in this exploratory analysis were rather small (ie only six ILCs were of grade III, Table 1).

Unsupervised hierarchical clustering analysis of six ILCs of histological grade III and six grade- and molecular subtype-matched IDCs-NST revealed, independent of the cluster algorithm used, two main clusters. One cluster was composed of only tumours of luminal molecular subtype, in which the ILCs clustered together, whereas the other cluster was composed of basal-like, luminal, and normal breast-like tumours, in which the ILCs clustered together (Supporting information, Supplementary Figure 4).

Figure 4.

Validation of gene expression results. Quantitative real-time RT-PCR assessment of (A) ANKRD28 and (B) AFF1 in 21 ILCs and 42 grade- and molecular subtype-matched IDCs-NST (Mann–Whitney U-test p = 0.0398 and p = 0.0127, respectively), and (C) HIF1A and (D) ADAM9 in ten classic ILCs and 11 pleomorphic ILCs (two-tailed, homoscedastic t-test p = 0.0495 and Mann–Whitney U-test p = 0.7045, respectively). Error bars represent standard deviation of mean

SAM analysis using the 7095 significantly regulated transcripts revealed only four transcripts to be significantly differentially expressed between grade III ILCs and molecular subtype-matched grade III IDCs-NST, which were all down-regulated in ILCs, with CDH1 as the top gene (FDR = 0%; Supporting information, Supplementary Table 7). Two genes, V-type proton ATPase subunit B (ATP6V1B1) and leucine-rich repeat protein (SHOC2), were also found in the 353 differentially down-regulated transcripts identified between all ILCs and grade- and molecular subtype-matched IDCs-NST (Supporting information, Supplementary Table 2). The only gene specifically down-regulated in grade III ILCs compared with grade- and molecular subtype-matched IDCs-NST was TPRG1 (encoding tumour protein p63-regulated gene 1 protein), which maps to 3q28 and is poorly characterized. Given the small sample size, these results should be interpreted with caution.

Classic and pleomorphic lobular carcinomas harbour remarkably similar transcriptomic profiles

Pleomorphic lobular carcinoma (PLC) of the breast is reported to have a more aggressive clinical behaviour than classic ILC 2, 6, 7, 12, 15. However, there is evidence to suggest that PLCs and classic ILCs evolve through the same molecular genetic pathways 8, 9, 12. Here, we sought to determine the differences between PLCs and classic ILCs at the transcriptional level. Of the 16 ILCs included in this study, seven were pleomorphic and nine were classic variants (Table 1). Supervised analysis using SAM using 7095 significantly regulated transcripts revealed that classic ILCs and PLCs were highly similar at the transcriptomic level, given that only seven transcripts were identified to be significantly differentially expressed between PLCs and classic ILCs (Supporting information, Supplementary Table 8) (three transcripts preferentially up- and four down-regulated in classic ILCs compared with PLCs; FDR = 0%). HIF1A, which plays a central role in response to hypoxia and angiogenesis 47, was found to be down-regulated in classic ILCs compared with PLCs, as was ADAM9, a disintegrin suggested to mediate cell adhesion and cell–matrix interactions, and to cleave and release a number of molecules with important roles in tumourigenesis and angiogenesis 48; and S100A8, which has been reported to be essential for H-Ras-mediated cell invasion and migration and, when overexpressed, to be a marker of poor prognosis in IDCs-NST 49.

Validation of microarray-based expression profiling findings

To validate the findings from the microarray-based gene expression analysis, the expression levels of two differentially expressed genes between ILCs and IDCs-NST (ANKRD28 and AFF1) and two differentially expressed genes between classic ILCs and PLCs (HIF1A and ADAM9) were verified by qRT-PCR in an independent, consecutive series of 21 ILCs and 42 grade- and molecular subtype-matched IDCs-NST. ANKRD28 and AFF1 were expressed at significantly higher levels in ILCs when compared with grade- and molecular subtype-matched IDCs-NST (Mann–Whitney U-test, p < 0.05; Figure 4 and Table 2). Our results also confirmed significantly higher expression levels of HIF1A in PLCs than in ILCs (two-tailed t-test, p < 0.05) (Table 3).

