Zimling, Zarah Glad MD*; Sørensen, Jens Benn MD, DmSc†; Gerds, Thomas Alexander PhD‡; Bech, Cecilia MD†; Andersen, Claus Bøgelund MD, DMSc*; Santoni-Rugiu, Eric MD, PhD*
Malignant pleural mesothelioma (MPM) arises from the mesothelium lining the pleural cavity, and development of this cancer is often associated with previous asbestos exposure. Because of the long tumor latency, the incidence of MPM is increasing, despite the ban issued on asbestos use 3 decades ago.1 MPM has a poor prognosis in advanced disease, and only a small minority of patients is diagnosed at an early stage, when curative multimodality treatment is possible. Inoperable patients are usually offered a two-drug platinum-based chemotherapy regimen.2 Platinum compounds are used in the treatment of a large variety of cancers, but their efficacy is often limited by the intrinsic or acquired resistance of the cancer cells toward their mechanism of action.3,4
The last few years have witnessed an increased focus on markers of resistance, which can be used to predict treatment efficacy and thereby guide treatment decisions. Cisplatin and carboplatin work by binding to the DNA forming adducts that lead to intra- or interstrand cross-links. The formation of these DNA cross-links inhibits the cell from replicating and drives it toward apoptosis. This proapoptotic signal can be counteracted by the cells' intrinsic ability to recognize and repair the DNA damage. Nucleotide excision repair is a highly conserved pathway that maintains DNA integrity by removing helix-distorting cross-links. This pathway seems to be a key element in mediating resistance toward platinum compounds. There are three important steps in this pathway. First, the DNA damage is recognized then excised, and finally, the excised area is resynthesized. Excision repair cross-complementation group 1 enzyme (ERCC1), a protein encoded by a gene located on chromosome 19q plays a rate-limiting step in this process by forming a complex with xeroderma pigmentosum complementation group F that excises the damaged DNA.3–5
An inverse relationship between ERCC1 mRNA levels and response to cisplatin-based therapy has been shown in several retrospective studies on tumor samples from clinical trials with ovarian cancer,6,7 colorectal cancer,8,9 and non-small cell lung cancer (NSCLC).10 Olaussen et al.11 subsequently published a retrospective study on the immunohistochemical determination of ERCC1 status in a large randomized trial, among completely resected NSCLC patients, randomized to receive cisplatin-based adjuvant chemotherapy or not. They reported an increased therapeutic benefit for patients with low ERCC1 expression, and also observed high ERCC1 expression to be a marker of good prognosis in the patient group not receiving adjuvant cisplatin, possibly due to a more efficient DNA repair and consequently fewer genetic abnormalities during tumor progression. The same methodology has more recently been used by Vilmar et al.12 in a retrospective study to test the association of ERCC1 expression with response to cisplatin therapy in the palliative setting in advanced NSCLC. Again, a better treatment response was seen in the ERCC1-negative tumor population.
Two studies have recently addressed the possible predictive and prognostic role of ERCC1 in MPM. In an observational study by Righi et al.,13 immunohistochemistry (IHC)was used to detect ERCC1 in a cohort of 45 MPMs treated with different platinum-based therapies (cisplatin + pemetrexed or carboplatin + pemetrexed in different regimens). In this series, there was no association between ERCC1 status and treatment response, but the authors did find high ERCC1 levels to be associated with a better prognosis regardless of the chemotherapy regimen used. More recently, Zucali et al.14 also used IHC to detect ERCC1 in a retrospective cohort of 67 MPMs treated with a combination of pemetrexed and carboplatin. Moreover, these authors found no association between ERCC1 protein status and clinical outcome in terms of disease control, progression-free survival (PFS) and overall survival (OS).
