Critical determination of the frequency of c-erbB-2 amplification in breast cancer.

Tissues from 323 methacarn-fixed and paraffin-embedded breast cancers were assessed for c-erbB-2 gene amplification by differential polymerase chain reaction (dPCR). The sensitivity of dPCR was ascertained using cell lines with c-erbB-2 amplification, and the relationship between dPCR ratio value and gene copy number was established. In clinical material the technique was not affected by the DNA contribution of normal tissue elements or by cancer DNA ploidy change. c-erbB-2 gene amplification was detected in 55% of invasive cancers and in 66% of in situ cancers. c-erbB-2 protein overexpression in breast cancer cells, as determined by specific immunohistochemistry, was only detected in 11% of invasive cancers and 43% of in situ cancers. Comparisons show that a substantial number of cancers with c-erbB-2 amplification lack detectable protein overexpression. This illustrates the complex nature of c-erbB-2 gene disregulation in cancer and suggests that multiple combinations of biological events and consequences are possible.

number was establshed. In clnical material the tehnue was not affected by the DNA contnbution of normal tissue elments or by canmr DNA ploidy change. c-erbB-2 gne amwti was detected in 55% of invasive cams and in 66% of ix situ cancse c-erbB-2 protein ovexession in breast cancer ceis, as determined by sPecif immunohistochenistry, was only detected in 11% of invasive cancrs and 43% of in situ cancem Comparisons show that a substantal number of cancrs with c-erbB2 ationlack detectable protein overexesso. This illustrates the complx nature of c-erbB-2 gme di la in cancer and sugests that multiple combinatons of biological events and consequences are possible.
Disregulation of the proto-oncogene c-erbB-2 (also known as HER-2/neu) has been implicated in the aetiology of breast cancer. Since the pubication of a study linking c-erbB-2 to poor prognosis in breast cancer patients (Slamon et al., 1987) there have been many studies examining c-erbB-2 gene amplification, mRNA production and protein overexpression. Recent reviews have collated the results from over 50 studies and found a geneal agreement between them on the frequency of c-erbB-2 disegulation in terms of gene amplification and protein overexpresson, as measured by Southern blotting and immunohistocmistry spectively (Perren, 1991;Singleton & Strickler, 1992). However, there are major differences in the association of c-erbB-2 disregulation with histopathological features and with prognosis, makcing its involvement in cancer development and progression difcult to determine. It is not clear whether differences in results between studies have been the result of variations in sample selection, experimental technique or genuine biologically relevant disparity between populations.
Of the techniques for measuring gene amplification, Southern or dot blotting suffers from the disadvantages that microgram quantities of DNA are required for analysis and tissue morphology is destroyed in the extration process. Recent advances in polymerase chain reaction (PCR) technology have made possible the analysis of minute quantities of DNA, with semiquantitative differential esimations (dPCR) demonstrating increased gene copy number in cell lines (Frye et al., 1989). The present work explores the sensitivity of dPCR in detecting an increased gene copy number in a large series of clnical cancers by extending the application of this technique to paraffin-embedded tissues, with a view to evaluating the relationship between c-erbB-2 gene amplification and expression.

Study set
The study tissues (336 cases) were colleted from primary operable (clinical stage I and II) breast cacers at routine operations, which included mastectomy and excisional biopsy for both palpable and non-palpable lesions. Sampks were restricted to the age group 50-65 and were collecte over the period of January 1988 to May 1990. They were fixed in methacar (6:3:1 methanol-chloroform-acetic acid) overnight at 4C, processed according to routine methods and embedded in praffin-Control tissue (43 cases) was obtained from breast tissue distant to the lesion site or from noncancer-bearing breasts. Pathological characterisation was taken from overall evaluation of material used for routine diagsis, and inchuded an evaluation of a 4 pzm section immediately adjacent to wctions taken for dPCR (see below). This section confirmed the nature of the tissue used in the PCR reaction, and in addition the cellularity of each specimen was assessed subjectively for the proportion of the cancer cellular content and desgnated as either 1 = more than 75%, 2 = 25-75% or 3 =<25%. In some cases samples of the lesion were taken and stored frozen in liquid nitrogen for RNA analysis.
