Patients

The study population was derived from the archives of the Departments of Pathology of the University Medical Center Utrecht, Utrecht, and the Radboud University Nijmegen Medical Centre, Nijmegen, The Netherlands. These comprised 483 cases of invasive breast cancer (operated between 1997 and 2007), of which 340 cases were part of a consecutive series (operated between 2003–2007). The series was enriched with a small consecutive series of lobular breast cancers and a consecutive series of 23 cases with a BRCA germline mutation as previously described [24].

Histological grade was assessed according to the Nottingham scheme [25], and mitotic activity index (MAI) was assessed as before [26]. From representative donor paraffin blocks of the primary tumors, tissue microarrays were constructed by transferring tissue cylinders of 0.6 mm (3 cylinders per tumor) from the tumor area, determined by a pathologist based on haematoxylin-eosin stained slides, using a tissue arrayer (Beecher Instruments, Sun Prairie, WI, USA) as described before [27]. Normal breast tissue was obtained from patients that underwent mammoplasty (and thus had no tumor at all). In case of matched tumor and normal tissue, we analyzed normal tissue in paraffin blocks that did not contain any tumor and thus were far away from the tumor. The use of anonymous or coded left over material for scientific purposes is part of the standard treatment contract with patients in The Netherlands [28]. Ethical approval was not required.

Immunohistochemistry

Immunohistochemistry was carried out on 4 μm thick sections for a panel of potential molecular membrane bound targets known to be expressed in a frequency of >10% in breast cancer. These were partly highly tumor specific, meaning that they have no or low intensity staining of the normal breast tissue (GLUT1, EGFR, insulin-like growth factor-1 receptor (IGF-1R), HER2, CAIX, hepatocyte growth factor receptor (MET)). We also included less tumor-specific, meaning that are known to have moderate or high intensity staining of the normal breast tissue (Mucin 1 (MUC1), CD44v6, Mammaglobin, transferrin receptor (TfR), carbonic anhydrase 12 (CAXII)), since cancers have usually increased cellularity compared to the normal breast and could thereby also provide adequate signal-to-noise in tumors compared to the normal breast.

After deparaffination and rehydration, endogenous peroxidase activity was blocked for 15 min in a buffer solution pH5.8 containing 0.3% hydrogen peroxide. After antigen retrieval, i.e. boiling for 20 min in 10 mM citrate pH6.0 (for progesterone receptor (PR), CD44v6, GLUT1, CAIX, MET, TfR, and CAXII), Tris/EDTA pH9.0 (estrogen receptor α (ERα), HER2, IGF1-R, MUC1, and Mammaglobin) or Prot K (0.15 mg/ml) for 5 min at room temperature (EGFR), a cooling off period of 30 min preceded the primary antibody incubation. CD44v6 (clone VFF18, BMS125 Bender MedSystems, Austria) 1:500; ERα (clone ID5, DAKO, Glostrup Denmark) 1:200; PR (clone PgR636, DAKO) 1:100; HER2 (SP3, Neomarkers, Duiven, The Netherlands) 1:100; GLUT1 (A3536, DAKO) 1:200; CAIX (ab15086, Abcam, Cambridge, UK) 1:1,000; IGF1-R (NB110-87052, Novus Biologicals, Cambridge, UK) 1:400; TfR (13–6800, Invitrogen, Breda, The Netherlands) 1:300; MUC1 (EMA, M1613 clone E29, DAKO) 1:400; Mammaglobin (clone 304-1A5, DAKO) 1:100; CAXII (HPA008773, Sigma Aldrich, Zwijndrecht, The Netherlands) 1:200 were incubated for 1 h at room temperature. Primary antibodies against EGFR (clone 31 G7, Zymed, Invitrogen) 1:30; MET (18–2257, Zymed, Invitrogen) 1:100 were incubated overnight at 4°C. All primary antibodies were diluted in PBS containing 1% BSA.

The signal was amplified using Powervision poly-HRP anti-mouse, rabbit, rat (DPVO-HRP, Immunologic, Duiven, The Netherlands) or the Novolink kit (Leica, Rijswijk, The Netherlands) (in the case of EGFR) and developed with diaminobenzidine, followed by counterstaining with haematoxylin, dehydration in alcohol and mounting.

