Background
Lobular carcinoma
in situ (LCIS) is a non-invasive breast lesion that is typically found incidentally on biopsy but is also often seen in the presence of invasive lobular carcinoma (ILC), which accounts for 10–15% of all invasive breast carcinomas. Hwang et al. showed that ILC and co-existing LCIS share many of the same genetic aberrations [
1]. Furthermore, Vos et al. demonstrated the presence of the same truncating mutations in
CDH1 and loss of heterozygosity (LOH) of the wild-type E-cadherin in the LCIS component and adjacent ILC [
2].
These studies suggest that LCIS is a non-obligate precursor of ILC in a manner analogous to ductal carcinoma
in situ (DCIS) preceding invasive ductal carcinoma of no special type (IDC). However, the risk of invasive cancer after LCIS is lower than that with DCIS (2–11 times greater than the risk in the general population, in contrast to DCIS with 20 times greater risk) [
3,
4], and the overall rate of progression of pure LCIS to ipsilateral ILC has been shown to be <10% ten years after the diagnosis of LCIS [
5,
6]. LCIS is also considered a risk factor for future breast cancer, as not all invasive disease post LCIS presents as ILC and, unlike DCIS, LCIS is also a risk factor for developing invasive cancer in the contralateral breast [
7]. Fisher et al. reported that 80% of invasive disease post LCIS is ILC; however, it is likely that the patients in that study did not represent typical cases of classical LCIS (cLCIS) as many were initially diagnosed as having DCIS, but on pathological review were determined to have LCIS [
8]. A more recent study has reported much lower rates of ILC post LCIS (27%), although this was still higher than the expected 10–15% [
6].
The timescale for the development of invasive carcinoma after an initial diagnosis of LCIS in either breast varies greatly between individuals; one study demonstrated that two thirds of patients developed invasive disease within 15 years; however, another study found that 50% of patients developed ILC up to 15 to 30 years later. This has led some to argue against LCIS as a non-obligate precursor lesion and to suggest that “pure” LCIS may have a different molecular profile compared to LCIS that co-exists with invasive disease, and that the molecular studies cited above have focused on LCIS with associated ILC, rather than pure LCIS.
There are limited studies of pure LCIS but generally these do show similar genetic changes to LCIS associated with ILC, with 16q loss and 1q gain being the most common chromosomal abnormalities. Whilst Mastracci et al. suggested that LOH at 16q was infrequent in 13 cases of pure LCIS [
9], comparative genomic hybridization studies on 17 cases of pure LCIS revealed 16q loss in 88% of cases, being the sole detected alteration in 29% [
10]. In the latter study, 1q gain was the second most common change, occurring in 41% of tumours and in all cases associated with 16q loss. There is also evidence of E-cadherin loss in both LCIS and atypical hyperplasia with
CDH1 mutations being common in LCIS, but rare in atypical lobular hyperplasia (ALH) [
11].
The increased breast biopsy rate associated with screening mammography has led to an increase in the diagnosis of pure LCIS in postmenopausal women [
12] with around 3% of needle biopsies identifying pure LCIS [
13]. Current guidelines for patients with a diagnosis of LCIS highlight the need for increased surveillance of both the affected and contralateral breasts; however, the optimum management of women with pure LCIS is unclear, as not all women with LCIS will develop invasive disease. Currently in the UK patients with pure LCIS do not receive any further treatment and, even if incompletely excised, no further surgery is performed. There is now convincing evidence from large randomized chemoprevention trials, that 5 years of endocrine therapy reduces the risk of invasive disease after a diagnosis of LCIS by 50%; the NASBP-P1 study demonstrated that 5 years of tamoxifen reduced the development of invasive disease after LCIS from 11% in the control group to 4% in the tamoxifen-treated group [
14]. Similarly 5 years of exemestane reduced invasive disease from 13 to 6%, respectively [
15]. However, despite this evidence, the use of chemoprevention for LCIS has not become common practice; seemingly clinicians and patients feel that as many cases of LCIS do not progress to invasive disease, the benefits of chemoprevention do not outweigh the potential side effects. It would therefore be invaluable to have biomarkers to predict the likelihood of progression of LCIS, so that appropriate screening and treatment can be offered.
Current biomarker data in LCIS are very limited [
16]. There is some evidence to suggest the risk of subsequent invasive disease is associated with high Ki67 expression [
17] and that increased expression of hsa-miR-375 contributes to lobular neoplastic progression [
18]. Other studies have shown that the five biomarkers known to be important in invasive breast cancer (oestrogen receptor (ER), progesterone receptor (PgR), c-erbB-2, p53 and Ki-67 expression) do not predict progression of LCIS [
19]. One of the problems with studies that have tried to identify biomarkers that predict LCIS recurrence is the small number of cases analysed due to the rarity of pure LCIS.