Previous studies 27, 29 have reported that CAV1, which is preferentially expressed in basal-like cancers, was overexpressed in ILCs when compared with IDCs-NST. In our series, we were unable to detect a significantly different expression of CAV1 between the two groups by microarray analysis (Supporting information, Supplementary Table 2). Given that CAV1 has been demonstrated to be expressed at high levels in endothelial cells, fibroblasts, adipocytes, and normal breast myoepithelial cells 43, 44, we hypothesized that the differences observed in previous studies would stem from a higher percentage of non-neoplastic cells in ILC samples when compared with samples of IDCs. To determine whether CAV1 would be more frequently expressed in the neoplastic cells of ILCs than in those of IDCs-NST, we performed an immunohistochemical analysis of CAV1 expression in 30 ILCs and 60 grade- and molecular subtype-matched IDCs-NSTs. Given the association between CAV1 expression and basal-like phenotype 43, 44, 50 and that all but three tumours were of luminal or HER2 phenotype, CAV1 was not expressed in neoplastic cells of any case, whereas stromal cells displayed variable expression. In addition, no association between the expression levels of CAV1 in the stroma of ILCs and grade- and molecular subtype-matched IDCs-NST was observed (Supporting information, Supplementary Table 1 and Figure 5). Interestingly, the distribution of CAV1 in the stroma was variable within each case, with desmoplastic areas displaying stronger staining than areas of paucicellular stroma (Figure 5C). Furthermore, fibroblasts encasing single tumour cells and groups of neoplastic cells consistently expressed CAV1 in ILCs and grade- and molecular subtype-matched IDCs-NST (Figures 5B and 5D).

Figure 5.

Immunohistochemical expression of caveolin-1. (A, B) Invasive lobular carcinoma, original magnification 40× and 400×, respectively; (C, D) invasive ductal carcinoma of no special type, original magnification 40× and 400×, respectively

Discussion

Previous studies have demonstrated that IDCs-NST and ILCs can be subclassified into the luminal A, luminal B, basal-like, normal breast-like, and HER2 molecular subtypes 3, 30, 37, 51. It should be noted, however, that lobular carcinomas have been shown to preferentially segregate with the luminal subgroups 3.

The transcriptomic differences between ILCs and IDCs-NST have been previously investigated 26–29; however, ILCs were compared with IDCs-NST without the differences in histological grade, oestrogen receptor status, and molecular subtype being taken into account. Our study provides a direct assessment of the molecular underpinnings of the lobular histological growth pattern, as we have grade- and molecular subtype-matched these tumours with IDCs-NST. Not surprisingly, the overlap between the list of genes identified in our study and those from other studies ranged from one to four genes (Supporting information, Supplementary Table 9). In fact, the only consistent finding in all of the previous publications is the CDH1 gene, which has repeatedly been shown to be markedly down-regulated in ILCs when compared with IDCs-NST 2, 12, 18, 19, 24–29.

Previous studies have suggested that CAV1 would be expressed at higher levels in ILCs when compared with IDCs-NST 27, 29. Our group 43, 44 and others 50, 52 have demonstrated that CAV1 is preferentially expressed in IDCs-NST of basal-like phenotype and metaplastic breast carcinomas and that ILCs rarely express this caveolar protein. In this study, we have demonstrated that there is no difference in the expression levels of CAV1 in the neoplastic cells of ILCs and grade- and molecular subtype-matched IDCs-NST. As CAV1 is also strongly expressed in adipose, endothelial, and normal breast stromal cells and myoepithelial cells 43, 44, 50, 52, and ILCs more frequently infiltrate in between normal breast structures, the differences in CAV1 expression between ILCs and IDCs found in the previous studies may stem from the more abundant presence of (entrapped) normal breast tissue within the bulk of ILC samples. It may be that the reported higher levels of CAV1 expression in ILCs than in IDCs-NST derive from a higher prevalence of CAV1-positive stromal cells in lobular cancers than in IDCs-NST. The data presented here and in another study 53 failed to show an association between expression of CAV1 in stromal cells and histological type. Instead, a strong association between lack of CAV1 expression and higher histological grade has been reported 53. Therefore an alternative explanation for the higher levels of CAV1 in lobular cancers reported in previous studies may be a higher prevalence of grade I/II cancers in the ILC group than in the IDC-NST group.

On the other hand, our results demonstrate that the classic and pleomorphic variants of ILCs have dysfunctional cell-to-cell signalling, actin cytoskeleton remodelling/signalling, and cell adhesion networks and pathways. This is not surprising, given that the main histological difference between grade- and molecular subtype-matched ILCs and IDCs-NST is the invasion pattern and the characteristic discohesiveness of ILC cells. Although we have validated the higher expression levels of ANKDR28 and AFF1 in classic and pleomorphic ILCs than in grade- and molecular subtype-matched IDCs-NST, it should be noted that these differences, albeit significant, are small and that these genes are unlikely to be useful as ancillary markers to differentiate between ILCs and IDCs. Furthermore, this study did not address the transcriptomic features of the alveolar or solid variants of ILCs 1, which despite having distinctive architectural patterns, are comprised of cells harbouring the typical lobular cancer cytological features and are also discohesive 1. Further studies to define the molecular underpinning of the alveolar and solid variants of ILCs are warranted.

