Mixed ductal‐lobular carcinomas: evidence for progression from ductal to lobular morphology

Abstract Mixed ductal–lobular carcinomas (MDLs) show both ductal and lobular morphology, and constitute an archetypal example of intratumoural morphological heterogeneity. The mechanisms underlying the coexistence of these different morphological entities are poorly understood, although theories include that these components either represent ‘collision’ of independent tumours or evolve from a common ancestor. We performed comprehensive clinicopathological analysis of a cohort of 82 MDLs, and found that: (1) MDLs more frequently coexist with ductal carcinoma in situ (DCIS) than with lobular carcinoma in situ (LCIS); (2) the E‐cadherin–catenin complex was normal in the ductal component in 77.6% of tumours; and (3) in the lobular component, E‐cadherin was almost always aberrantly located in the cytoplasm, in contrast to invasive lobular carcinoma (ILC), where E‐cadherin is typically absent. Comparative genomic hybridization and multiregion whole exome sequencing of four representative cases revealed that all morphologically distinct components within an individual case were clonally related. The mutations identified varied between cases; those associated with a common clonal ancestry included BRCA2, TBX3, and TP53, whereas those associated with clonal divergence included CDH1 and ESR1. Together, these data support a model in which separate morphological components of MDLs arise from a common ancestor, and lobular morphology can arise via a ductal pathway of tumour progression. In MDLs that present with LCIS and DCIS, the clonal divergence probably occurs early, and is frequently associated with complete loss of E‐cadherin expression, as in ILC, whereas, in the majority of MDLs, which present with DCIS but not LCIS, direct clonal divergence from the ductal to the lobular phenotype occurs late in tumour evolution, and is associated with aberrant expression of E‐cadherin. The mechanisms driving the phenotypic change may involve E‐cadherin–catenin complex deregulation, but are yet to be fully elucidated, as there is significant intertumoural heterogeneity, and each case may have a unique molecular mechanism. © 2018 The Authors. The Journal of Pathology published by John Wiley & Sons Ltd on behalf of Pathological Society of Great Britain and Ireland.


Introduction
The full extent of breast cancer heterogeneity is being recognized through genomic analysis, although its visible manifestations have been essential to the histopathological classification for many decades. Breast tumours are broadly categorized into two morphological groups: tumours showing specialized growth patterns, and those with no distinguishing features [invasive carcinoma of no special type (IC-NST), or invasive ductal carcinoma (IDC), which we will use for simplification of expression]. Invasive lobular carcinoma (ILC) is the most common special type (∼15%), and is defined by discohesive cells individually dispersed or arranged in single file, linearly in the fibrous stroma [1]. Three to five per cent of tumours show both ductal and lobular morphology [1], and are classified as mixed ductal-lobular carcinoma (MDL) if the ductal component constitutes at least 10% of the tumour and the lobular component constitutes ≥50% [1][2][3]. However, this definition is interpreted variably, as it does not account for tumours showing small foci of lobular differentiation. Some therefore report MDL as IDC with lobular features or simply as 'ductal or lobular carcinoma', whereas others also include those tumours with predominantly ductal features [3].
Several studies have investigated the clinical significance of MDLs, and have shown that they are associated with a better prognosis than IDC, but a poorer prognosis than ILC [3,4], although the distinction was mostly lost after adjustment for grade [3], and that the prognosis for MDLs was poorer after stratification for oestrogen receptor (ER) positivity [5]. Whereas a number of studies have shown that preinvasive lesions confined to the duct [lobular carcinoma in situ (LCIS) and ductal carcinoma in situ (DCIS)] and their associated invasive carcinomas are clonally related [6][7][8], there are insufficient studies to support this idea in MDLs. Limited evidence suggests that lesions with mixed morphology are clonally related and arise from a common ancestor [9]; however, this study used low-resolution technologies on relatively small cohorts. MDLs represent a unique model for interrogating intratumoural heterogeneity, clonal evolution, and the mechanisms driving acquisition of infiltrative growth patterns. In contrast to ILC and IDC, there are limited data on the underlying pathobiology of MDLs [2,3].