Flow cytometric analysis Paraffin-embedded tumours were processed for DNA flow cytometry according to the method of Hedley et al. (1983). Briefly, two 50 pm sections were dewaxed using two changes of xykne and rehydrated. The tissue was incubated for 30min at 3TC in 0.5% pepsin (Sigma) in 0.9% saline adjusted to pH 1.5 with 2 N hydrochloric acid. The isolated nuclei were counted and analysed usng an EPICS C flow cytometer (Coulter Electronics, Hileah, FL, USA), after staining with 0.1% propidium iodide containing 0.004% RNAse. Ten thousand nucli were counted at 480 nm excitation and the coefficient of variation calclated using STAT-PACK software (Coulter Electonics). Ploidy was as either diploid (DNA index, DI, between 0.9 and 1.10) or aneuploid (DI > 1.10 and <1.90 or >2.10). Tetraploids were classified as DI between 1.90 and 2.10 with more than 20% of the cells apparently in G2 plus M phase of the cell cycle. For inclusion the coefficient of variation for the peak value had to be less than 8%.
Cell lines and culture conditions Human breast cancer cell lines known to have an amplification of c-erbB-2 were used to calibrate the relationship between differential PCR ratio values and gene copy number. T'he epitheil cell line 21MT2 was obtained from R. Sager (Dana-Farber Cancer Institute, Boston, MA, USA) and contains a 40-fold increase of the c-erbB-2 gene (Band et al., 1989). The cell line UISO BCA1 was obtained from R.R. Mehta (University of Illinois, Chicago, IL, USA) and contains a 10-fold increase in the c-erbB-2 gene (Sasi et al., 1991). c-erbB-2 AMPLIFICATION IN BREAST CANCER 435 Each cell line was grown at 37C in air with 5% carbon dioxide added. 21MT2 was cultured in alpha minimum essenial medium (MEM) (Gibco) containing 10% fetal calf serum, 2 mM L-glutamie, I mM sodium pyruvate, 0.1 mM non-essential amino acids, 1 ig ml-' ins 2.8 pM hydrocortisone 12.5 ng ml-' epidrmal growth factor and 10 mM HEPES. UISO BCAI was cultured in Glasgow's minimum essential medium (GMEM) (Gibco), 10% fetal calf serum and 2 mM L-glutmine. For cahbration experiments, DNA was prepared from each cdl line (Sambrook et al, 1989) and was mixed with control DNA derived from normal placenta (p258, one c-erb-B-2 gene copy), in proportions which gave a series of known c-erbB-2 copy numbers. The 21MT2 DNA was diluted to give c-erbB-2 copy numbers of 32, 24, 16, and 8, and UISO BCA1 was diuted to give copy numbers of 9, 6, 5, 4 and 3.

Immunohistochemistry
Overexpression of c-erbB-2 was ascertained using the rabbit polyclonal antibody, 21N, to the c-erbB-2 protein (Gullick et al., 1987). Four micron sections of fixed tissue were dried at 56-C then stained in a thre-stag peroxidase-antiperoxidase technique (Sternberger, 1986). The primary antibody, 21N, was used at a concentration of 3.3igml'in 0.1M Trisbuffered sahne (pH 7.6) containing 5% normal swine serum. Each section was incubated at room temperature for 90 min. Endogenous peroxidase was blocked by exposure to 1% hydrogen peroxide in nmthanol for 30 min before staining.
Overexpression of c-erbB-2 was defined as the presence of brown Staining of surface membrane of cancer cells. To score positive, more than 10% of cells had to show moderate to strong staining. Controls a known positive cas and a negative control employing a preincubation of the antibody with its corresponding peptide (1 mg ml-').
mRNA Meger RNA was extracted from frozen tumour samples and analysed by Northern blot (Thompson et al., 1990).
