The prevalence of immunohistochemically determined oestrogen receptor positivity in primary breast cancer is dependent on the choice of antibody and method of heat-induced epitope retrieval – prognostic implications?

Abstract

Background. Oestrogen receptor (ER) status is important for the choice of systemic treatment of breast cancer patients. However, most data from randomised trials on the effect of adjuvant endocrine therapy according to ER status are based on the cytosol methods. Comparisons with immunohistochemical methods have given similar results. The aim of the present study was to examine whether different ER antibodies and heat-induced epitope retrieval (HIER) methods influence the prevalence of ER-positivity in primary breast cancer. Material and methods. This study is based on patients included in a clinical trial designed to compare the effect of two years of adjuvant tamoxifen versus no adjuvant systemic treatment in premenopausal women. From 1986 to 1991, 564 patients from two study centres in Sweden were enrolled and randomised. Patients were randomised independently of ER status. In the present study, ER status was assessed on tissue microarrays with the three different ER antibody/HIER combinations: 1D5 in citrate pH 6 (n = 390), SP1 in Tris pH 9 (n = 390) and PharmDx in citrate pH 6 (n = 361). Results. At cut-offs of 1% and 10%, respectively, the prevalence of ER-positivity was higher with SP1 (75% and 72%) compared with 1D5 (68% and 66%) and PharmDx (66% and 62%). At these cut-offs, patients in the discordant groups (SP1-positive and 1D5-negative) seem to have a prognosis intermediate between those of the double-positive and double-negative groups. Comparison with the ER status determined by the cytosol-based methods in the discordant group also showed an intermediate pattern. The repeatability was good for all antibodies and cut-offs, with overall agreement ≥ 93%. Conclusion. The present study shows that the choice of antibody and HIER method influences the prevalence of ER-positivity. We suggest that this be taken into consideration when choosing a cut-off for clinical decision making.

In breast cancer, the determination of oestrogen receptor (ER) status in tumour tissue is important in the choice of adjuvant therapy. For patients with ER-positive breast cancer, adjuvant endocrine therapy is indicative, whereas for patients with ER-negative disease this treatment is not beneficial [1,2]. Not only the presence of ER, but also the level of ER-positivity, seems to be associated with the sensitivity to endocrine therapy [3].

Over the years, the methods for ER analysis have changed. They started with ligand-binding assays (LBA) and enzyme immunoassay (EIA) in cytosol samples from fresh frozen tissue. These techniques were replaced by immunohistochemical (IHC) methods in formalin-fixed paraffin-embedded tissue, using monoclonal antibodies. Most of our knowledge about the importance of ER as a predictive marker for the response to endocrine therapy, especially about long-term clinical follow-up, is based on cytosol techniques. However, in a number of studies, ER has also been measured with IHC, alone or in addition to LBA/EIA, and related to clinical outcome after endocrine therapy. The overall conclusion is that LBA/EIA and IHC provide similar predictive value for the response to endocrine treatment [4–7].

Though ER has been used for decades now as a treatment-predictive test, no standardisation of the IHC test has emerged and thresholds predicting response to endocrine treatment have varied. To distinguish ER-positive and ER-negative breast cancer in clinical routine, the previous commonly applied cut-off of 10% positive tumour cells has recently been replaced by 1%, according to international recommendations [8,9]. Another classification, the Allred score, considers not only the percentage of positive cells but also the staining intensity of positive cells [10]. The continuous development of the IHC method has reached a limit with very intense dark staining using new epitopes and powerful staining systems. Pathology departments, adhering to quality assurance programmes such as United Kingdom External Quality Assurance System (UK NEQUAS) or NordiQC, often obtain excellent marks for the staining when using the most modern methods for ER assessment. However, these quality assurance programmes only consider the quality of the IHC staining itself in a limited number of cases. They do not consider a possible influence on the prevalence of ER-positivity and its predictive value. To improve the accuracy of testing, the American Society of Clinical Oncology/College of American Pathologists recommended that only methods with pre-analytical and analytical components conformed exactly to clinically validated assays or compared with the clinically validated assays showing 90% concordance for the ER-positive category and 95% concordance for the ER-negative category, with positivity defined as ≥ 1% stained nuclei, be used to predict response to endocrine therapy [9]. If methodological changes [11] [e.g. new tissue processing, new antibodies, changes in heat-induced epitope retrieval (HIER) and/or detection methods] are implemented without such validation [12], the long-term consequences will influence the prevalence of ER-positivity and the predictive value of ER. The importance of the choice of antibody for the prevalence of ER-positivity has been demonstrated in a study by Cheang and co-workers [13] where the proportion of ER-positive breast cancers increased from 63% to 71% when SP1 antibody was used instead of 1D5. In that study [13], SP1 was found to be a better prognostic marker for breast cancer-specific survival than 1D5.

