Assay development and measurement of Her2 levels as absolute and continuous variables in 332 FFPE samples
We first developed QDB-based immunoassay for Her2 measurement in FFPE samples by defining the linear range of the assay. Total tissue lysate was extracted from 2 × 15 µm FFPE specimen by de-paraffinization and solubilization with Triton-X 100 lysis buffer. Lysates from four Her2 + samples based on IHC analysis were pooled together in equal amount. The linear range of the assay was explored using pooled lysates and recombinant Her2 protein in serial dilutions in QDB measurement with both 4B5 and EP3 antibodies respectively (Additional file 1).
The Her2 protein levels were measured using total tissue lysates extracted from 332 FFPE samples with both 4B5 and EP3 antibodies within their defined linear range respectively. Samples were provided as 2 × 15 µm FFPE slices sequentially and non-selectively by local hospital with clinicopathological characteristics listed in Table 1. The flow diagram was shown in Fig. 1. Her2 levels measured with 4B5 and EP3 antibodies were high correlated, with r = 0.963, p = 0.0000, n = 332 when evaluated with Pearson’s correlation coefficient analysis (Fig. 2). For simplicity, we limited our analysis in this study with Her2 levels measured with EP3 antibody.
Table 1
The clinicopathological characteristics of the patients.
Variable | | No. of patients | Average ± SEM | Percentage |
Age (y) | Total | 332 | 53.3 ± 0.6 | |
< 50 ≥ 50 Unknown | 122 209 1 | | 36.7% 63.0% 0.35 |
Histological Grade | I II III Unknown | 36 145 119 32 | | 10.8% 43.7% 35.8% 9.6% |
Tumor Size (mm) | Total | 332 | 2.3 ± 0.6 | |
≤ 20 20 ~ 50 > 50 Unknown | 173 151 5 3 | | 52.1% 45.5% 1.5% 0.9% |
Histological Type | Ductal Lobular Other Unknown | 298 9 23 2 | | 89.85 2.7% 6.9% 0.6% |
Nodal Status | Negative Positive | 220 112 | | 66.3% 33.7% |
Her2 (IHC) | 0 1+ 2+ 3+ | 77 65 108 82 | | 23.2% 19.6% 32.5% 24.7% |
Her2 (FISH) | Negative Equivocal Positive Unknown | 95 6 43 188 | | 28.6% 1.8% 13.0% 56.6% |
A total of 332 breast cancer FFPE tissues in 2 × 15 µm slices were provided by a local hospital. MCF-7 and BT474 cell lysates were used as internal controls. FFPE tissue lysates (about 0.5 µg /unit) and cell lysates (about 0.3 µg/unit) were applied onto the QDB plates at 2 µl/unit in triplicate for the QDB measurements with clone EP3 and 4B5 respectively. A set of serially diluted Her2 recombinant protein were included in each plate to develop plate-specific standard curve. All results were averaged from three independent experiments, with each sample in triplicate. The correlation of Her2 levels measured with 4B5 and EP3 was analyzed with Pearson’s correlation coefficient analysis using Graphpad software, r = 0.963, p < 0.0001.
The distributions of Her2 levels among these samples were shown in Fig. 3A. The absolute Her2 level was distributed from non-detectable (signal below two-fold of background, and enter 0 as final result) to as much as 31.31nmole/g. The group average was 1.953 ± 0.254 nmole/g with 25th and 75th percentile at 0 and 0.987 nmole/g respectively.
Her2 levels in all 332 breast cancer FFPE sample lysates were measured with QDB method using EP3 antibody. The lysates were diluted to about 0.25 µg/µl, and then 2 µl lysate was used for each sample. (A) the distribution of Her2 levels among 332 samples. Her2 levels were ranging from 0 (chemiluminescence readings less than two times the background) to 31.31 nmole/g, with the mean at 1.953 ± 0.254 nmole/g. The 25th percentile was at 0 nmole/g and the 75th percentile was at 0.987 nmole/g. (B) All samples were grouped by their IHC scores provided by local hospital. The distributions of Her2 levels in each IHC group were recorded as following: 0, 0 ~ 0.205 nmole/g, mean = 0.045 ± 0.006 nmole/g, n = 77; 1+, 0 ~ 0.41 nmole/g, mean = 0.049 ± 0.008 nmole/g, n = 65; 2+, 0 ~ 7.25 nmole/g, mean = 0.537 ± 0.122 nmole/g, n = 108; and 3+, 0.329 ~ 31.31 nmole/g, mean = 7.12 ± 0.773 nmole/g, n = 82. The intra- and inter-CV were 8.98% and 9.89% respectively.
The samples were grouped into 0, 1+, 2 + and 3 + groups based on IHC scores with average at 0.045 ± 0.006 (n = 77), 0.049 ± 0.008 (n = 65), 0.537 ± 0.122 (n = 108), and 7.12 ± 0.773 (n = 82) nmole/g respectively (Fig. 3B). The differences between each individual group were with statistical significance with one exception when analyzed using unpaired two-tailed Student’s t-test (p < 0.005). There was no statistical difference between group 0 and group 1+.
Validation of QDB method
The only method for absolute quantification of Her2 levels was SRM-MS, which was still in developmental stage[8, 10, 15]. Consequently, we had to rely on the results from both IHC and FISH analyses to validate our results indirectly, as the results from QDB measurement were absolute and continuous variables, while those from FISH and IHC analyses were relative and discrete variables. Therefore, receiver operating characteristic (ROC) analysis was used to evaluate QDB method with provided Her2 results from local hospital.