To further understand intratumoural heterogeneity and clonal origins of the tumour, we assembled a cohort of 82 MDLs and investigated the expression of the E-cadherin adhesion complex, which is lost in classic ILC, accounting for its characteristic morphology. We also performed detailed genomic analysis of representative MDLs, using chromosomal comparative genomic hybridization (cCGH) (n = 4) and whole exome sequencing (WES) (n = 4) to investigate clonality. We show that the lobular and ductal components within MDLs arise from a common ancestor, as opposed to the collision of two independent tumours, and that the phenotypic diversity is a result of clonal progression.

Clinical cohort
Clinical and pathological data were obtained for a total of 82 patients, and samples were obtained for 51 patients (supplementary material, Table S1). Archival formalin-fixed paraffin-embedded (FFPE) tissue was obtained from local pathology laboratories, and frozen tissue and blood DNA were obtained from the Bris-

Immunohistochemistry (IHC)
IHC for the E-cadherin complex was performed (on whole tissue sections; antibodies and staining conditions are detailed in supplementary materials and methods) and scored centrally. E-cadherin and its binding partners were scored as negative in the absence of expression, or positive when there was complete, linear membranous staining. Aberrant staining was defined as either fragmented membranous staining (i.e. non-linear) or staining with a diffuse cytoplasmic localization. Positive/aberrant staining was recorded in those cases that showed both positive membranous staining and aberrant staining. The expression of ER, progesterone receptor (PR) and human epidermal growth factor receptor 2 (HER2) for each case was retrieved from pathology reports.

Genomic analyses
Tumour-rich regions were microdissected and DNA was extracted with the Qiagen (Melbourne, Australia) QIAamp DNA Micro Kit. cCGH was performed and the results were analysed as previously described [10], and WES was performed and the results were analysed as described in supplementary materials and methods.

Unique clinical and pathological features of MDLs
The clinical and pathological features of 82 MDLs were compared with those of a cohort of 256 IDCs and 64 ILCs from the Queensland Follow-Up (QFU) cohort [11][12][13][14] (Table 1). The average age at MDL diagnosis was 57 years, which was significantly younger than that for ILC (62 years; p = 0.0083), but not that for IDC (58 years). The frequency of grade 2 MDL tumours was significantly higher than that of IDC tumours (p = 0.0254), but lower than that of ILC tumours (p = 0.0006). MDL patients presented more frequently with lymph node metastases than did IDC and ILC patients (p = 0.0033 and p = 0.0097, respectively).
In situ carcinoma was diagnosed in 58 of 82 cases, with DCIS in 51 of 58 cases (87.9%). Of these, 35 (60.3%) presented with DCIS alone, 16 (27.6%) presented with DCIS coincident with LCIS, and just seven (12%) presented with LCIS only. The distribution of DCIS and LCIS in MDLs was significantly different from that in the QFU cohort (p < 0.0001; Table 1). For instance, there were significantly fewer cases with LCIS only in the MDL cohort than in the ILC cohort (p < 0.0001). Regarding hormone receptor expression, MDLs expressed ER and PR more frequently than IDCs

The E-cadherin adhesion complex is aberrant but not lost in the lobular component of MDLs
A characteristic feature of the ILC phenotype is cellular discohesion as a result of E-cadherin complex dysfunction [15,16], and is classically observed as complete loss of E-cadherin and β-catenin staining, with relocalization of p120 catenin to the cytoplasm. The integrity of E-cadherin and its binding partners was assessed ( Figure 1). Both in situ and invasive components of ductal morphology typically showed normal, membranous staining for E-cadherin, β-catenin, and p120 catenin. The majority of LCISs were negative for E-cadherin (70%) and β-catenin (88.9%), with cytoplasmic p120 catenin (62.5%). However, in the invasive lobular compartment, just 17.6% of cases (9/51) had lost E-cadherin expression, with most showing aberrant staining.