The aim of the present study was to compare three different antibodies (1D5, SP1, and the ER part of ER/PR PharmDx^TM, in this study denoted PharmDx) with different HIER in relation to the prevalence of ER-positivity and recurrence-free survival (RFS) in a well-defined cohort of breast cancer patients with primary stage II breast cancer.

Material and methods

Study design

The patients included in the present study were enrolled in a clinical trial designed to compare the effect of two years of adjuvant tamoxifen (TAM) versus no adjuvant systemic treatment in premenopausal women with stage II breast cancer. From 1986 to 1991, 564 patients were randomised from two study centres in Sweden. Patients were randomised independently of ER and progesterone receptor status. The characteristics of this trial have previously been described in detail [14]. Adjuvant poly-chemotherapy was administered to fewer than 2% of the patients.

Tissue microarray

From representative areas of the formalin-fixed and paraffin-embedded invasive breast cancer tissue samples, two core biopsies (0.6 mm in diameter) were punched out and mounted into the recipient block using a tissue array machine in accordance with the manufacturer's instructions (Beecher Instruments, MD, USA). The tissue microarray (TMA) were sectioned and stained for the ER clones 1D5, SP1 and PharmDx.

Paraffin-embedded material for TMA was initially available for 500 of 564 patients in the study [15] (Figure 1). Due to intensive use of the TMA blocks for other studies, some of the core biopsies were missing and of the 500 cases in the TMA, 1D5 and SP1 were evaluable in 390 (69%) cases and PharmDx, which was stained later, in 361 (64%) cases. All three antibodies were available for scoring in 321 cases (Figure 1). Comparing patients with and without TMA scores for 1D5 and SP1, no significant differences appeared in histological type or grade, tumour size or lymph node status. However, patients with evaluable TMAs were slightly older than patients without TMA data, at 45 years (26–57 years) and 43 years (25–55 years), respectively.

Figure 1. Flowchart of patients.

Immunohistochemical analyses

After deparaffinising, sections stained with 1D5 were pretreated in a microwave oven in citrate buffer pH 6. The primary antibody 1D5 (DAKO, Glostrup, Denmark) diluted 1:35 was incubated for 25 minutes. The detection system was labelled streptavidin biotin-horseradish peroxidase (LSAB-HRP) used in a DAKO TechMate500+, and the reaction product was visualised with diaminobenzidine (DAB). For SP1, HIER was performed in microwave oven in Tris buffer pH 9. The primary antibody (RM-9101-S, clone SP1, Neomarkers, AH Diagnostics, Stockholm, Sweden) was diluted 1:100 and incubated for 25 minutes. The Envision system (K5001, DAKO, Glostrup, Denmark) was used in a DAKO TechMate500+, and the reaction product was visualised with DAB. PharmDx (ER/PR PharmDx kit code K4071, DAKO, Glostrup, Denmark) was applied with HIER in epitopal retrieval solution (citrate pH6) in an autoclave for 5 minutes at 125°C. Staining was performed in a DAKO Techmate 500/1000 automated staining instrument (BioTek Solutions, Winooski, VT, USA), and visualisation was based on dextran technology.

Cytosol-based method

ER was analysed in cytosol samples with LBA or EIA. ER-positive and -negative samples were classified as previously described [14].

