- Email: [email protected]
Appendix 2
Appendix 2
269
APPENDIX 2
INCREASING SENSITIVITY AND ACCURACY OF QUANTITATIVE IMMUNOFLUORESCENCE IN FFPE TISSUE WITH SPECTRAL IMAGING Introduction Understanding cancer and its complexity can be advanced significantly with better tools for measuring proteins in situ in formalin-fixed, paraffin-embedded tissue sections. Separation technologies, such as Western 270
blot, microarrays, and mass spectrometry are used widely, but lose key architecturally-specific signals that reside at the cellular level, and blend signals from multiple cell types. Laser capture micro-dissection attempts to address this issue, but is expensive and laborious to perform. IHC has the significant advantage of retaining tissue architecture and heterogeneity, and presents views of protein expression, but is non-quantitative and variable. Most separation approaches disaggregate tissues, blending signals from many cells and tissues.
Appendix 2
monochrome image, acquired with emission filtering centered on the AlexaFluor™555 emission peak. Three locations in the sample were used as reference points: in cytoplasm; in a nucleus; and off the sample. A signal-tobackground ratio and autofluorescence contribution were calculated for each image from the cytoplasm and off-sample signal counts.
Fluorescence microscopy is becoming increasingly important for these endeavors, compared to chromogenic immunohistochemistry, due to higher multiplexing capability, larger and more linear signal range, and less interference among labels. However, immunofluorescence poses challenges, including the presence and effects of autofluorescence and the inability to distinguish overlapping signals due to cross-talk. Multispectral imaging eliminates these issues through the use of spectral unmixing, which enables isolation of individual biomarker signals, even when signals are substantially overlapping spatially and spectrally, and are obscured by autofluorescence signals. Sensitivity and accuracy can be increased several-fold. The purpose of this application note is to provide an example of the quantitative advantage of multispectral imaging.
Methods A tissue microarray of lung cancer tissue was labeled with a multicolor immunofluorescence cocktail commercially available from Cell Signaling Technology (Pathscan Node Kit #8999). The kit comprises two tubes of mixed reagents, one of primaries and one of conjugated secondaries. Phospho-AKT is labeled with AlexaFluor™ 488, phospho-ERK with AlexaFluor™ 555, and phospho-S6 with AlexaFluor™ 647. DAPI nuclear counterstain is included. Multispectral images were acquired with a Caliper Nuance multispectral imaging system and spectrally unmixed into label and autofluorescence signals. For comparison, conventional immunofluorescence images were acquired with fluorescence emission filters centered on the emission peak of the respective fluorophores. For this analysis, we compared the signals for the phospho-ERK (AlexaFluor™ 555) signal in typical microarray cores, as indicated in the spectrally unmixed component image and in the conventional fluorescence emission
Results Measured signals in the conventional immunofluorescence images were significantly higher than in the spectrally unmixed component images, due to the presence of a high autofluorescence signal present, in addition to the pERK signal. In the spectrally unmixed component image, the autofluorescence is removed, and thus the pERK signal is pure and more accurate. A further benefit of spectral unmixing is the removal of cross-talk from overlapping fluorophores spectra, in multi-label assays. Signal-to-background in the spectrally unmixed images is 50 and in the conventional monochrome image is 3.9, a 13× improvement. More importantly, the data suggest that 34% of the signal measured with conventional epifluorescence was actually autofluorescence or cross-talk from another fluorophores labels ((2991−1959)/2991). 271
APPENDIX 2
Figure 1 Lung tissue labeled with AlexaFluor™ 488, 555, and 647 and a DAPI counterstain. Panel A shows an RGB image of the sample with pAKT immunolabeled with AF488, pERK immunolabeled with AF555, and pS6 immunolabeled with AF647. Panels B through E display these unmixed component images along with the autofluorescence component in Panel F, whose border colors correspond to the pseudocolors used to form the composite image, G. The autofluorescence spectrum is unmixed in the black channel so it is not visible in the unmixed composite image.
Table 1 Signal counts found from the two pERK images in Figure 2. The signal-to-background ratio of each image was calculated by dividing the signal from the cytoplasm by the off-sample background signal Signal (counts)
272
Figure 2 Images of pERK signal generated with conventional fixed filters (above) and with spectral unmixing (below). Signal counts were found from the same three pixel locations in both images, in cytoplasm where pERK is likely expressed, in the nucleus, and a third off the sample to provide a background signal.
Conventional
Multispectral
Cell #1
2991
1959
Cell #2
2521
1614
Stroma
701
34
Signal/Background
3.9
50
Conclusion Spectral imaging enables reliable and accurate assessments of weakly expressing proteins in FFPE sections, even when fluorescence emissions are substantially weaker than background autofluorescence. The approach also automatically corrects for cross-talk between labels in multiplexed assays, thus giving pure signals of biomarkers and greatly increasing the signal-to-noise ratio compared to conventional methods.
Appendix 2
RECEPTOR TRAFFICKING ANALYSIS WITH MULTI-COLOR FLUORESCENCE LABELING AND MULTISPECTRAL IMAGING Example: EGFR Activation
Introduction Analysis of signaling pathway activity in FFPE tissue is essential in oncology research. For example: the epidermal growth factor receptor (EGFR) is a transmembrane receptor tyrosine kinase that belongs to the HER/ErbB protein family. Dysregulation of EGFR and other receptor signaling pathways through activating mutations or gene amplification has been implicated in the pathogenesis of many human carcinomas, leading to extensive clinical study. Immunohistochemical studies have limitations due to the difficulty in interpreting the results, as well as specificity of staining. Multiplexing allows simultaneous detection of multiple biomarkers, revealing specific signaling configurations and tumor heterogeneity. In this example, we demonstrate multiplex labeling of total EGFR epitopes and Phospho-EGFR epitopes to determine percent of activated receptors within tumor-positive cells. The antibodies recognize total EGFR and phosphorylated-EGFR specifically upon activation of tyrosine residues. The phosphorylation EGFR is rapidly endocytosed, and either recycled back to the plasma membrane or targeted for lysosomal degradation. The ratio of total EGFR versus Phospho-EGFR reveals the percentage of receptor being recycled by tumor cells. Thus, Caliper’s Vectra™ multispectral imaging,
automated tumor and cell finding software analysis offers precise detection of individual proteins through spectral unmixing based on their spectral signatures. inForm™ analysis software segments the individual tumor cells and provides quantitation for each cellular compartment.