The aim of our study was to investigate the potential role of ERCC1 protein status as a predictive marker for efficacy of cisplatin-based chemotherapy in MPM. For this purpose, we used immunohistochemical analysis to detect ERCC1 protein expression on diagnostic tissue samples, from inoperable patients, treated uniformly with a combination of cisplatin and vinorelbine, in a previously reported phase II clinical trial.15
PATIENTS AND METHODS
The study population consisted of 54 consecutive inoperable patients enrolled into a phase II trial conducted at our institution between 2003 and 2006.15 Eligibility criteria were as follows: Histologically verified MPM judged to be inoperable because of age >70 years, anatomical extension of disease (N2/N3 nodal status), nonepitheloid histology, or physiological inoperability (poor cardiopulmonary function, etc.). Patients were required to have measurable disease; Eastern Cooperative Oncology Group performance status 0–2; an estimated survival expectancy ≥3 months; age ≥18 years and had given informed written consent. Treatment was vinorelbine 25 mg m−2 intravenously weekly and cisplatin 100 mg m−2 intravenously every 4 weeks. Staging was done according to the system defined by the International Mesothelioma Interest Group.16
Patients were assessed with spiral computed tomographic scans at baseline before every other treatment cycle (every 8 weeks) and every 2 months after treatment completion. Treatment response was assessed by modified RECIST criteria.17 Change in disease was assessed by measuring tumor thickness perpendicular to the chest wall or mediastinum in up to three involved areas at least 2 cm apart. A reduction of at least 30% or an increase of 20% on two occasions 4 weeks apart was defined as a partial response (PR) or progressive disease (PD), respectively. A complete response (CR) was defined as complete absence of signs and symptoms of disease. In case of patients having not evaluable disease, progression (PD) was defined as the appearance of a new disease parameter. Response rate (RR) was defined as the proportion of patients, having evaluable disease, who achieved a CR or PR. The primary endpoint was PFS, defined as time from onset of treatment until documented progression or death from any cause, because this is an established marker of treatment efficacy in MPM supported by the European Organisation for Research and Treatment of Cancer.18 For patients not evaluable by modified RECIST criteria, PFS was calculated from onset of treatment until the appearance of a new disease parameter or death. OS was the time from initiation of treatment until death from any cause. For patients without progression at the time of analysis, OS and PFS were censored at the date of the last follow-up.
The formalin-fixed paraffin-embedded (FFPE) bioptic tumor specimens from the enrolled patients were obtained from Departments of Pathology throughout Denmark. Hematoxylin and eosin-stained sections from each specimen were reviewed by three independent observers (Z.G.Z., E.S.-R., and C.B.A.), to confirm the diagnosis and evaluate the suitability of the samples for the immunohistochemical analysis.
Thin (2 μ) sections were cut from each biopsy and mounted on coated glass slides. Sections were deparaffinized, and heat-induced antigen retrieval was performed in DAKO-PT-link module using target retrieval solution pH9 for 20 minutes at 97°C (DAKO, Glostrup, Denmark). Endogenous peroxidase was blocked, and sections were incubated at room temperature for 20 minutes with mouse monoclonal antihuman ERCC1 Ab-2 (clone 8F1; Thermo Fisher Scientific, Freemont, CA) diluted 1:200. Staining was visualized with EnVision flex + kit (DAKO) with diaminobenzidine as chromogen, and the sections were then counterstained with Meyer's hematoxylin for 1 minute. The immunostaining was carried out using DakoAutostainer PLUS.
Background staining was evaluated by omitting the primary antibody, in which case no staining was observed. Sections from two FFPE glioblastoma multiforme, known to have loss-of-heterozygosity at the 19q locus, served as negative controls and showed only weak nuclear staining in single sporadic tumor cells. FFPE cell pellets from the ovarian cancer cell line A2780 served as a positive control19 (provided by N.Z. Sroczynski, TopoTarget, Copenhagen, Denmark). The cell pellet showed moderate nuclear staining. Controls are shown in Figure 1 A, B.
The staining was separately evaluated by two independent observers (Z.G.Z., and E.S.-R.) blinded to the clinical data. In discordant cases, consensus was reached by the use of a two-headed microscope. ERCC1 nuclear staining was assessed using a semi-quantitative H-score as previously described.11,12 Briefly, staining intensity was evaluated against a positive control and assigned a value between 0 and 3 (0 = no staining, 1 < control, 2 = control, and 3 > control). Lymph node endothelium stained along with the tumor specimens served as an external control with staining intensity 2 (Figure 1C), and tumor vessels were used as internal controls. The percentage of positive tumor cells were evaluated and a proportion score assigned (0.1 < 10%, 0. 5 = 10–50%, 1 ≥ 50%). The proportion score was determinate by counting at least 500 cells per sample. The final H-score was calculated by multiplying the staining intensity by the proportion score. To establish a reference range, we examined the staining pattern of mesothelium in a cohort of FFPE non-neoplastic pleural samples (n = 34) from patients with benign thoracic diseases. The cutoff point was then chosen as the median H-score in this cohort to separate ERCC1-positive (H-score > median) from ERCC1-negative (H-score ≤ median) tumor samples.