The aim of the present study was to identify genetic changes that could be used as biomarkers of progression of LCIS to invasive disease. Ideally this should be done in a cohort of patients with pure LCIS, who have progressed to invasive disease, compared to a cohort that have not, but such patients are very rare. We therefore chose to do our discovery phase using patients with pure LCIS and comparing their genetic profiles to patients with LCIS who presented contemporaneously with associated ILC, based on the hypothesis that the latter represents LCIS that has already progressed.
Discussion
Our study confirms the finding of previous studies that LCIS has the same molecular changes as co-existing ILC. We have also shown that pure LCIS and LCIS associated with ILC (inv-LCIS) have very similar SCNAs, supporting the hypothesis that pure LCIS is a precursor lesion. Only five regions were significantly different between pure-cLCIS and inv-cLCIS and these were all more common in pure LCIS, so did not represent markers of progression. Three of the regions were small, containing just one gene: chr17 - RAB11FIP4; chr20 - MACROD2, overexpression of this gene has been implicated in oestrogen-independent growth [
27]; and chrX - KLF8, oncogenic in ovarian but not breast cancer cell lines [
28]. There was also a similar region on chr18, that was more common in both pure and inv-cLCIS compared to ILC, containing YES1, a SRC proto-oncogene that has recently been identified as a possible therapeutic target in basal breast cancer [
29]. It is possible that these regions contain genes that hinder development of the invasive phenotype. A similar finding was made in DCIS (albeit with different genomic regions) and the authors hypothesized that these regions could contain genes that provide a selective advantage under local conditions but also inhibit invasion [
30].
We identified four SCNAs that increased in frequency from pure cLCIS to inv-cLCIS and finally ILC: loss of 6q14.1-27, 8p23.2-23.1, and 22q13.31-13.33 and gain of 11q13.3. For the three regions of loss the most significant increase in frequency came in the transition from inv-cLCIS to cILC. Both 6q and 8p23 loss have also been found to be more common in invasive ductal carcinoma (IDC) than in paired DCIS [
30] suggesting that there may be tumour suppressor genes in these chromosomal regions important in the transition of
in situ to invasive carcinoma in both DCIS and LCIS. A recent study integrating SCNAs, promoter methylation and gene expression profiles in luminal B breast cancers showed that 88% of the potential tumour suppressor genes were located on 6q [
31]. Deletions of 8p have been described in 50% of IDC and 37% of lobular cancers [
32] and the minimal region of loss in our study (8p23.2-23.1) includes
MCPH1, a potential tumour suppressor gene [
33,
34] encoding a DNA damage response protein which has also been implicated in breast cancer predisposition [
35]. Similarly, loss of 22q13 has been described in IDC (not present in paired DCIS) and ILC (but also present in paired LCIS) [
36,
37].
Gain/amplification of 11q13.3 was considered the most useful SCNA to take forward as a potential practical biomarker of LCIS progression as unlike the regions of loss, the main increase in frequency occurred between the pure-cLCIS and inv-cLCIS sub-groups. There was also evidence of increasing amplitude of copy number gain between the paired inv-cLCIS and ILC samples. The region contains
CCND1 and other potential oncogenes such as
FGF3, FGF4 and
MYEOV and there is good evidence that this region is relevant to breast cancer [
38]. Two large studies of DCIS (approximately 400 patients) have reported amplification of
CCND1 in 10–12.6% of patients with pure DCIS and 14.8–17.4% of patients with DCIS associated with invasive breast cancer, with the majority of patients having amplification in the paired invasive component [
39,
40]. In a much smaller series of 20 patients with florid LCIS (a subtype putatively more likely to associated with invasive disease) 25% had
CCND1 amplification [
41]. This, together with the finding that
CCND1 amplification is usually homogeneous within breast carcinomas, suggests that it is an early event in the development of some breast cancers [
42].
Other studies have identified correlation between over-expression of cyclin D1 and amplification of 11q13 and this over-expression has been associated with an increased risk of recurrence in ILC [
43]. One early study of ILC suggested that cyclin D1 protein overexpression may play a role in the transition of LCIS to ILC, as the majority of ILC samples had over-expression but with little evidence in surrounding LCIS, although the number of patients with LCIS was not stated.