Evaluation

ER status was assessed in the invasive component, at least 10 cancer cells being required for scoring. The staining for the antibodies 1D5, SP1 and PharmDx were categorised at seven levels with cut-offs of 0%, 1%, 10%, 25%, 50%, 75% and 90% positive cells. In the following only two cut-offs, 1% and 10%, were considered for ER status (negative vs. positive). For repeatability, 50% was also considered because this cut-off has been suggested as defining a patient group with a good response to endocrine therapy [8]. PharmDx was also scored as 0: negative, 1: > 0 and ≤ 1%, 2: > 1% and ≤ 10%, 3: > 10% and ≤ 33%, 4: > 33% and ≤ 66% or 5: > 66% and the intensity was scored as 0: negative, 1: weak, 2: moderate or 3: strong to produce the Allred score by adding the intensity score to the percentage score. One pathologist (DG) evaluated the staining results on two different occasions (repeatability). To compare different antibodies, the highest score from the two TMAs was used in each case. The pathologist was blinded to the clinical data, the follow-up and the results of the other scorings. For evaluation of the prognostic value of ER, cut-offs of 1% and 10% were used for 1D5, SP1 and PharmDx and an Allred score 3 where cases with a score ≥ 3 were considered positive.

Statistical analysis

The Mann-Whitney test was used for the two-group comparison of continuous variables, the χ²-test for the comparison of categorical variables, and the log-rank test for the comparison of survival curves. All tests were two-sided and p-values < 0.05 were considered significant. McNemar's test was used to test equality of paired proportions. The overall agreement, i.e. the proportion of samples with the same receptor status, was calculated, as were exact 95% confidence intervals (CIs) based on the binomial distribution. Furthermore, positive and negative agreements were used as summary measures. When one of the two assays compared can be regarded as a reference, the positive agreement was defined as the proportion of positive specimens that were also positive with the test assay. Negative agreement was defined similarly [12]. When the same assay was evaluated twice, to assess repeatability, symmetric definitions of positive and negative agreement with approximate 95% CIs were used, following Graham and Bull [16]. The upper limits of the approximate CIs exceeded 1.00 in a few cases but were then set to 1.00. In the analysis of RFS, the event was defined as local, regional, or distant recurrence or breast cancer death. Long-term follow-up is available, but only the first five years after diagnosis were used because the effect of ER is known to diminish with time. It should be emphasised that our study was not designed to have the power to study the prognosis of the discordant group in relation to the double-positive or -negative groups or the effect of adjuvant TAM vs. no TAM. The statistics packages Stata version 12 and SPSS version 20 were used.

Results

Repeatability of staining evaluation

A comparison of evaluations 1 and 2 for each antibody at different cut-offs is shown in Table I. The repeatability was excellent with overall agreement between 93% and 100% for 1D5, SP1 and PharmDx respectively at cut-offs 1%, 10%, and 50%. The positive agreement (≥ 93%) and negative agreement (≥ 92%) were also acceptable.

Table I. Repeatability for 1D5 (n = 390), SP1 (n = 390) and PharmDx (n = 361). The figures below each of the nine two-by-two tables for overall agreement are, from left to right, the ratios (numerator/denominator), the percentages and the exact binomial 95% confidence intervals. If the cell counts in these two-by-two tables are labelled a, b, c and d, from left to right beginning with the first row, negative agreement is defined as 2a/(2a+ b + c) and positive agreement as 2d/(2d+ b + c), measures that are symmetric in evaluation order. For the latter measures, approximate 95% confidence intervals based on the delta method are presented.

			1D5		SP1		PharmDx
					Evaluation 1
			−	+	−	+	−	+
Cut-off 1%	Evaluation 2	−	124	3	98	1	122	7
		+	1	262	2	289	7	225
Overall agreement			(124 + 262)/390	99 (97–99)	(98 + 289)/390	99 (98–100)	(122 + 225)/361	96 (94–98)
Negative agreement				98 (97–100)		98 (97–100)		95 (92–97)
Positive agreement				99 (99–100)		99 (99–100)		97 (95–99)
Cut-off 10%	Evaluation 2	−	133	2	109	1	141	7
		+	0	255	0	280	1	212
Overall agreement			(133 + 255)/390	99 (98–100)	(109 + 280)/390	100 (99–100)	(141 + 212)/361	98 (96–99)
Negative agreement				99 (98–100)		100 (99–100)		97 (95–99)
Positive agreement				100 (99–100)		100 (99–100)		98 (97–99)
Cut-off 50%	Evaluation 2	−	166	3	131	2	216	15
		+	25	196	4	253	1	129
Overall agreement			(166 + 196)/390	93 (90–95)	(131 + 253)/390	98 (97–99)	(216 + 129)/361	96 (93–98)
Negative agreement				92 (89–95)		98 (96–100)		96 (95–98)
Positive agreement				93 (91–96)		99 (98–100)		94 (91–97)