Categorical variables were compared by Fisher's exact test. Survival curves for OS and PFS were obtained with the Kaplan-Meier method. Multivariate Cox regression was used to analyze OS and to investigate the association of ERCC1 status with PFS adjusting for the following known risk factors: patient's age and gender, tumor histology (epitheliod versus nonepitheliod), performance status (Eastern Cooperative Oncology Group performance-status score 0, 1, or 2), and disease stage (Ia/Ib/II, III, or IV). Results are presented as hazard ratios (HR) with 95% confidence intervals (CIs). Time-dependent sensitivity and specificity were computed for the PFS status when predicted by patients' ERCC1 status alone and supplied with exact binomial 95% CIs at each time-point. Progression before time t is regarded as a positive response and disease control until time t is regarded as a negative response for the computation of the diagnostic performance of the marker. Thus, in our setting, the true positive rate (TPR), or sensitivity, is the probability of a patient being ERCC1 positive given progression by time t. Correspondingly, the true negative rate (TNR), or specificity, is the probability of a patient being ERCC1 negative given he or she is progression-free until time t.
Fifty-four consecutive inoperable patients with MPM were enrolled between February 2003 and September 2006. Treatment results and patient characteristics were described in detail by Sørensen et al.15 and briefly summarized in Table 1. The FFPE diagnostic tumor tissue was available from 50 patients (93%). Patients were predominantly male (84%) with epitheloid histology (78%), performance status (PS) 0–1 (94%), and International Mesothelioma Interest Group stages III and IV (80%). Median age was 64 years (range, 31–78 years). On progression, 22 patients (44%) received pemetrexed as second-line treatment. All patients, except two, died during the follow-up period. One had a CR and the OS and PFS times were censored at 77 months. Another patient had a PR but progressed after 40 months and received second-line pemetrexed, the OS time was censored after 44 months.
Immunohistochemical Evaluation of ERCC1 Status
The cohort of non-neoplastic pleura, used to establish the normal reference range, consisted of 34 samples. Twenty-seven were lung resections from patients with relapsing pneumothorax, which showed reactive mesothelium hyperplasia. Seven were pleura-covered lung-tissue sections from patients undergoing lung transplantation because of cystic fibrosis. The median H-score was 2 (lower quartile, 2; upper quartile, 2) and this was then set as the cutoff point. A representative example of ERCC1 immunostaining in the mesothelium of a patient with cystic fibrosis is shown in Figure 1D.
The tumor cohort showed a wider range of H-scores, but here the median H-score was also 2 (lower quartile, 2; upper quartile, 3). The previously established cutoff point separated the tumor samples into 20 positives (H-score >2) and 30 negatives (H-score ≤2). Representative samples of ERCC1-positive and -negative MPMs are illustrated in Figure 1E, F. Demographics and tumor characteristics of patients being ERCC1 positive or ERCC1 negative were comparable with respect to the distribution of gender, age, stage, and histology (p > 0.05). Only three patients in the study-population had PS 2 and they were all ERCC1 positive (p = 0.008). The demographics and tumor characteristics of the study population according to ERCC1 status are shown in Table 1.
Survival and ERCC1 Expression
The median OS in the study population was 16.8 months (95% CI, 12.5–18.6 months) (Table 1 and Figure 2 A). There was no significant difference in OS between ERCC1-positive and -negative patients (HR, 1.29; 95% CI, 0.7–2.36; p = 0.4) (Figure 2). Multivariate Cox regression showed a significant association between PS and OS (HR, 2.15; 95% CI, 1.04–4.42; p = 0.03). The remaining factors did not show a significant association to OS (p > 0.05) (see Table 2).
Treatment Efficiency and ERCC1 Expression
The overall RR in the study population was 35.7%, and there was no significant difference between RR when patients were grouped by ERCC1 status (p = 0.7) (Table 1). The median PFS, however, was 10.9 months (95% CI, 5.6–16.7 months) in the ERCC1-negative tumor group and 6.7 months (95% CI, 4.4–7.2 months) in the ERCC1-positive group (p = 0.053) (Table 1 and Figure 2B). Multivariate Cox regression showed that the ERCC1-positive patients had a significantly shorter time to progression or death (HR, 2.24; 95% CI, 1.16–4.34; p = 0.0163) compared with ERCC1-negative patients (Table 3). The remaining factors (age, gender, histological subtype, stage, and PS) were not significant (p > 0.05).