The samples were grouped into Her2 + and Her2- based on both IHC and FISH analyses (IHC/FISH), eliminating all the equivocal cases in the process. We achieved area under the ROC curve (AUC) at 0.9753 ± 0.01026, 95% CI at 0.9551 to 0.9954, with p < 0.0001 (n = 322) (Fig. 4A). We also identified the optimized cutoff at 0.261 nmole/g to achieve sensitivity at 94.23% (95% CI: 87.87–97.85%) and specificity at 94.5% (95% CI: 90.58–97.12%) respectively. Using this cutoff, we achieved concordance rate with IHC at 99.6%, and with FISH analysis at 88.6% (κ = 0.732, n = 140 with Cohen’s kappa analysis) respectively.
Samples were separated into negative (her2-) and positive (Her2+) groups using provided results from local hospital based on the recommendations from ASCO/CAP. (A)The negative group consisted of 142 samples of IHC 0 and 1 + and 76 samples of IHC 2 + with FISH negative. The positive group consisted of 82 samples of IHC 3 + and 22 samples of IHC 2 + with FISH positive. Samples scored as IHC 2 + without FISH results (n = 7) and with FISH results equivocal (n = 3) were excluded. Absolute Her2 levels from QDB analysis were used for ROC analysis with Graphpad Prism7.0 software. The ROC curve of QDB analysis was obtained with area under the Curve (AUC) at 0.9753 ± 0.01026; 95% CI: 0.9551 ~ 0.9954; P < 0.0001. (B) Samples were grouped based on their IHC scores alone, with 142 samples in the negative group (IHC 0 and 1+), and 82 samples in the positive group (IHC 3+). Samples with IHC score of 2 + were excluded from the analysis. The ROC curve of QDB analysis was obtained with area under the Curve (AUC) at 0.9998 ± 0.0002; 95% CI: 0.9994 ~ 1; P < 0.0001. (C) Samples were grouped based on FISH results alone, with 95 samples as negative (her2-) group and 43 samples as positive (Her2+) group. Equivocal cases (n = 6) were excluded in the analysis. The area under the curve (AUC) was at 0.978 ± 0.0112, with 95% CI at 0.9561 ~ 0.9999; P < 0.0001. (D) Samples were grouped by IHC scores, and the suggested cutoff values from ROC analyses in (B) at 0.267 nmole/g (solid line), in (A) (C) at 0.261 nmole/g (dashed line) were shown to demonstrate the effectiveness of these cutoff values to separate samples from Her2 + to Her2- groups. Her2 levels were plotted in log scale to better demonstrate the distribution of QDB results among these samples. For those samples with undetectable Her2 level, a value of 0.001 nmole/g was arbitrarily entered to avoid omitting any sample in the log scale graph.
ROC analyses were also performed using provided IHC scores alone, eliminating all the 2 + group, and we achieved AUC at 0.9998 ± 0.0001, 95% CI at 0.9994 to 1, with p < 0.0001 (n = 224) (Fig. 4B). The optimized cutoff was identified at 0.267 nmole/g to achieve 100% sensitivity (95% CI: 95.6–100%) and 99.3% specificity (95% CI: 96.14–99.98%). Likewise, for FISH analysis, we were able to achieve AUC at 0.978 ± 0.0112, 95% CI at 0.9561 to 0.9999, with p < 0.0001 (n = 138) (Fig. 4C). The optimized cutoff was identified at 0.261 nmole/g to achieve 93.02% sensitivity (95% CI: 80.94–98.54%) and 93.68% specificity (95% CI: 86.76–97.65%). These cutoffs were also shown in a log scale plot of all the samples grouped by their IHC scores (Fig. 4D).
Using the 0.261 nmole/g cutoff derived from ROC analysis of IHC/FISH results, we identified 16 samples (11.4%) in disagreement with provided FISH results. To rule out potential misdiagnosis, these samples were submitted to a third party for independent FISH analysis (Additional file. 2). We were able to re-categorized 6 samples, with 2 samples as equivocal case. The overall concordance rate between QDB and IHC/FISH was also increased from 94.4–96.6% consequently.
The correlation between Her2 copy numbers from FISH analysis, reflected by the ratio of Her2 number over chromosome 17 number (Her2/CEP17), with Her2 protein level as continuous variables was analyzed in Additional file. 3. We found a strong correlation between DNA amplification level and Her2 protein level, with r = 0.75 with Pearson’s correlation coefficient analysis (n = 122).
Exploration of the correlation between clinicopathologic factors and Her2 as absolute and continuous variables
The quantitated Her2 levels in FFPE samples allows us to investigate the correlation between Her2 levels as absolute and continuous variables with other clinicopathologic factors including age, histological grade by Nottingham grading system, tumor size and metastasis status. Her2 levels were found to be associated significantly with histological grade based on Nottingham grading system using Spearman’s rank correlation analysis (ρ = 0.195, p = 0.001), a conclusion consistent with previous studies based on IHC analysis[16, 17]. In the same study, we found that age was negatively associated with Her2 with statistical significance based on IHC analysis (ρ=-0.117, p < 0.05), but not based on the absolutely quantitated Her2 levels (ρ=-0.084, p = 0.127) (Additional file. 4).
Her2 distribution by histological grade was further analyzed in Fig. 5. We observed the average of these samples by Grades at 0.791 ± 0.555, 1.554 ± 0.330, 3.271 ± 0.535 nmole/g for Grade I, Grade II, and Grade III respectively. There were statistical significance between Grade I vs Grade III (p < 0.05) and grade II vs Grade III (p = 0.005) using unpaired two-tailed Student’s t-test. We also calculated the percentage of Her2 + in each grade with 8.3% for Grade I, 29.7% for Grade II and 47.1% for Grade III. Thus, the possibility of Her2 + for Grade III patient was 5.7 fold over that of Grade I patient.