We compared patterns of E-cadherin complex expression between lobular components of the MDLs with a cohort of 148 pure ILCs and associated LCISs [17]. Both in situ and invasive components of pure ILCs showed loss of E-cadherin and β-catenin significantly more frequently than the lobular components of MDLs; conversely, the lobular components of MDLs showed The E-cadherin adhesion complex shows similar expression patterns in LCIS lesions in both MDLs and genuine ILCs, whereas the expression of these proteins in the invasive lobular components of MDLs is significantly different from that seen in ILCs. Chi-square analysis: ****p < 0.000000001; ***p = 0.0008; **p < 0.005; *p < 0.01. E-cad, E-cadherin; neg, negative; ns, not significant; pos, positive; β-cat, β-catenin. aberrant expression of the complex significantly more frequently than pure ILCs ( Figure 1B; p < 0.0001).
Interestingly, among cases with both LCIS and ILC, the staining pattern for E-cadherin was concordant in these components in eight of 10 cases. In all cases with DCIS in the absence of LCIS, the invasive lobular component showed aberrant E-cadherin staining. As exemplified by MDL4 in Figure 2A, some cases had a 'transition zone' in which it appeared that there was a direct morphological transition from ductal to lobular phenotypes coinciding with a progression from linear ductal E-cadherin staining to an aberrant appearance as fragmented membranous and/or punctate, cytoplasmic staining in lobular regions.

Morphologically and spatially distinct components of MDLs can arise from a common precursor
To explore clonal relationships between MDL components, morphologically distinct regions of DCIS, LCIS and invasive tumours with a ductal or lobular growth pattern were laser capture-dissected or needle-dissected whereas the E-cadherin staining in the lobular component remains strong, but is observed as cytoplasmic or discrete perinuclear dots, and β-catenin staining is weak to negative. Note the insets showing high-power fields. (B) Exome sequencing analysis of the invasive components identified a number of alterations; 134 variants were common to both lesions, including non-synonymous changes in the breast cancer driver genes (albeit at low frequencies in the ductal component) SF3B1, TBX3, and TP53. Nine variants were unique to the ductal region, and included the cancer driver gene HRAS; 27 variants were unique to the lobular component, including ESR1. Variants included in this analysis were both synonymous and non-synonymous, and also low-frequency alleles not definitively called as 'mutations' by the analysis pipeline. Numbers in black represent shared variants along 'trunks'. Numbers unique to each branch are coloured according to morphological compartment; branch length is defined as SNV number, and is to scale as shown. (C) cCGH analysis of the invasive components identified DNA copy number alterations common to both lesions, suggesting that they were derived from a common neoplastic clone. Chromosomal regions in blue/red were gained/deleted in the relevant lesion. from seven cases (supplementary material, Table S1), and analysed for DNA copy number alterations by cCGH (MDL1-MDL4, n = 4) or nucleotide variation by exome sequencing (MDL4-MDL7, n = 4). All samples were from FFPE diagnostic blocks, because appreciating the detailed morphology/cytology of individual components and understanding the topographical relationships between lesions within a case were critical to understanding the clonal nature of the molecular data obtained. Fresh frozen material may provide better-quality input DNA for these types of analysis, but lack the morphological detail provided by FFPE samples.
Detailed morphological annotation and molecular data for cases analysed by cCGH are shown in Figure 2 and supplementary material, Figures S1-S3. Illustrating the complexity of some cases, MDL1 contained DCIS, LCIS and both invasive ductal and lobular components within the same tissue block. DCIS and LCIS were sometimes admixed within the same duct, but showed different cytological features and staining for E-cadherin and β-catenin (supplementary material, Figure S1). We found that all morphological components analysed within the same case shared a considerable number of chromosomal aberrations (more than five events), suggesting they derived from a common ancestor before undergoing genotypic and phenotypic (morphological) change (Figure 2; supplementary material, Figures S1-S3 and Table S2).