Predictive Value of ERCC1 Expression
Figure 3 shows time-dependent sensitivity and specificity for predicting patients' PFS-status based on ERCC1. The figure shows sensitivity of ERCC1 below 50% during the whole study period. The specificity, however, is high during the first 3 years. Taken together, these results indicate that patients who experience long-term disease control (implying treatment benefit) are more likely to be ERCC1 negative.
The evidence to support the role of ERCC1 as a marker of cisplatin sensitivity in various cancer types is growing. The clinical utility of this biomarker to guide treatment decisions is currently being addressed in prospective randomized trials, e.g., in NSCLC and epithelial ovarian cancer (clinical trials.gov). The impact of ERCC1 expression on treatment response and prognosis in MPM is not yet well established, and only two previous studies addressing this issue have been published.13,14 Our study aimed at exploring the role of ERCC1 as a predictive biomarker in a well defined and uniformly treated patient population. We chose to evaluate ERCC1 status by IHC because this technique is fast, cheap, and easily applicable to routinely collected histological specimens. We have used the same 8F1 anti-ERCC1 antibody that has been used in several recent publications describing the predictive value of ERCC1 in other human cancer types.11–13,20,21 On the basis of antibody evaluation performed in cultured cell lines, Bhagwat et al.22 have raised concern about the specificity of the 8F1 antibody, but this has later been efficiently rebutted by a specific validation of the 8F1 antibody in human cancer cell lines and FFPE NSCLC tissue sections.23 We have tried to take this issue into account by performing relevant controls. As the ERCC1 gene is located on chromosome 19q, we use as a negative control, FFPE tumor samples from two patients with glioblastoma multiforme, with LOH at the 19q locus, collected and investigated for 19q deletion as part of routine histopathological practice. In these specimens, we saw nuclear positivity in the endothelium and lymphocytes, but most of tumor cells were ERCC1 negative with only few occasional positive tumor cells, as could be expected due to tumor cell heterogeneity. As a positive control, we used a FFPE cell pellet from the ovarian cancer cell line A2780, known to express ERCC1. Here, we saw a moderate nuclear staining intensity. On the basis of these controls, we conclude that the antibody is specific for ERCC1.
We found no correlation between ERCC1 status and objective treatment response evaluated by the modified RECIST criteria applicable to MPM. No difference in OS based on ERCC1 status was seen. Nevertheless, we did find low ERCC1 expression to correlate with longer PFS and a multivariate analysis of PFS-yielded ERCC1 status as the only variable significantly influencing PFS. We also report the time-dependent sensitivity and specificity of ERCC1. Our analysis indicates that ERCC1 status could potentially be useful for discriminating patients with expected long-term disease control from patients with expected early progression, but this requires further validation.
What we observed was a stabilization of disease in terms of longer PFS in the patient group with low ERCC1 expression. The median PFS in the ERCC1-negative group was 10.9 versus 6.7 months in the ERCC1-positive group. A 4-month increase in PFS seems a relevant difference for a relentlessly PD with a short life expectancy. Unfortunately, this difference in PFS did not translate into an OS advantage because the OS in the ERCC1-positive group was 17.4 months opposed to 16.5 months in the ERCC1-negative group. It has been suggested that high ERCC1 expression is associated with a better prognosis in untreated NSCLC, possibly because tumors with more efficient DNA repair accumulate less DNA aberration.11 The difference in OS in our study is not statistically significant, and so we hesitate to believe that the results reflect a truly opposite effect of ERCC1 on OS and PFS, respectively. One reason why we do not see an OS difference between ERCC1-positive and -negative groups might be that on tumor progression, 44% of the study population received pemetrexed as second-line therapy, which might have affected OS and even obscured the effect of ERCC1 on OS.
Our results support the hypothesis that ERCC1 is indeed a marker of cisplatin resistance in MPM. These results may motivate further and larger studies of the predictive utility of ERCC1 as a guide to individualized and optimized treatment. Our study results are consistent with those from similar studies in other cancer types, showing increased benefit from cisplatin-based therapy in patients having ERCC1-negative tumors.