Methods The tissues used in this analysis are mouse xenografts of lung cancer cell lines. EGFR signaling has been found to be dysregulated in several types of this cancer. Expression and/or activation have been linked to therapeutic response, angiogenesis, and metastasis. Labeling is done with a commercially available kit from Cell Signaling Technologies (Catalog #7967), used to detect simultaneously expression, localization, and activation state of EGFR, and additionally the downstream signaling through Erk1/2.
Figure 1 Multispectral 20x images acquired from multiplexed non-small cell lung carcinoma samples with pEGFR immunolabeled with AlexaFluor™ 488, total EGFR immunolabeled with AlexaFluor™ 555, pERK with AlexaFluor™ 647, and a DAPI counterstain. (A) Amplified H3255 cell line. (B) Amplified Kyse450 cell line. (C) Unamplified H1975 cell line. (D) Unamplified H1703 cell line.
Sample: Four human tumor xenografts of non-small cell lung carcinoma cell lines. n Two amplified: H3255/L858R and Kyse450/wild-type. n Two unamplified: H1975/L858R and H1703/wild-type. n
Immunofluorescence Staining: Cell Signaling Technology PathScan EGF Receptor Activation Multiplex IF Kit #7967. n Primary antibodies: pEGFR; total EGFR; pERK n Secondary antibodies: AlexaFluor™ 488; AlexaFluor™ 555; and AlexaFluor™ 647. n DAPI counterstain. n
Imaging and Analysis: Caliper’s Vectra automated imaging system and inForm image analysis software. 273 n Acquired four 20× multispectral images per sample. n
APPENDIX 2
Figure 2 Image cubes of amplified H3255 cells. (A) RGB representation of multispectral cube. (B) Composite image of unmixed signals with pEGFR represented in green, EGFR represented in red, and co-localization shown in yellow. (C) RGB image overlayed with an inForm tissue segmentation map, where red areas are tumor, green is stroma, and the background is in blue. (D) RGB image overlayed with an inForm cell segmentation map, where each cell in the tumor area has been segmented into nucleus, cytoplasm, and membrane compartments.
Cytoplasm
Membrane
3.5
7
3
6
2.5
5
2
4
1.5
3
1
2
0.5
1
0
H1703
H1975
H3255
Kyse
0
EGFR pEGFR pErk
H1703
H1975
H3255
Kyse
Figure 3 Comparison of total EGFR, pEGFR, and pERK expression levels in different cellular compartments for each xenograft cell line.
Automated machine-learning-based image analysis to: n Detect areas of tumor n Segment the cells into subcellular compartments n Extract IF signals on a per-cell basis
n
Results Evaluation of individual biomarkers following multispectral unmixing revealed changes in pEGFR expression compared to total EGFR in xenograft expression specific tyrosine kinase mutations. Since cytoplasmic pEGFR represents an activated form of surface EGFR, it was not observed within the proximity of total EGFR. Data analysis revealed tyrosine kinase EGFR mutant xenografts showed lower expression of pEGFR, while wildtype/Kyse450 xenograft samples showed increase in total EGFR, pERK, and a decrease in cytoplasmic pEGFR expression. The data demonstrates that the ratio of pEGFR versus total EGFR is consistent with cytoplasmic to membrane receptor recycling. Activation of pERK did not correlate with EGFR expression, which is not unexpected as it 274 could be activated by RAS/MEK signaling pathway.
Conclusion Multiplexing provides individual biomarker signals whose expression correlates with dynamic changes in receptor trafficking, activation of specific signaling pathways, and modifications in tumor cell morphology. In conclusion, high-throughput imaging platform along with image analysis software and a panel of phosphorylationspecific antibodies could reveal molecular subtypes of tumors for effective personalized therapy.
AUTOMATED IMAGE ANALYSIS OF TUMOR-INFILTRATING LYMPHOCYTES Introduction Tumors contain variable numbers of CD3+, CD4+, CD8+ and FoxP3+ lymphocytes referred to as tumorinfiltrating lymphocytes (TILs). Initially, TILs were thought to reflect the origin of cancer at sites of chronic inflammation. Recent studies revealed a relationship between the intensity as well as the ratio of CD3+, CD4+ CD8+, and FoxP3 lymphocytes with prognostic outcome
Appendix 2
Figure 1 (A) Multispectral 20x images from four example cores acquired from a TMA of ovarian cancer stained with hematoxylin, Vector Red labeling cytokeratin, and DAB labeling anti-CD3 antibody. (B) The same four cores with tissue and cell segmentation maps overlaid with red areas representing tumor, blue areas representing other tissue, and green outlines indicating DAB+ TILs.
Methods Sample: Tissue microarray (TMA) containing 618 ovarian cancer samples. n The TMA samples were stained with anti-CD3 antibody to identify T-lymphocytes. n The samples were further stained with epithelial cell specific marker (cytokeratin) to assist in automated segmentation of tumor and stroma. n
Automated Scoring: The TMA slides were scanned using Caliper’s Vectra™ multispectral imaging system. The scanned images were processed using Caliper’s inForm™. A machine-learning algorithm was trained to segment tumor from stroma and identify CD3 cells labeled with DAB, indicating T-lymphocytes. n Density of TILs within the tumor areas was calculated. n
Visual Scoring: A pathologist rated lymphocyte density. A semi-quantitative scale of 0 to 3 was assigned to each core.
n n
Analysis: Automated counts were compared to visual scores. Manual and automated scoring were compared with survival.
n n
80
Hematoxylin-CK-CD3 automated count
of different solid tumors. Potentially, tests that characterize TIL phenotype, location, and density in tumor and surrounding stroma could aid the selection of immunotherapy and define targets for individualized treatment. However, TIL assessment today, based on human visual perception, is prone to inter- and intra-observer variability, due to complicated tissue architectures, ambiguous histology revealed through hematoxylin counterstain, and other human factors. Also, visual TIL assessment is tedious and time consuming. In this study, performed in collaboration with Dr.Michael Feldman and Dr. Ian Hagemann at University of Pennsylvania, we investigate computer-aided histologic event counting, using as our sample use-case the counting of lymphocytes in serous ovarian carcinoma specimens. Samples were prepared as a tissue microarray (TMA) consisting of 618 cases, with clinical follow-up. Caliper’s inForm™ image analysis software automatically detected regions of tumor, and counted TILs in tumor regions, thus determining TIL density. Automated results were compared with visual assessments. Kaplan-Meier survival curves were generated for both automated and visual scores.