To  Table S6). These proportions are expected, as not all of the genes or mutated alleles will be expressed [18]. A total of 261 single-nucleotide variants (SNVs) and small insertion-deletions were identified (supplementary material, Tables S3 and S4); 142 SNVs were validated by iPlex or RNAseq (supplementary material, Tables S5 and S6). The proportion of shared variants within a case was considerable, clearly indicating clonal relatedness between all components within individual tumours. Unique alterations were also identified, permitting assessment of the genetic distance between different components. Phylogenetic relationships between different components in four independent cases were investigated using a deductive reasoning approach, supported by PyClone analysis [19] (supplementary material, Data S1) and results are discussed below (Figures 2 and 3).
Invasive ductal and lobular components of MDL4 were analysed with cCGH and by exome sequencing ( Figure 2B, C). Both invasive components were genetically complex, with considerable numbers of gains and losses, and SNVs. The two components shared 62% of gains and losses, and 72.7% of SNVs ( Figure 2; supplementary material, Tables S3 and S4), with shared SNVs affecting the cancer driver genes TP53 [p.Tyr220Cys, moderate effect (SNPeff), common mutation with 354 reported in COSMIC], ESR1 (p.Ser154Leu, moderate effect, none recorded in COSMIC), and TBX3 (p.Glu402_Pro403insGluGlu moderate effect, none recorded in COSMIC). Strong evidence supports a common ancestral relationship between lesions and likely progression of DCIS to IDC to ILC (in the absence of LCIS). The transition from a ductal to a lobular growth pattern was underpinned by a change in E-cadherin complex function, as demonstrated by disparate immunohistochemical staining between the components (Figure 2).
In MDL5, DCIS, IDC and ILC were sequenced and, although lesions showed a relatively low mutational load, >75% of SNVs identified were shared among lesions, including in the driver gene TBX3 (splice site alteration; predicted high effect, none recorded in COS-MIC). Although CDH1 was not mutated, the lobular component was negative for E-cadherin, β-catenin and p120 catenin by IHC ( Figure 3A; supplementary material, Figure S4).
In MDL6, DCIS, IDC, pleomorphic LCIS and pleomorphic invasive lobular carcinoma (PLC) growth patterns were identified. The ductal components were E-cadherin-positive, whereas the pleomorphic lobular components were negative, and DCIS and pleomorphic lobular carcinoma in situ (PLCIS) were admixed in some ducts ( Figure 3B; supplementary material, Figure S5). Seventy-eight variants were shared by all four components, including mutations in several driver genes (e.g. BRCA2, p.Lys936fs, high effect; SMAD4, p.Arg361His, moderate effect, 76 events reported in COSMIC; TBX3, p.Pro42fs, high effect). The SNV load was higher than in other cases, and the PLC component was associated with the highest proportion of unique variants (53%). The pathway to invasion from the PLCIS branch featured 327 unique variants, including changes in several driver genes (e.g. NCOR1, p.Ser1117*, high effect, none recorded in COSMIC; PARP4, p.Gln147Glu, moderate effect, alternative missense mutations recorded in COS-MIC). A CDH1 nonsense mutation (p.Gln610*, high effect, reported missense mutations at that nucleotide) was identified in the DCIS (mutant allele frequency of 16%), PLCIS (85%) and PLC (75%) components; this variant was also present in three of 257 reads (1%) in the IDC. We restricted the CDH1 mutation annotation to the pleomorphic branch in the tree after considering both the high frequency of the mutations in the pleomorphic regions and the admixture of cells. It is expected that the presence of this variant in ductal components results from contaminating 'lobular' cells, particularly given the admixture of DCIS and PLCIS within the same ducts, which is only evident on E-cadherin-stained sections (as indicated by the arrow in Figure 3B). Interestingly, PLCIS also harboured a 4-bp deletion in CDH1 (p.Arg796fs, high effect at 10% frequency, not recorded in COSMIC; supplementary material, Table S4).