The other published studies on ERCC1 expression in MPM13,14 did not report an association between ERCC1 expression and time-to-progression or PFS after platinum-based palliative chemotherapy. A possible reason for the discrepancy between these results and our own findings might be that all patients in the study by Zucali et al.14 and approximately half the patients in the study by Righi et al.13 were treated with carboplatin-based therapy instead of cisplatin, which is used in our study population. There is no preclinical data suggesting different mechanisms of resistance among platinum compounds, but the influence of ERCC1 on the action of carboplatin, in the clinical setting, is less well explored, and varying results have been observed.24 The fact that both previous studies13,14 used pemetrexed as first-line treatment might also influence the impact of ERCC1 on PFS.
Righi et al.13 identified a significant lower OS in patients having very low expression of ERCC1, when ERCC1 expression was dichotomized using a data-dependent cutoff value (tertile). Such findings could, however, not be reproduced in our data. In the study by Righi et al., the difference in OS, according to ERCC1, occurs at 16 to 18 months after initiation of therapy. One reason why we cannot see the same OS difference might be that 44% of our cohort received second-line treatment (22 patients received poststudy pemetrexed), this might affect and even obscure the effect of ERCC1 on OS.
Both previous studies used the same 8F1 antibody as we use, but the H-score used to evaluate the staining was slightly different from ours as the percentage of positive tumor cells was assigned a continuous score instead of the categorical score we use. This might be another reason for the differences between results highlighting a well-known problem with the quantitative evaluation of IHC. Even when using a semi-quantitative H-score relying on formal criteria and internal controls, this technique is still subject to interobserver and interlaboratory variability. This problem can perhaps be circumvented in the future by using newer techniques like Automated Quantitative Analysis (AQUA technology; HistoRx Inc., New Haven, Conneticut), a fluorescence-based method that provides objective and continuous protein expression scores in tissues using automated fluorescence microscopy and image analysis software.25,26
In our study and in the two previous ones, an intensely positive ERCC1 expression in the non-neoplastic pleura and in most of the tested MPM cases was observed. This could suggest that one of the reasons for the common resistance toward platinum-based chemotherapy, seen in MPM, might reside in the intrinsic ability of mesothelium cells to express high levels of ERCC1.
Our study used archival samples from a prospective, uniformly treated cohort but the study was not originally designed to address the medical utility of this marker in guiding treatment decisions. Nevertheless, because this is to the best of our knowledge the first report on ERCC1 expression in a prospective, uniformly treated MPM cohort, we believe our data can further contribute to establish the clinical validity of this marker in the treatment of MPM. Further studies are of course required to validate our findings and establish the clinical utility of ERCC1 in guiding treatment choices in patients with MPM.
Immunohistochemical evaluation of ERCC1 expression in pretreatment diagnostic pleural biopsies from inoperable MPM patients is feasible and reveals that low ERCC1 expression may predict longer PFS in patients treated with cisplatin-based chemotherapy. These results should be further validated in subsequent studies for confirmation or refutation as false positive. If they are confirmed, these findings may find use in the planning of future customized treatments of MPM, a disease in which better treatment options are urgently needed.
Supported by the Danish Cancer Society and the Danish Lung Association.
1.Churg A, Roggli V, Galateau-Salle F, et al. Mesothelioma. In Travis WD, Brambilla E, Müller-Hermelink HK, et al. (Eds), Pathology and Genetics of Tumours of the Lung, Pleura, Thymus and Heart. Lyon: IARC Press, 2004. Pp. 128–136.
2.Sorensen JB. Current concepts in chemotherapy for malignant pleural mesothelioma. Clin Respir J 2008;2:74–79.
3.Martin LP, Hamilton TC, Schilder RJ. Platinum resistance: the role of DNA repair pathways. Clin Cancer Res 2008;14:1291–1295.
4.Siddik ZH. Cisplatin: mode of cytotoxic action and molecular basis of resistance. Oncogene 2003;22:7265–7279.
5.Smeets H, Bachinski L, Coerwinkel M, et al. A long-range restriction map of the human chromosome 19q13 region: close physical linkage between CKMM and the ERCC1 and ERCC2 genes. Am J Hum Genet 1990;46:492–501.
6.Dabholkar M, Vionnet J, Bostick-Bruton F, et al. Messenger RNA levels of XPAC and ERCC1 in ovarian cancer tissue correlate with response to platinum-based chemotherapy. J Clin Invest 1994;94:703–708.