Twenty micrograms of total RNA was denatured with formamide and formldhyde at 55-C for 20 min and RNA species separated by edetophoresis on a 1.1% agarose gel. The RNA was tansferred to a nylon filter (hybond-N, Amersham, UK) by capillary action usng 10 x SSC and covaently fixed to the membrane using a UV tansil-lIuminator. To detect c-erbB-2 mRNA the filters were hybridised with 1107, a 1.7 kb fragment of v-erbB-2 (Semba et al., 1985), according to the method of Church and Gilbert (1984), washed to remove non-specifically attached probe and exposed to preflashed Kodak XAR film at -70C. Filters were stripped and reprobed with a-actin (Minty et al., 1981) as an internal control for loading. The extent of hybridisation of radiolabelled probe to the mRNA specis was determined using densitometry and expressed with respect to hybridisation to the actin probe.
Primes and the polymerase chain reaction Primers used in the differential PCR are listed in Table I. Tlhey were DNA sequences specific for interferon gamma (IFN-y150), c-erbB-2 and interferon beta (IFN-P). The singlecopy reference sequence was the 150 bp sequence from the IFN-y gene. For dPCR four 10.pm sections of fixed paraffinembedded tissue were added to 100 id of lysis buffer (50 mM Tris-HCL pH 8.4, 1 mM EDTA, 0.5% Tween 20) and boiled for 8 min (Hubbard & Anderson, 1993). Differential PCR was performed on a Techne PHC3 thermal cycler incorporating 5 tl of prepred lysed paraffin section or 200 ng of extracted DNA, 0.25 gtM each primer (except for primers for IFN-P, 0.125 gM), 200mM dNTPs (Pharmada), x 1 Taq polymerase buffer (Northumbria Biotechnology Limited, NBL), 1 unit of Taq polymerase (NBL) and 3 tCi of [2PJCTP (New England Nuclear). Cycling parameters were one cycle of 94-C for 5 mi 50C for 1 mi 70-C for 1 mi, followed by 30 cycles of 94C for 1 mi 50-C for 1 , 70-C for 1 mi, and one cycle of 94-C for 1 min, 50C for 1 mi, 70 C for 5 min. Duplicate PCR products were separated by size on 2% agarose gels, and stained with ethidium bromide.
Visible bands were excised, finely chopped and added to 5 ml of Optiphase-safe scinillation fluid and radioactivity present assessd as counts per minute (c.p.m.) on a Beckman scintillation counter. A correction factor was applied to compensate for the differences in numbers of CTP bases between reference and test gene. Al specimens were assed in duplicate experiments.
The results from dPCR are expressed as ratio values and were callated by averaging the c.p.m. from duplicate gel tracks and subtracting the average experimental blank. For c-erbB-2 a correction factor of 1.25 was applied to compensate for differences in dCTP content between IFN-y150 (69 C bases) and c-erbB-2 (55 C bases). To ascertain the relative quantity of c-erbB-2 gene with respect to the reference gene, the corrected average c.p.m. for c-erbB-2 was divided by the average c.p.m. for IFN-'yl5O, giving in each case a result expressed as a ratio value. Similar correction factors were calculated and applied to dPCR involving amplification of IFN-P.

Reset
Validation ofd&fferential PCR method Calibration of ratio values Differential PCR was assesed for sensitivity and reproducibility in detemining gene copy number. As defined here, 'one gene copy' corresponds to the normal dipkoid content of one cell. The results of differential PCR on the various DNA solutions of p258 and cell lines 21MT2 and UISO BCAI with known c-erbB-2 gene copy number, using primers for IFN-150 and c-erbB-2, are shown in Figure 1. A comparison of the known copy numbers in each cell line sample and differential PCR ratio values showed that increasing gene copy number resulted in increasing ratio values, mean valus for I and 40 gene copy numbers were 1.66 and 11.46 repcively. While comparison of the ratio values obtained for given copy numbers shows some variation between experiments, there was a consistent irement in this value within each experint. Sampks with large amplifiations (>32) showed increased variation between duplicate tests. For the purposes of standardisation it was consided best to work on a mean value for these Table I  Note that each ratio value is a derived value a equate with but is directly proportional to number.
Factors affecting dPCR ratio values Application nique in a series of paraffin-embedded specime stringent controls. Confirmation that IFN-y was ] single-copy gene was obtained in 57 cancer and specimens by performing differential PCR with both IFN-yl50 and IFN-P. The ranges of r detected were similar for cancers (0.81-1.9) and sues (0.4-1.7), suggesting that IFN-y was presen a single-copy gene.
