Digital immunohistochemistry: new horizons and practical solutions in breast cancer pathology

DA performed on the IHC tissue multi-control sections provided continuous output data that could be efficiently summarized (Table 1) and visualized (Figure 1) to reveal variation in the serial sections of tissue multi-controls used in consecutive IHC batches. While linear plot of the mean values (for all 4 tissue samples in the multi-control) of the average membrane intensity and percentage of cells with complete membranous staining, indicated individual IHC batches with unexpected variation (Figure 1), the analyses performed on each multi-control sample uncovered even greater variation (not shown), potentially reflecting an impact of the variable content of the control tissue in the serial sections. The variation was not significantly related to neither the IHC staining time (morning/afternoon/overnight run), nor the weekday (by ANOVA, not shown). The pathologist’s visual review of the controls mostly detected samples with lower than expected IHC HER2 mean intensity score (Table 1). Taking into account only 2+ and 3+ tissue controls given a lower than expected intensity score, only 3 IHC batches would have been considered as “under-stained”. Interestingly, additional review of digital images of the 2+ serial sections arranged consecutively on computer monitor (Figure 2) revealed staining intensity variation, in particular, increased intensity that was missed by conventional microscope review but detected by the DA. To explore possible “long-term” drifts of the IHC sensitivity, we plotted intercepts of the parameters (Figure 3) along the consecutive tests: a mild decrease of the average membrane staining intensity was noted, further supported by corresponding weak correlation between the staining intensity and section number (r=0.32, p<0.0001). Nevertheless, the mean percentage of cells with complete membranous staining remained stable.

The DA-based HER2 IHC quality monitor provided a measurement of IHC staining quality useful in two aspects: as a quality control tool to alert on unexpected variation prospectively and as a quality improvement measure disclosing potential assay drifts based on retrospective analysis of the data. Although the use of DA to improve analytical phase of the IHC test has been suggested [7], we were not able to find published data on practical solutions. In our study, the DA parameters analysed on “per batch” basis revealed a rather broad IHC staining intensity variation underestimated by microscope-based IHC quality control review, while only a few cases with potential quality issues were detected by additional retrospective pathologist’s review.

The application of DA for IHC quality monitor is not as straightforward as it may appear. Several methodological issues have to be considered with caution. Although the same DA algorithm was used for the series, the impact of the tissue variation does not allow for a “pure” IHC tissue control system where, ideally, always an identical tissue section would be used as a reference. Although serial sections are expected to provide rather continuous changes of the tissue properties, cutting and other artefacts may impact the DA results. Furthermore, in our experiment we used a sophisticated DA algorithm, based on automated tumour tissue recognition, which may be dependent on section thickness, hematoxylin staining variation, etc. For this matter, the use of IHC multi-controls containing 4 different tumour samples could be regarded as an “internal control” for tissue-related DA variation, however, further optimization of the approach is needed.

Strong correlation was observed between the numbers of microscope- and DA-estimated HER2 and CEP17 signals/cell (r=0.83, p<0.0001 and r=0.68, p<0.002, retrospectively), and HER2/CEP17 ratio (r=0.71, p=0/0006); the corresponding correlations between two microscope evaluations was close to perfect. However, ANOVA analysis revealed significant bias to lower values of the DA-counted HER2 signals when compared to the microscope evaluation (average difference 1.35, CI 0.4-2.3, p<0.05). Meanwhile, no significant difference was observed between the methods when comparing CEP17 signal counts and HER2/CEP17 ratio.

We present our initial data on the DotFinder algorithm to detect HER2 and CEP17 signals, which is followed by other TissueQuest functionalities to select populations of non-overlapping tumour cells in paraffin sections. Although the final aim of the analysis is to achieve an automated cell-based quantification of the FISH signals providing a robust and high-throughput tool to investigate tumour tissue heterogeneity, in this first stage of validation, we have explored the accuracy of the signal detection. One major obstacle to achieve full automation of the analysis is related to spatial clustering of HER2 signals in some amplified cases; other algorithms (e.g., AutoVysion, Metafer) switch to area-based estimate of the “clusters” [8, 9]. This latter approach is clinically valid to detect amplified cases, however, to investigate intratumoral heterogeneity, a cell-based measurement providing actual number of FISH signals would be preferable. In our experiment, we did not include the “cluster” cases detected by conventional microscope evaluation, nevertheless, underestimation of the HER2 (but not CEP17) signal is likely to be related to spatial confluence of the signals in some HER2-amplified cells. Since the images for the analyses are produced by projecting 9 Z-stack images into one, some information allowing better signal discrimination is inevitably lost. We therefore suggest that with the increase in computing and storage capacity, 3-dimensional FISH-detection algorithms may be the basic way to progress further.

We have recently published [6] a study on breast cancer TMA stained for 10 IHC markers proving a concept that important biological interdependencies can be detected at the level of tumour tissue immunophenotype based on the factor analysis of DA data. This “automated readout” of the IHC data in the TMA revealed independent biological processes standing behind the IHC profile variability in the disease entities. Integral characteristics (factor scores) of individual patients were associated with main conventional categories of the breast ductal carcinoma. In particular, we found that major factor of the IHC profile variation was characterized by a strong inverse relation between the expression of hormone (estrogen, progesteron, androgen) receptors along with anti-apoptotic marker BCL2, on one side, and Ki67 (proliferation) and HIF-1α (hypoxic stress, angiogenesis) on the other side. We named this factor the “i-Grade” since its pattern reflected the interdependent variance of the IHC markers known to represent the axis from aggressive (Ki67, HIF-1α) to more indolent (hormone receptor-positive, BCL2) behaviour of the disease and was associated with the Nottingham histological grade. Also, we were able to test independent informative value of conventional and less explored IHC biomarkers and their combinations.