Our data also demonstrate that classic ILCs and PLCs are remarkably similar at the transcriptomic level as only less than 0.1% of the significantly regulated transcripts were found to be differentially expressed. These findings support our previous work which suggests that PLCs and classic ILCs harbour similar patterns of genetic aberrations and may evolve along a common molecular genetic pathway 8, 9, 12. The up-regulation of ADAM9 and S100A8 in PLCs, genes mediating cell adhesion, invasion, and migration 48, 49, may provide an explanation for the more aggressive clinical behaviour described for PLCs compared with classic ILCs. High-grade breast cancers more often display expression of hypoxia markers (eg CAIX, HIF1A) and harbour necrotic areas than low-grade breast cancers 54, 55; hence, one could hypothesize that the higher levels of HIF1A identified here in PLCs than in classic ILCs may stem from a higher hypoxic/necrotic environment in PLCs.

Finally, we and others have suggested that low-grade IDCs and ILCs and their respective precursors would form a family of lesions (ie low-grade breast neoplasia family) 45, 46, 56. Although these lesions display similar clinical presentations and immunohistochemical profiles, differences in metastatic pattern and response to chemotherapy have been reported 13–17. Here we have demonstrated that low-grade ILCs have distinct expression profiles when compared with grade- and molecular subtype-matched IDCs. Therefore although these lesions may form a spectrum or family, there are important differences in the transcriptomic profiles, metastatic pattern, and clinical behaviour that warrant their separation as discrete entities.

SUPPORTING INFORMATION ON THE INTERNET

The following supporting information may be found in the online version of this article.

Table S1. Histological grade (Bloom and Richardson) and molecular subtypes according to an immunohistochemical surrogate (Nielsen et al42) of the validation cohort of 30 ILCs and 60 grade- and molecular subtype-matched IDCs-NST, and scoring results of caveolin-1 immunohistochemical staining.

Table S2. List of 414 differentially expressed transcripts between invasive lobular carcinomas (ILCs) and histological grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDCs-NST) identified by SAM.

Table S3. Top networks and canonical pathways identified by Ingenuity Pathway Analysis in genes differentially expressed between invasive lobular carcinomas (ILCs) and grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDCs-NST).

Table S4. Top ten KEGG-pathways identified by the ‘Functional Annotation Tool’ of the DAVID database in genes differentially expressed between invasive lobular carcinomas (ILCs) and grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDCs-NST).

Table S5. List of 270 differentially expressed transcripts between grade I and II invasive lobular carcinomas (ILCs) and molecular subtype-matched grade I and II IDCs-NST identified by SAM.

Table S6. Top networks identified by Ingenuity Pathway Analysis in genes differentially expressed between grade I and II invasive lobular carcinomas (ILCs) and histological grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDCs-NST).

Table S7. List of four differentially expressed transcripts between grade III invasive lobular carcinomas (ILCs) and molecular subtype-matched grade III invasive ductal carcinomas of no special type (IDCs-NST) identified by SAM.

Table S8. List of seven differentially expressed transcripts between classical invasive lobular carcinomas (ILCs) and pleomorphic ILCs identified by SAM.

Table S9. Overview of transcriptomic differences identified between ILCs and IDCs-NST in previous studies.

Figure S1. Unsupervised hierarchical average-linkage clustering of 16 invasive lobular carcinomas (ILC) and 16 histological grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDC-NST), using 7095 significantly regulated transcripts.

Figure S2. Unsupervised hierarchical average-linkage clustering of ten invasive lobular carcinomas (ILC) of histological grades I and II and ten grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDC-NST).

Figure S3. Ingenuity Pathway Analysis.

Figure S4. Unsupervised hierarchical average-linkage clustering of six invasive lobular carcinomas (ILC) of histological grade III and six grade- and molecular subtype-matched invasive ductal carcinomas of no special type (IDC-NST).

Acknowledgements

We are grateful to Professor Alan Ashworth for his insightful comments. This study was funded in part by Breakthrough Breast Cancer. Britta Weigelt is supported by a Cancer Research UK Fellowship. We also acknowledge NHS funding to the NIHR Biomedical Research Centre.

Ancillary