Satisfactory analysis of DNA ploidy by floA was obtained from 240 cancers. In 117 the phe diploid, and 123 were aneuploid or tetraploid. Th of amplification of c-erbB-2 in specimens assess cytometry was found to be highest in cancers diploid (60%), with lower percentages of aneu; and tetraploid (42%) cancers being amplil differences were not significant.
A third potentially confounding factor was th effect of normal cells present within the cancer haps reducing the detection frequency of amplifi proportion of amplified and non-amplified cases cancer ranked according to section cancer cellular in Figure 2. Amplification was found in each of including those specimens in which cancer cells less than 25% of total cellularity. c-erbB-2 amplification and overexpression in breast Gene anplification determined by differential PCR 323 breast cancer specimens and 43 controls wer c-erbB-2 amplification using primers for c-erbB-y150. Figure 3 shows representative PCR produc from fixed tissue specimens of three different canc DNA extracted by routine phenol/choroform from fresh tissue preserved at -70°C from oi cancers. Differential increase of c-erbB-2 product amplification is illustrated, with corresponding rat 1.4, 2.1 and 3.6 for the fixed cancer tissue, 3.7 f specimen 3 and 1.2 for control DNA. of c-erbB-2 amplification in 277 breast cancers. Cancer cellularity was assessed visually as > 75% cancer cells = 1, 25-75% cancer cells = 2, <25% = 3. 0, Specimens with a dPCR ratio value less r and differ-than 2; *, Specimens considered to be amplified (dPCR ratio red by dilu- fied. These ie dilutional tissue, per-The ratio values obtained using primers for c-erbB-2 and ication. The IFN-yl50 from both normal and cancer tissues are shown in ,of invasive Figure 4. The ratio range for 43 normal tissues fell conity is shown sistently between 0.6 and 1.9 (mean 1.2, s.d. 0.36), and the groups, therefore values of 2 or above were considered to signify gene constituted amplification. This value corresponds to approximately five gene copies (see Figure 1), and indicates that dPCR, in its present form, is unsuitable for exact specification of those cancers cases with low copy number (<5  Protein overexpression assessed by immunohistochemistry Immunohistochemistry for c-erbB-2 overexpression was performed on 336 breast cancer specimens. Overexpression of c-erbB-2 was detected in 23 of 54 (43%) of in situ carcinomas and in 31 of 282 (11%) invasive carcinomas. In cases in which in situ and invasive forms of cancer were present on the same slide, no detectable differences in the staining pattern between them was observed. Staining was concentrated on epithelial cell membranes and stained cells were present evenly throughout the cancer, except in one cancer in which focal staining of cancer cells was observed.
Overexpression was not observed in normal epithelial or stromal cells.
Comparative evaluation of protein overexpression and gene amplification A case comparison of gene amplification determined by dPCR with protein expression determined by immunohistochemistry is shown in Table II. Thirty-nine of 49 immunopositive cases (80%) had gene amplification (with ratio values ranging from 2.0 to 19.2). There were ten cases in which differential PCR did not detect gene amplification in the presence of protein overexpression. However 146 of 274 immunonegative cases (53%) had PCR-detectable amplification of the c-erbB-2 gene, and this included 43 cases with ratio values >3, indicating high copy number. The range of differential PCR values was similar between the immunopositive and immunonegative groups ( Figure 5) and applied to both in situ and invasive cancers. Of the 13 samples assessed by immunohistochemistry but not available for PCR, five were immunopositive. mRNA measurement Specific messenger RNA was measured in 26 breast cancer cases, and increased levels of c-erbB-2 mRNA corresponding to densitometry values four times control or greater were found in 11 cases (42%). The correlation between c-erbB-2 mRNA levels and gene amplification and overexpression is shown in Table III immunohistochemistry contained elevated levels of c-erbB-2 mRNA. Furthermore, 4 of 19 cases negative for immunohistochemistry also had elevated levels of mRNA; gene amplification determined by dPCR was present in two of these cases.