MDL7 presented a vastly more complex situation, with the diagnostic pathology report showing a clear topography of four coincident tumours: ILC, IDC, and two MDLs (MDL1 and MDL2) ( Figure 3C;  Table S4. supplementary material, Figure S6). Despite the morphological differences and physical distance between tumours, exome sequencing demonstrated that the most physically distant lesion contained the most unique variants (ILC, n = 3), as did MDL1, with which the ILC also shared several alterations ( Figure 3C; supplementary material, Table S4). Indeed, one is reminded of the original 'sick lobe' studies; despite the physical distance, the shared changes indicate a clear ancestral history [20]. Notably, no breast cancer driver genes were noted to be altered in this case.

Discussion
This study investigated the clinical, pathological and molecular features of a large cohort of MDLs, and corroborates that MDLs constitute a distinct morphological entity [3,4] with both unique biological features and some similarities with pure IDCs and ILCs. There is now strong evidence independently confirming that MDLs: are mostly grade 2 (58%); are more frequently associated with DCIS than with LCIS; are frequently lymph node-positive (68.3%); are more frequently ER-positive than IDCs; and, from a clinical point of view, have a prognosis that may lie somewhere between that for IDCs and that for ILCs [3,4].
Here, we wished to gain a better understanding of the underlying aetiology of mixed tumours. The current multistep model of progression considers that IDC arises from DCIS, whereas ILC arises from LCIS, via a low-grade or high-grade pathway of tumourigenesis [21]. It has so far been unclear how MDLs would fit into this theory. Two observations are worthy of elaboration as important insights. First, MDLs are more frequently associated with DCIS than with LCIS. In our series, DCIS was present in 87.9% of cases (60.3% as pure DCIS; 27.6% as coincident DCIS and LCIS), similar to what was reported by Rakha et al [3] (89%), whereas just 12.1% of cases had pure LCIS. It is possible that LCIS was present in more cases but was just not detected, or was completely transformed into the invasive lobular component. However, the differing distributions of DCIS and LCIS imply that the invasive lobular component of an MDL may be the result of evolution from a tumour clone of ductal origin in cases without LCIS.
Second, the expression status of E-cadherin and the associated complex protein β-catenin in the lobular components of MDLs was different from that seen in the ductal components of MDLs, from that seen in IDCs, and, importantly, from that typically seen in ILCs [22]. In the vast majority of ductal lesions, E-cadherin is expressed and linearly distributed on the cell membrane. ILCs show complete loss of E-cadherin and β-catenin expression in 90% of cases, and expression of these proteins but aberrant localization in the remaining 10% of cases [23]. E-cadherin expression in MDLs has previously been characterized as either positive (ductal-like) or negative (lobular-like [22]). Here, we separately recorded the expression patterns in ductal and lobular areas, and, in most cases, the ductal components showed normal membranous staining. In 70.6% of cases, E-cadherin and β-catenin were expressed but were distributed in an aberrant localization, either as fragmented membrane staining, as diffuse cytoplasmic accumulation, or as strong focal perinuclear accumulation. This observation suggests that dysfunction of cell-cell adhesion may underpin the lobular growth pattern in MDLs, as it does in ILCs, although the mechanism of disruption is likely to be different between ILCs (CDH1 mutation, loss, or silencing [16]) and MDLs.
The primary objective of the molecular approach used was therefore to determine whether an invasive lobular growth pattern can evolve from a ductal origin, and, if so, whether we can uncover the molecular mechanism underpinning this progression. Although frozen samples are best for molecular analyses, we studied FFPE diagnostic cases to provide morphological clarity in the topographical relationships between the lesions analysed. Using both cCGH and exome sequencing we demonstrated that all morphological components within a given case shared a significant number of mutations (copy number alterations or SNVs), unequivocally demonstrating that all lesions within a case shared a common ancestry. Historically, these molecular approaches have demonstrated similar clonal relationships between columnar cell lesions and DCIS, between DCIS and associated IDC, between DCIS and coincident LCIS, and between LCIS and coincident ILC [10,24], but, as far as we are aware, not between the different invasive components of MDL. This supports findings obtained with methylation analysis of the HUMARA gene to demonstrate that these components were related [25].