60
40
20
0 0
1
2
0
Visual TIL Score
Figure 2 Correlation of visual scoring and automated scoring yielded an R-value of 0.6455.
Results The machine-learning algorithm determined tumor TIL density for 70% of cores and stromal TIL density for 42%. With triplicate representation of each tumor on the array, 93% of tumors had at least one core informative for intratumoral TILs, and 71% had at least one core informative for stromal TILs. There was a significant strong positive correlation between total visual and machine counts (r = 0.6581, p<0.0001 by Spearman’s nonparametric test). Kaplan-Meier analysis shows equivalent P values (~0.03) for visual and automated 275
APPENDIX 2 Table 1 Automated segmentation results showing that 436 of 618 TMA cores successfully segmented. Errors were due to issues with the tissue, over- or under-segmentation of tissue, and over- or under-segmentation of lymphocytes. Total histospots evaluated
618
Pre-algorithmic failures Spot fell off Unsuitable tissue (e.g., colon or fat only)
37 77
Tissue segmentation failures Tumor interpreted as stroma Stroma interpreted as tumor
26 49
Cell segmentation failures Overdetection of lymphocytes Underdetection of lymphocytes
9 3
Spots successfully segmented
436
CD8+/FoxP3+), enabled by Vectra’s multispectral capability. n TMAs, although useful for research investigations, do not support routine clinical work. Future investigations will involve whole biopsy sections. n These results demonstrate the feasibility of a practical and viable clinical workflow, in which TIL counting is automated by computer and results are reviewed by pathologists to assure data quality.
AUTOMATED QUANTITATIVE IMAGE ANALYSIS OF DIABETIC NEPHROPATHY Introduction Late stage diabetic nephropathy is histologically characterized by either diffuse or nodular expansion of the glomerular matrix. Biochemical studies have provided evidence that the microfibrillar collagen type IV is increased in diabetic nephropathy, which is presumed to correlate with functional impairment of the kidney. Highthroughput image acquisition and analysis along with
scoring approaches. Although correlation between visual and automated scoring was high, automated scoring consistently determined larger numbers than visual. Larger numbers were due primarily to over-splitting of segmented lymphocytes, inclusion of lymphocytes lacking nuclear counterstain which are ignored during visual scores, and inclusion of lymphocytes at the periphery of tumor areas. Additionally, we found that a pathologist’s involvement was essential, to review segmentation results and assure data quality by rejecting data from areas improperly stained, out-of-focus, folded, or otherwise inaccurately segmented.
Conclusions Preliminary results indicate machine scoring can meaningfully capture TIL status of tumors and yield quantitative, normalized feature count using consistent rules. n The prognostic power of the test can be extended by adding labels for lymphocyte phenotyping (e.g., CD3+/ n
Figure 1 Representative multispectral 20x images from PAS and Collagen IV stained (A) negative and (B) positive control rats.
Visual TIL score
Automated Hem-CK-CD3 count > = 20 <20
80 60 40 20 0
p=0.0303 0
20
40 Months
276
60
80
> = 1.5 <1.5
100 Percent survival
Percent survival
100
80 60 40 20 0
p=0.0292 0
20
40
60
80
Months
Figure 3 Kaplan-Meier curves for visual and semi-automated scoring compared with survival. These curves have essentially equivalent P-values.
Appendix 2 immunohistochemical assessment was designed to evaluate the extent and exact morphologic location of Collagen IV deposition at various stages of diabetic glomerulosclerosis (GS). In late stage nephropathy, intrinsic basement membrane components are no longer produced. Instead, massive accumulation of Collagen IV occurs. This application note describes work performed by Dr. Bruce Homer at Pfizer. It demonstrates the capabilities of the Vectra imaging system and inForm image analysis software (Caliper Life Sciences, Inc.) for segmenting nodular glomeruli and quantitating expression of Collagen IV in Steptozotocin (STZ) induced diabetic mice. The study also compares automated results with data generated with stereological software (Stereologer 2000).
Methods Sample: Male Wistar rats were injected with Steptozotocin (STZ) to induce diabetes. n Animals with FBG (3 days post-STZ) > 200 mg/dl were included in the study. n All treatment was initiated 3 days after STZ injection. n FFPE kidney sections were stained with immunohistochemistry for Collagen IV. n
Multispectral Imaging: 144 slides from a diabetic nephropathy study were scanned with Caliper’s Vectra imaging platform. n An algorithm was created in Caliper’s inForm image analysis software to automatically identify the cortex of the kidney. n A threshold was set so that regions containing > 40% of cortex region were selected for high-power 20× imaging and further analysis. n
Image Analysis: 1440 20× multispectral images (10 per kidney) were acquired by Vectra and analyzed with inForm. n Caliper’s inForm image analysis software was used to automatically identify glomerular tufts within the cortical region of each image. n Once the glomeruli were identified, the mesangial matrix within each was segmented. n Mesangial matrix volume was calculated by multiplying the kidney tuft volume by the mean percent mesangial matrix area. n Percent mesangial matrix and mesangial matrix volume determined from the automated platform and from stereological assessment were compared. n
Figure 2 4x mosaic of entire scanned kidney section. Red boxes indicate sections automatically chosen for high-power 20x imaging based on an algorithm created to locate the cortex.
Figure 3 (A) RGB representation of a 20x multispectral image acquired with Vectra imaging system. (B) The same image with tissue and cell segmentation masks created by inForm image analysis software. The software was trained to segment the tissue into glomeruli (red) and other tissue (green). inForm was then able segment the mesangial matrix (green outlines) and output quantitative data from these segmented regions.