7.Weberpals J, Garbuio K, O'Brien A, et al. The DNA repair proteins BRCA1 and ERCC1 as predictive markers in sporadic ovarian cancer. Int J Cancer 2009;124:806–815.
8.Kim SH, Kwon HC, Oh SY, et al. Prognostic value of ERCC1, thymidylate synthase, and glutathione S-transferase pi for 5-FU/oxaliplatin chemotherapy in advanced colorectal cancer. Am J Clin Oncol 2009;32:38–43.
9.Shirota Y, Stoehlmacher J, Brabender J, et al. ERCC1 and thymidylate synthase mRNA levels predict survival for colorectal cancer patients receiving combination oxaliplatin and fluorouracil chemotherapy. J Clin Oncol 2001;19:4298–4304.
10.Simon GR, Sharma S, Cantor A, et al. ERCC1 expression is a predictor of survival in resected patients with non-small cell lung cancer. Chest 2005;127:978–983.
11.Olaussen KA, Dunant A, Fouret P, et al. DNA repair by ERCC1 in non-small-cell lung cancer and cisplatin-based adjuvant chemotherapy. N Engl J Med 2006;355:983–991.
12.Vilmar AC, Santoni-Rugiu E, Sorensen JB. ERCC1 and histopathology in advanced NSCLC patients randomized in a large multicenter phase III trial. Ann Oncol 2010;21:1817–1824.
13.Righi L, Papotti MG, Ceppi P, et al. Thymidylate synthase but not excision repair cross-complementation group 1 tumor expression predicts outcome in patients with malignant pleural mesothelioma treated with pemetrexed-based chemotherapy. J Clin Oncol 2010;28:1534–1539.
14.Zucali PA, Giovannetti E, Destro A, et al. Thymidylate synthase and excision repair cross-complementing group-1 as predictors of responsiveness in mesothelioma patients treated with pemetrexed/carboplatin. Clin Cancer Res 2011;17:2581–2590.
15.Sørensen JB, Frank H, Palshof T. Cisplatin and vinorelbine first-line chemotherapy in non-resectable malignant pleural mesothelioma. Br J Cancer 2008;99:44–50.
16.Rusch VW. A proposed new international TNM staging system for malignant pleural mesothelioma from the International Mesothelioma Interest Group. Lung Cancer 1996;14:1–12.
17.Byrne MJ, Nowak AK. Modified RECIST criteria for assessment of response in malignant pleural mesothelioma. Ann Oncol 2004;15:257–260.
18.Francart J, Legrand C, Sylvester R, et al. Progression-free survival rate as primary end point for phase II cancer clinical trials: application to mesothelioma—The EORTC Lung Cancer Group. J Clin Oncol 2006;24:3007–3012.
19.Ferry KV, Hamilton TC, Johnson SW. Increased nucleotide excision repair in cisplatin-resistant ovarian cancer cells: role of ERCC1-XPF. Biochem Pharmacol 2000;60:1305–1313.
20.Ueda S, Shirabe K, Morita K, et al. Evaluation of ERCC1 expression for cisplatin sensitivity in human hepatocellular carcinoma. Ann Surg Oncol 2011;18:1204–1211.
21.Kim KH, Do IG, Kim HS, et al. Excision repair cross-complementation group 1 (ERCC1) expression in advanced urothelial carcinoma patients receiving cisplatin-based chemotherapy. APMIS 2010;118:941–948.
22.Bhagwat NR, Roginskaya VY, Acquafondata MB, et al. Immunodetection of DNA repair endonuclease ERCC1-XPF in human tissue. Cancer Res 2009;69:6831–6838.
23.Olaussen KA, Soria JC. Validation of ERCC1-XPF immunodetection–letter. Cancer Res 2010;70:3851–3852.
24.Vilmar A, Sorensen JB. Excision repair cross-complementation group 1 (ERCC1) in platinum-based treatment of non-small cell lung cancer with special emphasis on carboplatin: a review of current literature. Lung Cancer 2009;64:131–139.
25.Camp RL, Chung GG, Rimm DL. Automated subcellular localization and quantification of protein expression in tissue microarrays. Nat Med 2002;8:1323–1327.
26.Gustavson MD, Bourke-Martin B, Reilly D, et al. Standardization of HER2 immunohistochemistry in breast cancer by automated quantitative analysis. Arch Pathol Lab Med 2009;133:1413–1419.