Historically, E-cadherin expression in MDLs has been characterized as either positive (ductal-like) or negative (lobular-like [22]); however, we showed previously the presence of a detectable but aberrant E-cadherin staining pattern in some lobular-like tumours [23]. This implies that the evolution of the invasive lobular component may differ according to the presence of LCIS; there may be more than one pathway to the development of a lobular phenotype. These hypothetical concepts regarding the evolution of different morphological components of MDLs are summarized in Figure 4.
The largest molecular study of MDLs to date was performed by Ciriello et al [15], involving multi-omic profiling of 88 MDLs as part of a larger breast cohort. The main focus of the analysis was to show whether these tumours could be classified as either ILC-like or IDC-like, based on the mRNA and somatic mutations; the conclusion was that MDLs do not represent a molecularly distinct entity, but are closely related to either IDC or ILC. Notably, the ILC-like subgroup constituted ∼25% of MDL cases, and showed a higher rate of CDH1 mutations and lower expression of E-cadherin mRNA and protein. These findings could be interpreted in two ways. First, they may not have any biological relevance, owing to a critical lack of morphological annotation of the samples, and hence may simply reflect the proportions of the ductal or lobular components that underwent profiling. Alternatively, the authors may have highlighted, with their broad analysis, that ILC-like cases represent those that may arise via LCIS and the loss of E-cadherin by gene mutation, whereas the IDC-like cases represent those arising via the ductal pathway, with no LCIS component and aberrant E-cadherin expression for unknown reasons.
The molecular analyses identified a number of defining genomic alterations; in particular, the exome sequencing identified several key cancer driver mutations, particularly in the common ancestral evolutionary arm of each case. For example, we identified a frameshift deletion in BRCA2 (c.2806_2809delAAAC) in the evolution of MDL6 that involved a pleomorphic lobular component; a relationship between BRCA2 variants and pleomorphic lobular breast cancer has previously been suggested [26]. Large-scale genomics consortia have previously identified a number of breast cancer driver genes, including TBX3 [27]; indeed, TBX3 mutations account for ulnar-mammary syndrome [28], and, in breast cancer, appear to result in the loss of transcriptional repressor function [29]. Intriguingly, TBX3 alterations are enriched in ILCs as compared with IDCs (9.5% versus 1.6% [18]), and we, too, report genetic alterations in TBX3 in three of four WES cases, all of which were early events. Functionally, TBX3 has been shown to repress E-cadherin expression in melanoma [30], and it is a downstream target of the Wnt-β-catenin pathway [31]. The interplay between TBX3 mutations and the E-cadherin adhesion complex in MDLs is a fascinating area for future research.
This study demonstrates that the different morphological components present within an MDL tumour are clonally related and not the result of a collision of multiple independent tumours, thus supporting the idea that MDL tumours represent a distinct clinical and biological entity. We show that, in some cases, the divergence of the morphological components may occur early during tumour evolution (both DCIS and LCIS are present) or later during tumour progression (only DCIS detectable). The cases with late-occurring divergence may arise via a ductal-like pathway of progression, and these data emphasize the possibility that a lobular-like phenotype can arise via a modified ductal pathway. As previously detailed [3], the clinical conundrum surrounding a mixed tumour is that the good prognostic features of one component may have no prognostic value in the presence of a component with more aggressive features, regardless of which is the most conspicuous component. Clinically, there is little value in broadly categorizing MDLs into either lobular-like or ductal-like, and a more detailed assessment is required. Indeed, we have shown that the morphologically disparate regions harbour mutations in several breast cancer driver genes, which may predict targeted therapeutic options in the future. The evaluation of the true molecular profile of breast tumours is greatly influenced by tumour cellularity, and, critically in terms of discovery approaches, known proportions of morphologically distinct clones. The molecular analysis presented here highlights the genomic differences and similarities between the morphological regions of the MDLs, and contributes to our broader understanding of the genetic picture of breast cancer.

Supplementary figure legends
Data S1. Pyclone analysis