Agreement between antibodies

The proportion of ER-positive cases for SP1 was significantly higher than that for 1D5 at all cut-offs. At cut-offs of 1%, 10% and 50%, a total of 29, 24 and 47 patients were positive with SP1 but negative with 1D5, whereas only three, 0, and 0 patients showed the opposite pattern (Table II). As a consequence, the positive agreement was high above 99%, whereas the negative agreement was lower: 77%, 82%, and 74% at cut-offs of 1%, 10% and 50%, respectively. In the subgroup where ER was also analysed with PharmDx (n = 321), the overall agreement was similar at cut-offs of 1% and 10% for PharmDx compared with 1D5 and SP1 (Table III), but at a cut-off of 50% the overall agreement was higher between PharmDx and 1D5 than between PharmDx and SP1. The intensity of positive cases, stained with PharmDx, was distributed as follows: 107 (42%) patients with weak, 108 (43%) with moderate, and 39 (15%) with strong intensity. The concordance in ER status between the Allred score and the 10% cut-off for PharmDx was 95% (92–97%). The addition of intensity to the percentage of positive cells resulted in 19 cases changing from negative with PharmDx to positive with the Allred score, whereas no cases were positive with PharmDx and negative with the Allred score. At a cut-off of 1%, no patients had discordant ER status.

Table II. Agreement of evaluations of the highest scores for 1D5 and SP1 (n = 390). The figures below each of the three two-by-two tables are, from left to right, the ratios (numerator/denominator), the percentages and the exact binomial 95% confidence intervals. 1D5 is used as a reference when calculating negative and positive agreements.

		1D5
	SP1	−	+
Cut-off 1%	−	95	3
	+	29	263
Overall agreement		(95 + 263)/390	92 (89–94)
Negative agreement		95/124	77 (68–84)
Positive agreement		263/266	99 (97–99)
Cut-off 10%	−	109	0
	+	24	257
Overall agreement		(109 + 257)/390	94 (91–96)
Negative agreement		109/133	82 (74–88)
Positive agreement		257/257	100 (99–100)
Cut-off 50%	−	131	0
	+	47	212
Overall agreement		(131 + 212)/390	88 (84–91)
Negative agreement		131/178	74 (66–80)
Positive agreement		212/212	100 (98–100)

Table III. Agreement of evaluations of the highest scores for PharmDx, 1D5 and SP1 (n = 321). The figures below each of the six two-by-two tables are, from left to right, the ratios (numerator/denominator), the percentages and the exact binomial 95% confidence intervals. 1D5 and SP1, respectively, are used as reference categories when calculating negative and positive agreements.

			1D5		SP1
			−	+	−	+
Cut-off 1%	PharmDx	−	93	17	76	34
		+	10	201	4	207
Overall agreement			(93 + 201)/321	92 (88–94)	(76 + 207)/321	88 (84–91)
Negative agreement			93/103	90 (83–95)	76/80	95 (88–98)
Positive agreement			201/128	92 (88–95)	207/241	86 (81–90)
Cut-off 10%	PharmDx	−	100	21	86	35
		+	10	190	5	195
Overall agreement			(100 + 190)/321	90 (87–93)	(86 + 195)/321	88 (83–91)
Negative agreement			100/110	91 (84–96)	86/91	95 (88–98)
Positive agreement			190/211	90 (85–94)	195/230	85 (79–89)
Cut-off 50%	PharmDx	−	140	49	103	86
		+	5	127	3	129
Overall agreement			(140 + 127)/321	83 (79–87)	(103 + 129)/321	72 (67–77)
Negative agreement			140/145	97 (92–99)	103/106	97 (92–99)
Positive agreement			127/176	72 (65–79)	129/215	60 (53–67)

Prevalence of ER-positivity

At a cut-off of 1%, the prevalence of ER-positivity was significantly (p < 0.001) higher with SP1 (75%; 292/390) compared with 1D5 (68%; 266/390) and PharmDx (66%; 211/321). The corresponding figures for a cut-off of 10% were 72%, 66% and 62%, and for a cut-off of 50% were 67%, 55% and 41% (Tables II and III).