Discussion
This study with fixed paraffin-embedded tissue has demonstrated that dPCR is a highly sensitive technique for the detection of gene amplification and is also sufficiently robust to be applied to tumours of differing cellularity and DNA ploidy. For invasive cancers the frequency of gene amplification (55%) was considerably higher than anticipated from reports of conventional methods based on Southern or dot blotting techniques. In ten major studies of breast cancer, each assessing 100 or more cancer cases, the frequency of amplification varied between 17% and 23% (see review by Singleton & Strickler, 1992). Because of the size of the disparity some initial comment on comparability of methods is appropriate.
Study of gene amplification is complicated by terminology for an increased gene number, which may be expressed as either a fold difference, increased copy number or both; fold difference is equated with copy number in some reports (Ali et al., 1988;Garcia et al., 1989). We have assumed that the fold differences ascertained for the cell lines used in calibration of the dPCR are valid reflections of gene copy number, and have therefore expressed the altered dPCR ratio values as increased gene copy number. Owing to the arbitrary cutoff point for 'amplification' outwith the range observed in normals, dPCR would appear to lack the specificity to identify low copy number. Caution must be applied when ratio values are extrapolated to gene copy number in clinical cases. Experimental variation and approximations inherent to DNA analysis, including Southern or dot blotting techniques, may affect the precise relationship between classifications. Yet studies using Southern or dot blotting claim to detect increases as low as 2-fold without quoting the full range of values observed, the experimental variation in duplicate tests  The overall frequency of amplification is 57% and overexpression 150o. All cancers with protein overexpression show elevated levels of mRNA. Four of 19 cases which were negative for protein overexpression have increased mRNA. Messenger RNA level appears to be independent of gene amplification.
.nA vF or recorded cancer cellularity differences. It is of interest that a large proportion of amplified cases show a low increase in gene copy number by all techniques; for eaample, 44% have 2-5 copies on Southern blotting (Borg et al., 1990), while in this study 61% have ratio values of 2-3. There remains some uncertainty about the most appropriate cut-off point on which to base an amplified finding, but for the purposes of this evaluation a ratio value of 2 was chosen, as this was always above the values obtained for control samples. Raising the cut-off point to a ratio value of 2.5 would reduce the numbers amplified to levels equivalent to those previously reported. However, differences in amplifiation frequency depending on technique have also been observed in studies of the ovary. dPCR detected c-erbB-2 amplifiction in 40% of cancers (Hruza et al., 1993); in contrast, pevious studies by Southern blotting detected amplification in 1-26% of ovarian cancers (Slamon et al., 1989;Zang et al., 1989;Imyanitov et al., 1992). This suggsts that there may be differences in sensitivity between these techniques. The possibility of artefactual elevation of dPCR ratio values in fixed tissue extracts was examined by comparing them with samples of DNA from the corresponding fresh tissue in a subset of cases, but we found no evidence for this (data not shown). There is also the issue of selection bias towards larger size of cancer where there is a requirment to submit tissue for extraction in DNA analysis. This does not apply to dPCR studies which, as in the present series, can be applid in a consecutive manner.
A higher degree of sensitivity than in the present study was claimed in a previous investigation of c-erbB-2 amplification using dPCR (Frye et al., 1989). One extra copy (2-fold increase) was detectable, but that study used high-quality, homogeneous DNA derived from cell lnes in a single experiment. Further developments of the technique on clinical material classified amplfication in terms of fold differences, the most sensitive level detecting a 2to 4-fold increase in c-erbB-2 product (Liu et al., 1992;Neubauer et al., 1992).
Those studies used a complex algorithm of experimetal exclusions involing four different dPCR reactions resulting in a selce population of cancers, and detected c-erbB-2 amplification in 48% of in situ canrs and in 21 % of invasive cancers (Liu et al., 1992). Details of interexperimental variation, ranges of dPCR ratio values and criteria for exclusion at each step of the algorithm were not stated. This makles direct comparison of amplation frequencies with the current study difficult. In addition Liu et al. (1992) restricted their series to stage H node-negative disease, whereas the present series was a consecutive group of operable cancer including both node-positive and nodenegative cases. However, despite the probklms of comparability, we consider that the technique as currently applied has major potential to give a valid but different perspetive of gene disregulation reklvant to study of the development and progression of ancer.