Our data support their finding that cyclin D1 over-expression is important in the transition of LCIS to ILC, but we have also shown that some cases of LCIS also over-express cyclin D1 and this may be a marker of progression to ILC [
44]. In a genome-driven classification of over 7500 breast tumours, amplification of 11q13.3 was associated with a sub-group of ER-positive breast tumours with a poor prognosis and chemo-resistance. This sub-group accounted for only 3.1% of tumours in this series of mainly IDC [
45]. However, in our study 24% of cILC had 11q13.3 amplification in keeping with other series showing 11q13.3 amplification is more frequent in ILC than IDC [
46]. In our validation set, albeit very small, we found that cyclin D1 over-expression may identify a subset of pure LCIS that is likely to progress to ILC. As around 40% of ILC have high expression of cyclin D1 [
43], this represents a significant subset of ILC.
Exome sequencing of this small number of LCIS and ILC did not identify any potential biomarkers of LCIS progression. It did, however, show that activating
PIK3CA mutations are as common in LCIS as
CDH1 mutations. Activating
PIK3CA mutations are well-described in both ILC and IDC, occurring in 48% of ILC (as the second most common mutations after
CDH1) and 33% of IDC [
47,
48].
PIK3CA mutations have previously been reported by Christgen et al. [
49] in 1/3 patients with LCIS associated with ILC and by Sakr et al. [
48] in 7/19 patients. We confirmed the frequency of
PIK3CA mutations in a larger set of LCIS by Sanger sequencing, which revealed no difference in the frequency of
PIK3CA mutations in inv-cLCIS compared to pure-cLCIS. The only other study to assess
PIK3CA mutations in pure LCIS was by Sakr et al. who found no evidence of
PIK3CA mutations in three patients with pure LCIS after targeted sequencing [
48]. The frequency of
PIK3CA mutations in ILC in our study is lower than that reported by The Cancer Genome Atlas (TCGA) and Sakr et al.; however, this is likely to be because both those studies used next-generation sequencing to assess the whole gene, whereas we targeted only the common mutations.
So, although they are not a useful biomarker of LCIS progression,
PIK3CA mutations are an early event in lobular tumorigenesis leading to abnormal proliferation of the breast epithelium, but importantly, they do not appear to be the critical event leading to invasive malignancy. This is supported by the findings of Ang et al. who identified frequent
PIK3CA mutations in non-invasive proliferative breast lesions including DCIS, inv-LCIS and one case of pure LCIS [
50,
51].
A comprehensive analysis of 127 ILC by the TCGA has shown that
CDH1, PIK3CA, TBX3, FOXA1 and
RUNX1 are the most commonly mutated genes [
47]. We found no evidence of
TBX3,
FOXA1 or
RUNX1 mutations in the four LCIS samples on which we performed exome sequencing, and identified only one
TBX3 mutation in the four ILC samples. The only other known driver gene mutated in LCIS was
ATRX, frequently mutated in neuroblastoma, low-grade glioma and glioblastoma but not common in breast cancer. In the present series we report splice site mutation in two additional well-known drivers of cancer,
MAP2K4 and RB1 in ILC, but not LCIS,
The TCGA data also showed that AKT signalling is strongly activated in ILC and homozygous losses of the
PTEN locus (10q23) occurred in 6% of ILC [
47]. We found no evidence of homozygous deletions of 10q23 in LCIS in our samples. One case of ILC did have a homozygous deletion at the PTEN locus, whilst this was not evident in the paired LCIS, suggesting it is a later event in lobular tumourigenesis. Interestingly the region encompassing
AKT3 was the locus most frequently amplified in both pure cLCIS and inv-cLCIS and paired cILC. So, although
PIK3CA mutations do not appear to be the trigger for malignant transformation in lobular cancer, it is possible that progression of LCIS may be related to acquisition of mutations or alterations in other components of the PI3K/Akt pathway; for example, expression profiling studies have shown that PIK3R1 is significantly downregulated in the stepwise progression from normal epithelium to LCIS to ILC [
52].
With the advent of next-generation sequencing intra-tumoural heterogeneity has been found to be widespread in invasive cancer; however, there are few data on the intra-tumoural heterogeneity of
in situ breast cancers. We developed a relatively crude method to assess heterogeneity using SNP arrays and this clearly showed that sub-clonal SCNAs increase in frequency from
in situ to invasive lobular carcinoma. It remains to be seen whether a more sensitive measure of clonal diversity could be used as a biomarker of progression to invasive disease, as it is in Barrett’s oesophagus [
53,
54]. Of interest, we have also shown evidence of passenger mutations in LCIS not transmitted to the invasive component, suggesting that, like invasive disease, there is an early sub-clone expansion process [
55], with at least one acquiring critical mutations and developing into invasive disease. Driver mutations that are sub-clonal in the pre-invasive state then become clonal in the invasive stage.