Recurrence-free survival

Breast cancer patients with tumours positive (cut-offs 1% and 10%) with both 1D5 and SP1 had a better five-year RFS compared with those negative with both antibodies. This was true for the entire patient cohort as well as for the TAM-treated sub-group (Figure 2a,b,d and e). For patients without adjuvant TAM the difference in RFS was not significant (Figure 2c and f). Discordant patients that were 1D5-negative, but SP1-positive, showed an intermediate RFS (Figure 2a–f). When 1D5 and SP1 were compared with PharmDx, similar results were obtained (data not shown). The staining intensity was also related to RFS (Figure 3a). When only the positive cases were included, staining intensity correlated weakly with RFS (p = 0.12). When the intensity was added to the percentage score, to obtain the Allred score, 19 cases became positive. These 19 patients had a prognosis intermediate between that of patients positive with Allred and PharmDx and that of patients negative with Allred and PharmDx (Figure 3b).

Figure 2. Recurrence-free survival, according to ER status assessed by 1D5 and SP1 at cut-offs of 1% (left panel) and 10% (right panel), for all patients (a and d), for TAM-treated patients (b and e) and for patients not treated with TAM (c and f).

Figure 3. Recurrence-free survival according to intensity of ER (a) and ER status assessed by PharmDx at a cut-off of 10% and Allred score (b), (n = 361).

Patient and tumour characteristics

The characteristics of the discordant group (1D5-negative/SP1-positive) were compared with those of the group positive (cut-offs 1% and 10%) with both 1D5 and SP1 and the group negative with both antibodies (Tables IV and V). The discordant patients were more often of histological grade 1 or 2 compared with the double-negative patients, and this difference was statistically significant (p < 0.01). At a cut-off of 1% the discordant group differed also significantly from the double-positive group (p < 0.01). The discordant patients were also significantly more often node-positive than were the double-negative patients (p = 0.01) at a cut-off of 10%. The double-positive patients were more often ER-positive with the cytosol-based methods compared with the double-negative patients (cut-off 1%: 84% vs. 8% and cut-off 10%: 85% vs. 11%). The discordant patients showed an intermediate level of ER-positivity based on the cytosol methods at both cut-offs (29% and 37%), significantly different from both the double-positive patients (p < 0.01) and the double-negative patients (p = 0.01). No differences appeared in age, tumour size, or histological type.

Table IV. Patient and tumour characteristics by group according to 1D5/SP1 results at a cut-off of 1%.

	1D5+/SP1 + n = 279	%	1D5−/SP1 + n = 42	%	1D5−/SP1 − n = 64	%	p −/+ vs +/+	p −/+ vs −/−
Age, years
≤ 40	47	18	4	14	28	29	0.46	0.39
41–45	82	31	13	45	34	36
46 – 50	101	38	8	28	24	25
> 50	33	13	4	14	9	9
Tumour size, mm
≤ 10	18	7	0	0	4	4	0.07	0.65
11– 20	97	37	8	28	20	21
21– 30	110	42	12	41	41	43
> 30	38	14	9	31	30	32
Tumour type
Ductal	225	86	25	89	80	84	0.05	0.67
Lobular	27	10	0	0	2	2
Other	9	3	3	11	13	14
Missing	2		1		0
Histological grade
1	41	16	1	3	2	2	< 0.01	< 0.01
2	143	55	10	34	8	9
3	77	30	18	62	77	89
Missing	2		0		8
Lymph node status
Negative	63	24	6	21	37	39	0.29	0.17
1–3 positive nodes	141	54	12	43	37	39
≥ 4 positive nodes	59	22	10	36	21	22
Missing	0		1		0
ER (cytosol method)
Positive	179	84	7	29	6	8	< 0.01	0.01
Negative	33	16	17	71	74	93
Missing	51		5		15

Excluded are three patients with 1D5+/SP12.

Table V. Patient and tumour characteristics by group according to 1D5/SP1 results at a cut-off of 10%.