277
APPENDIX 2
Figure 4 (A) Correlation between percent mesangial matrix per glomerulus when analyzed with the automated platform versus stereological assessment. (B) Correlation between mesangial matrix volume normalized to renal volume when analyzed with the automated platform versus stereological assessment. (C) Correlation between mesangial matrix volume normalized to tuft volume when analyzed with the automated platform versus stereological assessment.
Results Conclusion Results indicate a significant correlation in the percentage of mesangial matrix volume analyzed by Caliper’s imaging system compared to a stereological analyzer for STZ treated versus non-STZ vehicle samples. n There was a significant correlation in the matrix volume and glomerular tuft volume analyzed by Caliper’s imaging system compared to a stereological analyzer. n An increase in Collagen type IV expression was observed within glomerular tufts following STZ treated samples. n High-throughput imaging with Vectra and automated image analysis with inForm are capable of automating analysis of diabetic nephropathy, thus accelerating studies and reducing manual analysis significantly. n
STUDIES OF RECEPTOR SIGNALING AND MUTATIONS IN ARCHIVAL TISSUE USING TISSUE MICROARRAYS AND MULTISPECTRAL IMAGING Example: Androgen receptor analysis in prostate cancer
Introduction The analysis of immunofluorescence in tissue microarrays offers an efficient and precise method to explore correlations between mutation-driven receptor expression patterns and clinical outcome. Analysis using automated multispectral imaging (with the Caliper Vectra™) enables precise and linear quantitation of receptor, in anatomically appropriate cellular structures, and in archival clinical tissues. Sensitive and precise quantitation is possible even if archival FFPE tissues exhibit strong autofluorescences, and receptor expression levels are significantly less than autofluorescence level. TMAs rather than whole sec278 tions offer substantial increases in throughput for image
acquisition and analysis, consistency of staining among samples, and efficient use of reagents. The example case is the assessment of androgen receptor in prostate cancer. Androgen receptor (AR) has a central role in normal growth, in carcinogenesis, and in progression. Androgens function predominantly through their action on the androgen receptor (AR), which belongs to the nuclear receptor family. In addition, 10–30% of prostate carcinomas treated by anti-androgens acquire somatic point-mutation in the AR gene. These genetic changes lead to AR over-expression and hypersensitivity and promiscuous mutant AR proteins activated by non-androgenic ligands or growth modulators. The AR molecule plays a major part in the regulation of androgenAR complex, and is a potential marker in the prognosis and hormonal responsiveness in PCa. IHC studies have shown variability in the expression of ARs in cancer and the ability to predict clinical progression and survival1.
Methods The data presented in this note was generated by Dr. Wei Huang of the University of Wisconsin. She is investigating the expression of AR as a potential marker to augment Gleason score for PCa staging. This study involved tissue microarrays (TMAs) built from formalin-fixed, paraffin-embedded (FFPE) prostatic adenocarcinoma tissue blocks from 183 patients (174 hormone-naïve, 9 castration resistant) and benign prostatic tissue with five year follow-up information. Immunofluorescence: The TMA sections were immunolabeled with rat anti-AR monoclonal antibody (mAb) (Abcam, 1:200) and mouse anti-e-cadherin monoclonal antibody (Dako, 1:50). AR was detected with the rat-AR mAb and was labeled with AlexaFluor™ 647, while the mouse anti-E-cadherin was labeled with AlexaFluor™ 488. DAPI was added as a nuclear counterstain.
n
Appendix 2 Multispectral Imaging: Multispectral fluorescence images from the center of each duplicate core of the TMAs was automatically acquired using the Caliper Vectra™ imaging platform. A spectral library for unmixing label and autofluorescence signals was generated from single-stain slides n Image Analysis: inForm™ image analysis software was applied to quantify the expression of AR in epithelial cells, in a per-cell and per-cell-compartment basis. inForm’s machine-learning pattern-recognition algorithm was trained by a pathologist to identify epithelial regions automatically. Individual cells within epithelial regions were segmented into associated subcellular compartments (nucleus, cytoplasm, and membrane). Spectrally unmixed AR protein expression level was then extracted from each cell for analysis. n Data Analysis: the AR expression levels were correlated with the Gleason scores (GS 5, 6, 7, 8, and 9), prostatic adenocarcinoma pathological stages (pT2, pT3a, and pT3b), and recurrence status (free, biochemical, and cancer). n
RESULTS Multispectral imaging revealed AR expression in the nucleus, cytoplasm, and membrane compartments of PCa, BPT, and CRecur patients. AR exits in the cytoplasm, bound to heat shock proteins. Upon ligand interaction, the AR homodimerizes, undergoes phosphorylation and translocation to the nucleus, where it binds to androgen response elements and induces transcription of genes. inForm™ analysis comparing AR proteins levels in PCa and BPT patients revealed elevated nuclear expression and lower expression in the cytoplasm. However, AR protein level in all the cellular compartments was lower in the PCa compared to the BPT patients. In addition, analysis of nAR and mAR levels in patients with cancer recurrence (CRecur) was significantly lower compared to recurrence-free (RF) patients. No significant difference in AR levels was observed between RF patients versus biochemical recurrence (BRecur), or between BRecurr and CRecur patients.
279
APPENDIX 2
Figure 1 Images from four example TMA cores. (A) RGB representations of multispectral images. (B) Unmixed DAPI component images. (C) Unmixed androgen component images. (D) Unmixed background and autofluorescence component images.
280
Figure 2 RGB representations from four example TMA cores displaying inForm™ epithelial tissue segmentation and epithelial cell segmentation overlays. (A) The tissue was segmented into three classes: epithelial (the target tissue); stroma; and background. (B) The individual cells inside the target epithelial tissue were segmented into nuclei, cytoplasm, and membrane.