Scoring of immunohistochemistry

All stainings were compared to normal breast tissue and scored as positive when a clear membranous staining was seen and when the expression in the tumor was clearly higher than in the normal breast tissue. All stainings were scored using the DAKO/HER2 scoring system for membranous staining. Scores 2+ and 3+ were considered as positive except for HER2 where only a score of 3+ was considered positive. Due to the strong intra-tumor heterogeneity of Mammaglobin expression, scoring was performed by estimating the percentage of positive tumor cells, considering cancers with more than 35% of the membrane stained tumor cells as positive. All scoring was done by a single experienced pathologist (PJvD) who was blinded to patient characteristics and results of other stainings. To take tumor-heterogeneity between the tumor cores into account, the average score per tumor was calculated and used for analyses. Only in case of GLUT1 and CAIX, the tumor was classified as positive when a single core showed positivity. In this study a maximum of 3 missing stainings per patient was allowed, these stainings were considered as negative in the analyses. This potentially results in underestimation of the percentage positivity of a marker.

Based on ERα, PR, and HER2 immunohistochemistry, tumors were classified as luminal (ERα and/or PR positive), HER2-driven (ERα-, PR-, HER2+), triple negative (ERα-, PR-, HER2-) or basal (ERα-, PR-, HER2-, EGFR+), the immunohistochemical surrogate [29] of the original Sorlie/Perou classification [30].

Immunofluorescence for quantification of protein expression in tumor and normal breast tissue

Several of the evaluated molecular membrane targets (CD44v6, MUC1, TfR, Mammaglobin, and CAXII) are known to be expressed to some extent in the normal breast epithelium. In order for these targets to be useful for breast cancer screening by optical imaging, the signal to background ratio needs to be high enough to be discriminative. We therefore performed immunofluorescence with these antibodies to allow quantification of expression ratios between normal breast and cancer tissue of four randomly selected patients by image analysis.

Immunofluorescence was performed as described above for immunohistochemistry, except that the primary antibodies were detected by incubation with Goat-anti-mouse/rabbit Alexa555 (1:1,000, Invitrogen) for 1 h at room temperature, followed by 4,6-Diamidine-2-phenylindole dihydrochloride (DAPI) counterstaining and mounting with Immumount (Thermo Scientific, Etten-Leur, The Netherlands).

Representative images of normal breast and breast cancers from the same sections were taken using identical settings at 20x magnification using a Leica DMI4000b inverted bright-field/fluorescence microscope.

Image analysis of tumor expression versus normal breast tissue

Conventional immunohistochemical slides were digitalized for image analysis using a digital slide scanner (Aperio Technologies Inc., Vista, CA, USA). Of each patient four representative areas of normal and tumor tissue were selected and the average membrane intensity was calculated with the IHC membrane algorithm (Aperio, v8.001). As the signal-to-noise ratio in vivo is determined by the difference in expression between cancer and normal cells as well as by cellularity, the number of cells in the selected area was obtained from the algorithm. Tumor-to-normal ratio was calculated as (membrane intensity*cellularity/area) of the tumor/(membrane intensity*cellularity/area) of normal tissue.

Tumor-to-normal ratios of the fluorescently labeled antibodies were calculated with ImageJ using the median intensity scores. Values are expressed as the average tumor/normal ratio ± SEM.

Based on experience in radiology with the blood-pool agent indocyanine green in studies assuming a leaky vessel model [31, 32], and from studies using NIRF labeled trastuzumab/bevacizumab in mouse models [33], a tumor-to-normal ratio larger than 3 was considered to be sufficient for optical imaging.

Statistics

Statistical analysis was performed using IBM SPSS Statistics version 18.0 (SPSS Inc., Chicago, IL, USA). Associations between categorical variables were examined using the Pearson’s Chi-square test. P-values <0.05 were considered to be statistically significant.

Detection rate of combinations of highly tumor-specific molecular targets in relation to grade, molecular and histological type

Because the frequency of expression (further denoted 'detection rate') of individual highly tumor-specific markers did not exceed 20.3% of the cases, we examined several combinations of markers by sequential addition of markers to the expression of GLUT1, the most widely expressed highly tumor-specific marker. GLUT1 in combination with EGFR resulted in 30.0% positive cases, GLUT1/IGF1-R in 28.8%, GLUT1/HER2 in 27.7%, GLUT1/MET in 25.2%, and GLUT1/CAIX in 22.3% positive cases. The panel GLUT1, EGFR, HER2, IGF1-R, MET, and CAIX resulted in 45.5% positive cases, although the contribution of CAIX and MET was minimal (Figure 2A).