Detection of overexpression of c-erbB-2 by immunohistochemistry is subject to considerable variation between studies (Singleton & Strickler, 1993) in part because of the different primary antibodies, fixation methods, study set composition and criteria for assessing positive staining. The dilution used in this study of antibody 21N has been calibrated as detecting around 12 or more copies of c-erbB-2 (Gusterson et al., 1988), therefore cases with an amplification of between five and 12 copies may appear to be immunonegative. Evidence from the present mRNA studies supports this lmitation to detecting expression as 21% of our immunonegative cases tested had increased mRNA levels. That changes in methods can affect the frequency of detection is evident from a recent report by Poller et al. (1992), in which modification of fixation and immunohistochemical techniques increased the proportion of invasive cancers with c-erbB-2 overexpression to 39.7% from 15% (Lovekin et al., 1991). As in other studie, we found good correlation between overexpression and amplification: 80% of immunopositive cancers had detectable gene ampliication. However, the ranges of gene copy values found by dPCR in immunopostive and immuno-negative ancers of the current study set indicate that for both invasive and in situ cancers amplfication does not necessarily mean an equivalent overexpression, and some cases with strong immunostaining showed normal or modest icreases in gene copy numbers. This sugests that factors which cause overexpression of c-erbB-2 in the absence of gene amplication may also play a role when gene amplification is present.
The disparity in frequency of c-erbB-2 gene activation between in situ (around 44%) and invasive cancer (around 22%) noted in previous studies (see review by Singleton & Stickler, 1992) is considerably diminished in the present analysis, but the impliations for reklance in cancer progression are uncertain. An evaluation to test a hypothesis of cancer natural history in the breast by Alred et al. (1992) commented on c-erbB-2 overexpression in slce groups of 45 hyperplastic and dysplastic lesions as well as 708 in situ and invasive cancers. They concluded that abnormal activation of the gene was likely to be a sigiint but not the sole initiating factor for many cancers. The limitations of simple immunohistochemistry as a measure of disregulated gene activity have been recognised (Anderson, 1992;Wynford-Thomas, 1992). Improved sensitivity of detecting abnormal gene activity through fluorescent (or other methods of) in situ hybridisation (Kallionemi et al., 1992;Smith et al., 1993) is hlkely to reveal considerably more about the heterogeneity and degree of gene disregulation within cell populations. The potential to explore mcnisms of gene control by further analysis of material slcted according to results of dPCR, Northern analysis and/or immuohistochemistry is however apparent from the results reported here.
The present tudies show that amplification of c-erbB-2 is a frequent event in breast cancer and that the relationship between gene amplification and overexpression may be complex. Although the insensitivity of current immunohistochemistry in detecting small inreases in protein is a factor compliating interpretation, it is neverteless likely that each part of the repication/transcription/translation process can be disregulated. Thus combininations of such events could account for the distribution of cases among the categories of Table II1. Indeed, it appears that the fiequency of these various disorders of gene number and expression is not equivalent. A small percentage of cases overexpress c-erbB-2 in the absence of amplfication, while a larger number fail to show a detectable overexpression of the gene in the presence of amplification even though a small number of this group also have increased mRNA levels. Factors acting as promoters or suppressors of gene function may directly affect transcription regardless of the amplification status. Further direct evidence of factors affecting transcription comes from studies of breast cancer cell lines in which c-erbB-2 protein can be down-regulated by oestrogen complexed with its receptor (Russell and Hung, 1992). Inreased c-erbB-2 mRNA levels resulting from elevated amounts of a transcription factor have also been observed in cancer cell ies which have no detectable gene ampltion (Hollywood & Hurst, 1993). Other explanations of disorder include physical damage to the gene, mutation or the absence of promoters. This variety of biological events and consequences suggest that a more realistic model to evaluate c-erbB-2 disregulation in breast cancer must encompass a greater number of circumstances and consider the interaction of other biological processes. The potential to study these in subgroups of suitably characterised breast cancer cases is apparent. This study was supported by the Cancer Research Campaign, London. We would like to thank the staff of the Longmore Breast Unit Tbeatre for their help in colecting specimens, Mr D. Bishop for assistanc rwith immunobistochemistry, Professor WJ. Gulick for the gift of antibody 21N and Professor A.H. Wylie for his constructive discussions.