Appendix 2
Subcellular AR Expression 6
Subcellular AR Expression 3.5
(A)
Subcellular AR Expression 3.4
(B)
(C)
3.2 5
3.0
3.0
2.8
4
2.6
2.5 3
nAR_A555 Mean
nAR_A555 Mes
cAR_A555 Mes
1
mAR_A555 Mes
8PT
GS5
GS6
GS7
GS8
GS9
nAR_A555 Mean
2.2
2.0 2
2.4
cAR_A555 Mean
mAR_A555 Mean
1.5 BPT
None
Biochemical
Cancer
cAR_A555 Mean
2.0 1.8
mAR_A555 Mean
BPT
P12
P13a
P13b
Figure 3 Correlation of subcellular androgen receptor expression with: (A) Gleason scores; (B) prostate cancer outcomes in terms of recurrence; and (C) prostate cancer pathological stages. These correlations are made from expression in the nucleus (nAR), cytoplasm (cAR), and membrane (mAR) of epithelial cells. Each graph displays the correlation with benign prostatic tissue (BPT) as a control.
Conclusion Automated multispectral imaging of tissue microarrays allows ability to identify multiple proteins and pathways involved in the progression of prostate cancer. inForm™ image analysis revealed the translocation of AR to the nucleus from the cytoplasm that is consistent with previous studies (Waltering et al., 2009; Qiu et al., 2008). In our study low AR expression was observed in patients with metastatic PCa that is consistent with data reported by Takeda et al. 1996 and Segawa et al., 2001. Although the changes in AR levels in all the three compartments were not significantly correlated with PCa stages or
Gleason Score, they are consistent with other reports. This could be due to the heterogeneity of AR immunostaining within a tumor. Thus, this application demonstrates the ability of automated multispectral imaging system in aiding pathologists, as well as researchers to identify multiple biomarkers involved in the prognosis of prostate cancer. In addition, inForm™ software with its machine-learning abilities is capable of segmenting tissue based on morphological features, performing cell segmentation, and biomarker expression on a per cell basis on large data sets for quantitative analysis.
281
269
APPENDIX 2
INCREASING SENSITIVITY AND ACCURACY OF QUANTITATIVE IMMUNOFLUORESCENCE IN FFPE TISSUE WITH SPECTRAL IMAGING Introduction Understanding cancer and its complexity can be advanced significantly with better tools for measuring proteins in situ in formalin-fixed, paraffin-embedded tissue sections. Separation technologies, such as Western 270
blot, microarrays, and mass spectrometry are used widely, but lose key architecturally-specific signals that reside at the cellular level, and blend signals from multiple cell types. Laser capture micro-dissection attempts to address this issue, but is expensive and laborious to perform. IHC has the significant advantage of retaining tissue architecture and heterogeneity, and presents views of protein expression, but is non-quantitative and variable. Most separation approaches disaggregate tissues, blending signals from many cells and tissues.
Appendix 2
monochrome image, acquired with emission filtering centered on the AlexaFluor™555 emission peak. Three locations in the sample were used as reference points: in cytoplasm; in a nucleus; and off the sample. A signal-tobackground ratio and autofluorescence contribution were calculated for each image from the cytoplasm and off-sample signal counts.
Fluorescence microscopy is becoming increasingly important for these endeavors, compared to chromogenic immunohistochemistry, due to higher multiplexing capability, larger and more linear signal range, and less interference among labels. However, immunofluorescence poses challenges, including the presence and effects of autofluorescence and the inability to distinguish overlapping signals due to cross-talk. Multispectral imaging eliminates these issues through the use of spectral unmixing, which enables isolation of individual biomarker signals, even when signals are substantially overlapping spatially and spectrally, and are obscured by autofluorescence signals. Sensitivity and accuracy can be increased several-fold. The purpose of this application note is to provide an example of the quantitative advantage of multispectral imaging.
Methods A tissue microarray of lung cancer tissue was labeled with a multicolor immunofluorescence cocktail commercially available from Cell Signaling Technology (Pathscan Node Kit #8999). The kit comprises two tubes of mixed reagents, one of primaries and one of conjugated secondaries. Phospho-AKT is labeled with AlexaFluor™ 488, phospho-ERK with AlexaFluor™ 555, and phospho-S6 with AlexaFluor™ 647. DAPI nuclear counterstain is included. Multispectral images were acquired with a Caliper Nuance multispectral imaging system and spectrally unmixed into label and autofluorescence signals. For comparison, conventional immunofluorescence images were acquired with fluorescence emission filters centered on the emission peak of the respective fluorophores. For this analysis, we compared the signals for the phospho-ERK (AlexaFluor™ 555) signal in typical microarray cores, as indicated in the spectrally unmixed component image and in the conventional fluorescence emission
Results Measured signals in the conventional immunofluorescence images were significantly higher than in the spectrally unmixed component images, due to the presence of a high autofluorescence signal present, in addition to the pERK signal. In the spectrally unmixed component image, the autofluorescence is removed, and thus the pERK signal is pure and more accurate. A further benefit of spectral unmixing is the removal of cross-talk from overlapping fluorophores spectra, in multi-label assays. Signal-to-background in the spectrally unmixed images is 50 and in the conventional monochrome image is 3.9, a 13× improvement. More importantly, the data suggest that 34% of the signal measured with conventional epifluorescence was actually autofluorescence or cross-talk from another fluorophores labels ((2991−1959)/2991). 271
APPENDIX 2
Figure 1 Lung tissue labeled with AlexaFluor™ 488, 555, and 647 and a DAPI counterstain. Panel A shows an RGB image of the sample with pAKT immunolabeled with AF488, pERK immunolabeled with AF555, and pS6 immunolabeled with AF647. Panels B through E display these unmixed component images along with the autofluorescence component in Panel F, whose border colors correspond to the pseudocolors used to form the composite image, G. The autofluorescence spectrum is unmixed in the black channel so it is not visible in the unmixed composite image.
Table 1 Signal counts found from the two pERK images in Figure 2. The signal-to-background ratio of each image was calculated by dividing the signal from the cytoplasm by the off-sample background signal Signal (counts)
272
Figure 2 Images of pERK signal generated with conventional fixed filters (above) and with spectral unmixing (below). Signal counts were found from the same three pixel locations in both images, in cytoplasm where pERK is likely expressed, in the nucleus, and a third off the sample to provide a background signal.