	1D5+/SP1 + n = 257	%	1D5 − /SP1 + n = 24	%	1D5− /SP1− n = 109	%	p − /+ vs +/+	p − /+ vs −/−
Age, years
≤ 40	45	18	4	17	32	29	0.62	0.21
41–45	80	31	8	33	42	39
46–50	100	39	7	29	26	24
> 50	32	12	5	21	9	8
Tumour size, mm
≤ 10	17	7	1	4	4	4	0.82	0.35
11–20	94	37	9	38	23	21
21–30	109	42	9	38	46	42
> 30	37	14	5	21	36	33
Tumour type
Ductal	220	86	21	91	89	82	0.66	0.41
Lobular	26	10	1	4	4	4
Other	9	4	1	4	16	15
Missing	2		1		0
Histological grade
1	41	16	1	4	2	2	0.12	< 0.01
2	140	55	12	50	10	10
3	74	29	11	46	89	88
Missing	2		0		8
Lymph node status
Negative	61	24	2	9	45	41	0.10	0.01
1–3 positive nodes	138	54	12	52	41	38
≥ 4 positive nodes	58	23	9	39	23	21
Missing	0		1		0
ER (cytosol method)
Positive	175	85	7	37	10	11	< 0.01	0.01
Negative	32	15	12	63	82	89
Missing	50		5		17

Discussion

Three antibody/HIER combinations, 1D5 with HIER in citrate pH 6, SP1 in Tris pH 9, and PharmDx in citrate pH 6, were compared for the prevalence of ER-positivity in patients with clinically stage II breast cancer participating in a clinical trial comparing two years of adjuvant TAM versus no adjuvant systemic treatment. The overall pairwise agreement varied between 93% and 100%. Staining with SP1 yielded a higher percentage of ER-positivity than did 1D5 and PharmDx at cut-offs of both 1% and 10%. It should be noted that the overwhelming majority of discordant patients went from negative with 1D5 to positive with SP1 (n = 29) at a cut-off of 1% whereas only three patients went in the opposite direction. At a cut-off of 10%, all discordant patients (n = 24) went from negative with 1D5 to positive with SP1. Patients 1D5-negative SP1-positive showed an intermediate clinical outcome (RFS). The same pattern appeared when comparing PharmDx at a 10% cut-off with the Allred score, where the addition of intensity resulted in a discordant group with an intermediate RFS.

The ER/PR guidelines [9,12] underline the details of concordance studies when a new method of determination of ER status is being tested. To be considered acceptable, positive agreement must be ≥ 90% and negative agreement ≥ 95%; positive results are defined as ≥ 1% immunoreactive cells compared with a clinically validated assay. In the present study the differences between the antibodies were greater with negative agreements of only 77%, 82% and 74% comparing SP1 with 1D5 at a cut-off of 1%, 10% and 50%. The positive agreements between 1D5, SP1 and PharmDx were all > 90%. This appears to be antibody- and HIER method-related because the repeatabilities within the antibodies were acceptable.

Similarly to the results of the present study, which shows an increase of 7 percentage points with SP1 compared with 1D5 at a cut-off of 1%, the ER-positivity increased by 8 percentage points when SP1 was used instead of 1D5 with a cut-off of 1% in the study by Cheang et al. [13] who reported a large population-based study from British Columbia, Canada. That study included 1450 patients who had ER assessment by the dextran-coated charcoal (DCC) method. In fact, the material used for TMA to examine ER by IHC was from frozen stores of residual tissue in excess of the requirements of the DCC assay [17], i.e. those patients with enough tumour tissue to make frozen samples. Consequently, the patients had a more advanced disease stage than population-based patient material. In the analysis of recurrence-free survival and breast cancer-specific survival, the cases positive with SP1 and negative with 1D5 followed the positive cases. In the present study, the discordant group had an outcome intermediate between the double-positive and -negative groups, but no significant differences were found.

The cut-off for IHC-detected ER-status has not been determined in randomised studies but has evolved by comparison with the cytosol-based methods [7]; the correlation is about 0.85 [18]. Today, 1% is the generally accepted cut-off used for clinical decision making [19]. In the randomised Stockholm adjuvant tamoxifen trial, including postmenopausal patients (STO-3), only seven of 777 patients had ER expression between 1% and 10%. In the present study, nine of 390 patients were classified between 1% and 10% with 1D5, 11 of 390 with SP1 and 19 of 361 with PharmDx, respectively. In the STO-3 trial using a cut-off of 10%, the predictive value of ER status determined by IHC was comparable with that of the DCC method [6]. In a population-based registry in Sweden including more than 6800 patients, the percentage of ER-positivity at a cut-off of 10% was 86% in 2011 (pers. comm.). The more sensitive antibodies and more aggressive HIER methods may contribute to a higher prevalence of ER-positivity [20].