Molecular targets that are expressed in normal breast tissue have sufficient signal-to-noise to detect lobular and luminal-type breast cancer

Since lobular and luminal-types of breast cancer appeared to hardly express tumor-specific antigens, antigens that are less tumor-specific are required for their detection. Like with tumor-specific markers, variation between histological and molecular subtypes was observed for TfR, Mammaglobin, and CAXII. Luminal-type ductal cancers and lobular cancers expressed significantly more CAXII (10.5% vs. 2.3%, p = 0.017) and Mammaglobin (19.9% vs. 5.9%, p = 0.002) compared to HER2-driven and basal/TN ductal cancers (Tables 3 and 4). TfR expression in lobular and luminal type ductal cancers was significantly lower than in HER2-driven and basal/TN cancers (11.9% vs. 27.9%, p < 0.001). For MUC1 and CD44v6, no differences in expression were found between lobular and ductal cancer (Tables 3 and 4).

Due to the expression of less tumor-specific antigens in the normal breast epithelium (Figure 1B), the signal-to-noise ratio (or tumor-to-normal) needs to be sufficiently discriminating to be applicable for imaging strategies. We determined therefore the tumor-to-normal ratio in a quantitative manner by image analysis of digital slides, considering a 3-fold tumor-to-normal ratio as sufficient. Image quantification using conventional IHC showed that the intensity of the staining was dependent on the cellularity of the tumor as expected. This resulted in tumor-to-normal ratios of 4.8 ± 0.56, 2.3 ± 0.27, 1.2 ± 0.095, 4.6 ± 0.62, and 2.4 ± 0.88 for CD44v6, MUC1, Mammaglobin, CAXII, and TfR, respectively.

Since conventional immunohistochemistry is not necessarily quantitative, we also performed immunofluorescence using directly fluorescently labeled antibodies. The results were comparable with conventional immunohistochemistry (Figure 3) resulting in tumor-to-normal ratios of 3.93 ± 0.14, 2.74 ± 0.46, 1.54 ± 0.11, and 1.66 ± 0.066 for CD44v6, MUC1, Mammaglobin, and CAXII, respectively. TfR expression was not detectable using immunofluorescence. Thereby, CD44v6 was the only less tumor-specific marker consistently meeting the required 3-fold tumor-to-normal ratio.

Detection rate of combined highly and less tumor-specific molecular targets

Including TfR, Mammaglobin, and MUC1 to the panel of highly tumor-specific markers GLUT1, MET, EGFR, IGF1-R, CAIX, and HER2 increased the detection rate from 45.5% to 49.8% (TfR), 56.4% (Mammaglobin), and 98.1% (MUC1), respectively. However, of these markers, only CD44v6 reached a sufficiently high tumor-to-normal ratio (see above), so adding CD44v6 to the panel of highly specific markers therefore realistically increased the overall detection rate to 80.1%. When CD44v6 was included, removal of CAIX or MET from the panel had no influence on the detection rate.

Especially the luminal-type ductal and lobular breast cancers were better detected by including CD44v6. Upon addition of CD44v6, the detection rate rose from 45.5% to 78.9% for luminal-type cancers, from 18.3% to 72.2% for the lobular breast cancers, and from 84.2% to 90.0% for basal/TN ductal breast cancers (Figure 4A). Moreover, the detection rate of the panel was not dependent on grade (76.4%, 74.0%, and 84.5% for grade 1, grade 2, and grade 3 tumors, respectively), tumor size (79.1%, 77.6%, and 85.7% for tumors ≤2 cm, >2 and ≤5 cm, and >5 cm, respectively), lymph node status (76.2% for lymph node negative, and 82.7% for lymph node positive cases), or age (78.8% for patients <60 years and 80.1% for patients >60 years) (Figure 4B).

Therefore, the optimal combination of membrane-expressed proteins to target by molecular imaging seemed to consist of CD44v6, GLUT1, EGFR, HER2, and IGF1-R by which about 80% of invasive breast cancers are predicted to be detectable.