Conventional
Multispectral
Cell #1
2991
1959
Cell #2
2521
1614
Stroma
701
34
Signal/Background
3.9
50
Conclusion Spectral imaging enables reliable and accurate assessments of weakly expressing proteins in FFPE sections, even when fluorescence emissions are substantially weaker than background autofluorescence. The approach also automatically corrects for cross-talk between labels in multiplexed assays, thus giving pure signals of biomarkers and greatly increasing the signal-to-noise ratio compared to conventional methods.
Appendix 2
RECEPTOR TRAFFICKING ANALYSIS WITH MULTI-COLOR FLUORESCENCE LABELING AND MULTISPECTRAL IMAGING Example: EGFR Activation
Introduction Analysis of signaling pathway activity in FFPE tissue is essential in oncology research. For example: the epidermal growth factor receptor (EGFR) is a transmembrane receptor tyrosine kinase that belongs to the HER/ErbB protein family. Dysregulation of EGFR and other receptor signaling pathways through activating mutations or gene amplification has been implicated in the pathogenesis of many human carcinomas, leading to extensive clinical study. Immunohistochemical studies have limitations due to the difficulty in interpreting the results, as well as specificity of staining. Multiplexing allows simultaneous detection of multiple biomarkers, revealing specific signaling configurations and tumor heterogeneity. In this example, we demonstrate multiplex labeling of total EGFR epitopes and Phospho-EGFR epitopes to determine percent of activated receptors within tumor-positive cells. The antibodies recognize total EGFR and phosphorylated-EGFR specifically upon activation of tyrosine residues. The phosphorylation EGFR is rapidly endocytosed, and either recycled back to the plasma membrane or targeted for lysosomal degradation. The ratio of total EGFR versus Phospho-EGFR reveals the percentage of receptor being recycled by tumor cells. Thus, Caliper’s Vectra™ multispectral imaging,
automated tumor and cell finding software analysis offers precise detection of individual proteins through spectral unmixing based on their spectral signatures. inForm™ analysis software segments the individual tumor cells and provides quantitation for each cellular compartment.
Methods The tissues used in this analysis are mouse xenografts of lung cancer cell lines. EGFR signaling has been found to be dysregulated in several types of this cancer. Expression and/or activation have been linked to therapeutic response, angiogenesis, and metastasis. Labeling is done with a commercially available kit from Cell Signaling Technologies (Catalog #7967), used to detect simultaneously expression, localization, and activation state of EGFR, and additionally the downstream signaling through Erk1/2.
Figure 1 Multispectral 20x images acquired from multiplexed non-small cell lung carcinoma samples with pEGFR immunolabeled with AlexaFluor™ 488, total EGFR immunolabeled with AlexaFluor™ 555, pERK with AlexaFluor™ 647, and a DAPI counterstain. (A) Amplified H3255 cell line. (B) Amplified Kyse450 cell line. (C) Unamplified H1975 cell line. (D) Unamplified H1703 cell line.
Sample: Four human tumor xenografts of non-small cell lung carcinoma cell lines. n Two amplified: H3255/L858R and Kyse450/wild-type. n Two unamplified: H1975/L858R and H1703/wild-type. n
Immunofluorescence Staining: Cell Signaling Technology PathScan EGF Receptor Activation Multiplex IF Kit #7967. n Primary antibodies: pEGFR; total EGFR; pERK n Secondary antibodies: AlexaFluor™ 488; AlexaFluor™ 555; and AlexaFluor™ 647. n DAPI counterstain. n
Imaging and Analysis: Caliper’s Vectra automated imaging system and inForm image analysis software. 273 n Acquired four 20× multispectral images per sample. n
APPENDIX 2
Figure 2 Image cubes of amplified H3255 cells. (A) RGB representation of multispectral cube. (B) Composite image of unmixed signals with pEGFR represented in green, EGFR represented in red, and co-localization shown in yellow. (C) RGB image overlayed with an inForm tissue segmentation map, where red areas are tumor, green is stroma, and the background is in blue. (D) RGB image overlayed with an inForm cell segmentation map, where each cell in the tumor area has been segmented into nucleus, cytoplasm, and membrane compartments.
Cytoplasm
Membrane
3.5
7
3
6
2.5
5
2
4
1.5
3
1
2
0.5
1
0
H1703
H1975
H3255
Kyse
0
EGFR pEGFR pErk
H1703
H1975
H3255
Kyse
Figure 3 Comparison of total EGFR, pEGFR, and pERK expression levels in different cellular compartments for each xenograft cell line.
Automated machine-learning-based image analysis to: n Detect areas of tumor n Segment the cells into subcellular compartments n Extract IF signals on a per-cell basis
n
Results Evaluation of individual biomarkers following multispectral unmixing revealed changes in pEGFR expression compared to total EGFR in xenograft expression specific tyrosine kinase mutations. Since cytoplasmic pEGFR represents an activated form of surface EGFR, it was not observed within the proximity of total EGFR. Data analysis revealed tyrosine kinase EGFR mutant xenografts showed lower expression of pEGFR, while wildtype/Kyse450 xenograft samples showed increase in total EGFR, pERK, and a decrease in cytoplasmic pEGFR expression. The data demonstrates that the ratio of pEGFR versus total EGFR is consistent with cytoplasmic to membrane receptor recycling. Activation of pERK did not correlate with EGFR expression, which is not unexpected as it 274 could be activated by RAS/MEK signaling pathway.
Conclusion Multiplexing provides individual biomarker signals whose expression correlates with dynamic changes in receptor trafficking, activation of specific signaling pathways, and modifications in tumor cell morphology. In conclusion, high-throughput imaging platform along with image analysis software and a panel of phosphorylationspecific antibodies could reveal molecular subtypes of tumors for effective personalized therapy.
