ESR1 gene amplification in breast cancer: a common phenomenon?

To the Editor

Using fluorescence in situ hybridization on tissue microarrays (FISH-TMA), Holst et al.¹ recently reported amplification of ESR1, the gene encoding estrogen receptor alpha, in 21% (358 of 1,739) of breast cancers. This prompted us to analyze ESR1 copy number using either FISH-TMA or array CGH (aCGH) in a combined series of 725 breast cancers (see Supplementary Methods online for details of series and methodology).

We analyzed a total of 334 cases by FISH-TMA using the same FISH probe (end-sequence verified) for ESR1 (RP11-450E24) as reported by Holst et al.¹ We carried out automated scoring of FISH signals using Metacyte (Metasystems) and considered cases to be amplified when the ESR1 to centromere 6 ratio was ≥2 (ref. ²). We found ESR1 to be amplified in four cases (1%, Supplementary Fig. 1 and Supplementary Table 1 online). Digital FISH images for ESR1 and centromere 6 on the breast cancer TMAs are publicly available (http://www.gpecimage.ubc.ca/).

Holst et al.¹ validated ESR1 gene amplification using a quantitative PCR (qPCR) assay comparing DNA copy numbers of ESR1 and ESR2 in four cases with and without ESR1 amplification (as determined by FISH). We applied the same qPCR assay to 125 breast tumors that were included on our TMA. We observed increased DNA copy number of ESR1 as compared to ESR2 in 20 of 125 breast tumors (16%): two cases were considered amplified by FISH and 18 cases had a normal ESR1 to centromere 6 ratio. The use of ESR2 as the reference gene introduced an additional bias to the well-known limitations of qPCR in scoring copy number gains; this locus was lost in 12% of cases using aCGH (data not shown). We therefore used as controls two additional genes rarely altered in breast cancer: EIF5B (2q11.1) and PVR (19q13.2). Using this more rigorous qPCR assay, we found that three samples showed amplification of ESR1 (and a further two when normalizing separately with each control gene) and only one of these was considered amplified by FISH.

We studied a further 391 breast cancers with aCGH, the methodology first used by Holst et al. to identify ESR1 amplifications. Three different platforms were used: a custom 30K oligonucleotide array^3,4 (n = 171), an OncoBAC array⁵ (n = 143) and the Agilent 244K array⁶ (n = 77). As shown in Table 1, both copy number gain (18 of 391, 5%) and amplification (4 of 391, 1%) at the ESR1 locus were rare events. In contrast, we observed the expected frequency of commonly amplified regions, such as ERBB2 at 17q12 and CCND1 at 11q13 (refs. ^3,5).

Table 1.

Analysis of copy number abnormalities at the ESR1 locus using three different acGH platforms

aCGH platform	No. of probes at ESR1 locus		Breast tumors		ER status
		N	Gains (%)	Amplification (%)	ER+ (%)	ER− (%)
Oligo aCGH2	4	171	13 (8)	2 (1)	113 (66)	58 (34)
BAC aCGH4	1	143	5 (3)	0 (0)	94 (66)	49 (34)
Agilent 244K aCGH	33	77	7 (9)	2 (3)	75 (97)	2 (3)

The reported ESR1 amplicon was approximately 600 kb in size¹, and each of the aCGH platforms used here has the capability to detect an amplicon of this small size (see Supplementary Methods for details). The oligonucleotide 30K array contained four probes at the ESR1 locus (Supplementary Fig. 2a online), and when we used our segmentation and calling algorithms, we found that the overall frequency of ESR1 copy number gains was low (Supplementary Fig. 2b and Supplementary Table 2 online), which contrasts with the high frequency of ERBB2 copy number gains observed in the same tumors (Supplementary Fig. 2c). We verified this result by application of an algorithm specifically designed to detect low-level focal amplifications. The OncoBAC array contained a single BAC clone that spanned the ESR1 locus (Supplementary Fig. 2a). The published OncoBAC array data was reanalyzed for this investigation (Supplementary Methods) and also showed a low percentage of copy number gains (Supplementary Tables 3 and 4 and Supplementary Fig. 3 online). The third aCGH platform contained 33 probes spanning from 152.2 to 152.5 Mb on 6q21, fully encompassing ESR1. Using this array, we observed a similar low frequency of amplification in the breast cancers (Supplementary Table 5 and Supplementary Fig. 4 online).

The results reported here (ESR1 amplification in 1% of breast cancers) are clearly different from those published in this journal (ESR1 amplified in 21% of breast cancers) by Holst et al.¹. Several explanations for this disparity could be possible. The most trivial, given that Holst et al.¹ reported that ESR1 amplification was exclusive of estrogen receptor (ER)-positive cases, would be that our series had a substantially larger proportion of ER-negative cases. However, that is not the case, as 69% of the 725 cases studied here were ER positive. Furthermore, the use of the CGH arrays described above rules out difficulty in identifying the amplicon because of its small size as a possible source of discrepancy. It is possible that natural copy number variation (CNV) in the reference DNA could mask our ability to observe amplification at the ESR1 locus in the aCGH experiments. However, Redon et al.⁷ reported no copy number variation at the locus where the clone used by Holst et al. maps. Moreover, we investigated this further in the oligonucleotide 30K array data by examining the signal in the reference channel at the ESR1 locus and found no evidence of CNV. Thus, CNV is unlikely to be the explanation for the discrepancy.

The key difference between our study and that of Holst et al.¹ is the methodology for scoring FISH-TMA (manual vs. automated) and the criteria used to call amplifications. Holst et al.¹ scored as amplified not only cases with an ESR1 to centromere 6 ratio ≥2 but also “tumors with tight signal clusters…independent of their ESR1/centromere 6 ratio.” They report using previously the same definition of amplification for CCND1, ERBB2, MDM2 and MYC⁸. However, review of this publication reveals use of a single amplification criterion: signal ratio ≥2. As the authors state that “most amplified cases showed a clustered arrangement of additional ESR1 copies”¹, we interpret this to mean that most of the ESR1-amplified cases were scored using subjective criteria. In contrast, the automated system we used to score FISH signals employs specific measurement algorithms to detect and quantify such clustered signals. We have previously reported a high correlation between manual and automated scoring of FISH signals and have implemented the use of this system for the scoring of gene amplification events².The system is FDA approved for the automated scoring of ERBB2 gene amplification (Metasystems). Using this objective set up, we found that only one case had a tight cluster of signals.

In summary, our data compiled from several institutions and obtained using two different techniques does not validate the findings of Holst et al., and we conclude that ESR1 amplification in breast cancer is a rare event of unknown clinical significance.