In accordance with the results of the present study discordant results of ER status has also been reported in studies comparing different methods to assess ER status. One study compared ER status determined by cytosol-based assays with IHC on frozen sections using the Abbot antibody [21]. The study included 2364 patients diagnosed between 1991 and 1993 and registered in the Danish Breast Cancer Cooperative Group (DBCG) register. Of the patients 68% were positive with the IHC method and 80% were ER-positive with the cytosol-based assay. The study showed superior sensitivity of the cytosol-based assay and overall survival of the discordant patients was comparable to that of the double-positive patients.

Another study used central re-assay with IHC with the antibody 6F11 in Tris EGTA pH 9 and compared results with previous results of cytosol-based and IHC assays [22]. The study included 1530 patients diagnosed between 1985 and 2001 and registered in the DBCG database. Overall 7% more patients were ER-positive at re-assay. The authors observed almost exactly the protective effect of endocrine therapy among tamoxifen-treated ER-positive women whose tumours expressed the ER on re-assay, compared with those ER-negative on re-assay.

In the present study, the addition of intensity does not add prognostic information. It should also be mentioned that no method to determine the exact amount of protein by weight in IHC sections exists and the influence of fixation and HIER on protein loss is largely unknown [23], which makes the estimation of intensity unreliable.

Another study compared ER status assessed by 1D5 and SP1 on whole sections from 508 breast carcinomas including in situ cancer and metastases, finding only two cases positive for SP1 and negative for 1D5 [24]. In that study, both antibodies had HIER in citrate pH 6; the dilution was 1:100 for 1D5 and 1:200 for SP1. The authors suggest that because SP1 has an eight-fold higher affinity for its epitope than does 1D5, it may simply detect more positive cells than does 1D5 when protein levels are extremely low. The low discordance rate of 2/508 might also be attributed to the fact that SP1 was examined in citrate pH 6 instead of Tris pH 9.

Rhodes et al. [20] examined the influence of HIER buffer and heating time, and showed for the ER-antibodies 6F11 and SP1 that HIER in Tris EDTA, pH 9 and a longer heating time resulted in higher Allred scores than for HIER in citrate pH 6 and shorter heating times. In the present study, Tris pH 9 was used with SP1, and this more sensitive IHC method and the higher affinity for its epitope resulted in an increase of SP1-positive cases of 7 percentage points at a cut-off of 1%, in line with the 8 percentage points increase reported by Cheang et al. [13]. It seems the antibody and HIER method is more important in the increasing prevalence of ER-positive patients than intra-rater variability (repeatability). This is reflected in the present study, where one pathologist showed good repeatability when scoring the antibodies twice, and in a recent study of patients referred for a second opinion where the ER status was changed in only two of 405 patients [25].

Using 1D5 in EDTA buffer at pH 8, Lau et al. [26] reported 18% of 55 lung carcinomas and 72% of 50 breast carcinoma to be ER-positive using a cut-off of > 0. In another study, Gomez-Fernandez et al. [27] also examined lung carcinomas. HIER was citrate pH 6 and the results showed 27% of 92 cases positive for SP1, 14% positive for 6F11 and 8% positive for 1D5. The higher affinity for the epitope of SP1 also causes more lung adenocarcinomas to be ER-positive, information very important to the diagnostic pathologist because ER status is involved in the distinction between primary lung adenocarcinomas and breast cancer metastases.

In conclusion, the present study shows that the choice of antibody and HIER method influences the prevalence of ER-positivity. We suggest that this be taken into consideration when choosing a cut-off for clinical decision making.

Acknowledgements

We are indebted to participating departments of the South and South-East Swedish Breast Cancer Groups for providing samples and clinical follow-up. Thanks to Kristina Anderson for performing staining with 1D5 and SP1 and to Kristina Lövgren for performing the staining with PharmDx. Thanks to Dako for providing the antibody ER PharmDx.

Declaration of interest: The authors report no conflicts of interest. The authors alone are responsible for the content and writing of the paper.

The study was supported by funding from the Swedish Cancer Society, the Swedish Research Council, the Gunnar B Nilsson Cancer Foundation, the Mrs. Berta Kamprad Foundation, the Anna and Edwin Bergers foundation, Skåne University Hospital Research Foundation, Skåne County Council's Research and Development Foundation, and Governmental Funding of Clinical Research within the National Health Service.