AUTOMATED IMAGE ANALYSIS OF TUMOR-INFILTRATING LYMPHOCYTES Introduction Tumors contain variable numbers of CD3+, CD4+, CD8+ and FoxP3+ lymphocytes referred to as tumorinfiltrating lymphocytes (TILs). Initially, TILs were thought to reflect the origin of cancer at sites of chronic inflammation. Recent studies revealed a relationship between the intensity as well as the ratio of CD3+, CD4+ CD8+, and FoxP3 lymphocytes with prognostic outcome
Appendix 2
Figure 1 (A) Multispectral 20x images from four example cores acquired from a TMA of ovarian cancer stained with hematoxylin, Vector Red labeling cytokeratin, and DAB labeling anti-CD3 antibody. (B) The same four cores with tissue and cell segmentation maps overlaid with red areas representing tumor, blue areas representing other tissue, and green outlines indicating DAB+ TILs.
Methods Sample: Tissue microarray (TMA) containing 618 ovarian cancer samples. n The TMA samples were stained with anti-CD3 antibody to identify T-lymphocytes. n The samples were further stained with epithelial cell specific marker (cytokeratin) to assist in automated segmentation of tumor and stroma. n
Automated Scoring: The TMA slides were scanned using Caliper’s Vectra™ multispectral imaging system. The scanned images were processed using Caliper’s inForm™. A machine-learning algorithm was trained to segment tumor from stroma and identify CD3 cells labeled with DAB, indicating T-lymphocytes. n Density of TILs within the tumor areas was calculated. n
Visual Scoring: A pathologist rated lymphocyte density. A semi-quantitative scale of 0 to 3 was assigned to each core.
n n
Analysis: Automated counts were compared to visual scores. Manual and automated scoring were compared with survival.
n n
80
Hematoxylin-CK-CD3 automated count
of different solid tumors. Potentially, tests that characterize TIL phenotype, location, and density in tumor and surrounding stroma could aid the selection of immunotherapy and define targets for individualized treatment. However, TIL assessment today, based on human visual perception, is prone to inter- and intra-observer variability, due to complicated tissue architectures, ambiguous histology revealed through hematoxylin counterstain, and other human factors. Also, visual TIL assessment is tedious and time consuming. In this study, performed in collaboration with Dr.Michael Feldman and Dr. Ian Hagemann at University of Pennsylvania, we investigate computer-aided histologic event counting, using as our sample use-case the counting of lymphocytes in serous ovarian carcinoma specimens. Samples were prepared as a tissue microarray (TMA) consisting of 618 cases, with clinical follow-up. Caliper’s inForm™ image analysis software automatically detected regions of tumor, and counted TILs in tumor regions, thus determining TIL density. Automated results were compared with visual assessments. Kaplan-Meier survival curves were generated for both automated and visual scores.
60
40
20
0 0
1
2
0
Visual TIL Score
Figure 2 Correlation of visual scoring and automated scoring yielded an R-value of 0.6455.
Results The machine-learning algorithm determined tumor TIL density for 70% of cores and stromal TIL density for 42%. With triplicate representation of each tumor on the array, 93% of tumors had at least one core informative for intratumoral TILs, and 71% had at least one core informative for stromal TILs. There was a significant strong positive correlation between total visual and machine counts (r = 0.6581, p<0.0001 by Spearman’s nonparametric test). Kaplan-Meier analysis shows equivalent P values (~0.03) for visual and automated 275
APPENDIX 2 Table 1 Automated segmentation results showing that 436 of 618 TMA cores successfully segmented. Errors were due to issues with the tissue, over- or under-segmentation of tissue, and over- or under-segmentation of lymphocytes. Total histospots evaluated
618
Pre-algorithmic failures Spot fell off Unsuitable tissue (e.g., colon or fat only)
37 77
Tissue segmentation failures Tumor interpreted as stroma Stroma interpreted as tumor
26 49
Cell segmentation failures Overdetection of lymphocytes Underdetection of lymphocytes
9 3
Spots successfully segmented
436
CD8+/FoxP3+), enabled by Vectra’s multispectral capability. n TMAs, although useful for research investigations, do not support routine clinical work. Future investigations will involve whole biopsy sections. n These results demonstrate the feasibility of a practical and viable clinical workflow, in which TIL counting is automated by computer and results are reviewed by pathologists to assure data quality.
AUTOMATED QUANTITATIVE IMAGE ANALYSIS OF DIABETIC NEPHROPATHY Introduction Late stage diabetic nephropathy is histologically characterized by either diffuse or nodular expansion of the glomerular matrix. Biochemical studies have provided evidence that the microfibrillar collagen type IV is increased in diabetic nephropathy, which is presumed to correlate with functional impairment of the kidney. Highthroughput image acquisition and analysis along with
scoring approaches. Although correlation between visual and automated scoring was high, automated scoring consistently determined larger numbers than visual. Larger numbers were due primarily to over-splitting of segmented lymphocytes, inclusion of lymphocytes lacking nuclear counterstain which are ignored during visual scores, and inclusion of lymphocytes at the periphery of tumor areas. Additionally, we found that a pathologist’s involvement was essential, to review segmentation results and assure data quality by rejecting data from areas improperly stained, out-of-focus, folded, or otherwise inaccurately segmented.
Conclusions Preliminary results indicate machine scoring can meaningfully capture TIL status of tumors and yield quantitative, normalized feature count using consistent rules. n The prognostic power of the test can be extended by adding labels for lymphocyte phenotyping (e.g., CD3+/ n
Figure 1 Representative multispectral 20x images from PAS and Collagen IV stained (A) negative and (B) positive control rats.
Visual TIL score
Automated Hem-CK-CD3 count > = 20 <20
80 60 40 20 0
p=0.0303 0
20
40 Months
276
60
80
> = 1.5 <1.5
100 Percent survival
Percent survival
100
80 60 40 20 0
p=0.0292 0
20
40
60
80
Months
Figure 3 Kaplan-Meier curves for visual and semi-automated scoring compared with survival. These curves have essentially equivalent P-values.
Appendix 2 immunohistochemical assessment was designed to evaluate the extent and exact morphologic location of Collagen IV deposition at various stages of diabetic glomerulosclerosis (GS). In late stage nephropathy, intrinsic basement membrane components are no longer produced. Instead, massive accumulation of Collagen IV occurs. This application note describes work performed by Dr. Bruce Homer at Pfizer. It demonstrates the capabilities of the Vectra imaging system and inForm image analysis software (Caliper Life Sciences, Inc.) for segmenting nodular glomeruli and quantitating expression of Collagen IV in Steptozotocin (STZ) induced diabetic mice. The study also compares automated results with data generated with stereological software (Stereologer 2000).
Methods Sample: Male Wistar rats were injected with Steptozotocin (STZ) to induce diabetes. n Animals with FBG (3 days post-STZ) > 200 mg/dl were included in the study. n All treatment was initiated 3 days after STZ injection. n FFPE kidney sections were stained with immunohistochemistry for Collagen IV. n
Multispectral Imaging: 144 slides from a diabetic nephropathy study were scanned with Caliper’s Vectra imaging platform. n An algorithm was created in Caliper’s inForm image analysis software to automatically identify the cortex of the kidney. n A threshold was set so that regions containing > 40% of cortex region were selected for high-power 20× imaging and further analysis. n
Image Analysis: 1440 20× multispectral images (10 per kidney) were acquired by Vectra and analyzed with inForm. n Caliper’s inForm image analysis software was used to automatically identify glomerular tufts within the cortical region of each image. n Once the glomeruli were identified, the mesangial matrix within each was segmented. n Mesangial matrix volume was calculated by multiplying the kidney tuft volume by the mean percent mesangial matrix area. n Percent mesangial matrix and mesangial matrix volume determined from the automated platform and from stereological assessment were compared. n
Figure 2 4x mosaic of entire scanned kidney section. Red boxes indicate sections automatically chosen for high-power 20x imaging based on an algorithm created to locate the cortex.
Figure 3 (A) RGB representation of a 20x multispectral image acquired with Vectra imaging system. (B) The same image with tissue and cell segmentation masks created by inForm image analysis software. The software was trained to segment the tissue into glomeruli (red) and other tissue (green). inForm was then able segment the mesangial matrix (green outlines) and output quantitative data from these segmented regions.
277
APPENDIX 2
Figure 4 (A) Correlation between percent mesangial matrix per glomerulus when analyzed with the automated platform versus stereological assessment. (B) Correlation between mesangial matrix volume normalized to renal volume when analyzed with the automated platform versus stereological assessment. (C) Correlation between mesangial matrix volume normalized to tuft volume when analyzed with the automated platform versus stereological assessment.
Results Conclusion Results indicate a significant correlation in the percentage of mesangial matrix volume analyzed by Caliper’s imaging system compared to a stereological analyzer for STZ treated versus non-STZ vehicle samples. n There was a significant correlation in the matrix volume and glomerular tuft volume analyzed by Caliper’s imaging system compared to a stereological analyzer. n An increase in Collagen type IV expression was observed within glomerular tufts following STZ treated samples. n High-throughput imaging with Vectra and automated image analysis with inForm are capable of automating analysis of diabetic nephropathy, thus accelerating studies and reducing manual analysis significantly. n
STUDIES OF RECEPTOR SIGNALING AND MUTATIONS IN ARCHIVAL TISSUE USING TISSUE MICROARRAYS AND MULTISPECTRAL IMAGING Example: Androgen receptor analysis in prostate cancer
Introduction The analysis of immunofluorescence in tissue microarrays offers an efficient and precise method to explore correlations between mutation-driven receptor expression patterns and clinical outcome. Analysis using automated multispectral imaging (with the Caliper Vectra™) enables precise and linear quantitation of receptor, in anatomically appropriate cellular structures, and in archival clinical tissues. Sensitive and precise quantitation is possible even if archival FFPE tissues exhibit strong autofluorescences, and receptor expression levels are significantly less than autofluorescence level. TMAs rather than whole sec278 tions offer substantial increases in throughput for image
acquisition and analysis, consistency of staining among samples, and efficient use of reagents. The example case is the assessment of androgen receptor in prostate cancer. Androgen receptor (AR) has a central role in normal growth, in carcinogenesis, and in progression. Androgens function predominantly through their action on the androgen receptor (AR), which belongs to the nuclear receptor family. In addition, 10–30% of prostate carcinomas treated by anti-androgens acquire somatic point-mutation in the AR gene. These genetic changes lead to AR over-expression and hypersensitivity and promiscuous mutant AR proteins activated by non-androgenic ligands or growth modulators. The AR molecule plays a major part in the regulation of androgenAR complex, and is a potential marker in the prognosis and hormonal responsiveness in PCa. IHC studies have shown variability in the expression of ARs in cancer and the ability to predict clinical progression and survival1.
Methods The data presented in this note was generated by Dr. Wei Huang of the University of Wisconsin. She is investigating the expression of AR as a potential marker to augment Gleason score for PCa staging. This study involved tissue microarrays (TMAs) built from formalin-fixed, paraffin-embedded (FFPE) prostatic adenocarcinoma tissue blocks from 183 patients (174 hormone-naïve, 9 castration resistant) and benign prostatic tissue with five year follow-up information. Immunofluorescence: The TMA sections were immunolabeled with rat anti-AR monoclonal antibody (mAb) (Abcam, 1:200) and mouse anti-e-cadherin monoclonal antibody (Dako, 1:50). AR was detected with the rat-AR mAb and was labeled with AlexaFluor™ 647, while the mouse anti-E-cadherin was labeled with AlexaFluor™ 488. DAPI was added as a nuclear counterstain.
n
Appendix 2 Multispectral Imaging: Multispectral fluorescence images from the center of each duplicate core of the TMAs was automatically acquired using the Caliper Vectra™ imaging platform. A spectral library for unmixing label and autofluorescence signals was generated from single-stain slides n Image Analysis: inForm™ image analysis software was applied to quantify the expression of AR in epithelial cells, in a per-cell and per-cell-compartment basis. inForm’s machine-learning pattern-recognition algorithm was trained by a pathologist to identify epithelial regions automatically. Individual cells within epithelial regions were segmented into associated subcellular compartments (nucleus, cytoplasm, and membrane). Spectrally unmixed AR protein expression level was then extracted from each cell for analysis. n Data Analysis: the AR expression levels were correlated with the Gleason scores (GS 5, 6, 7, 8, and 9), prostatic adenocarcinoma pathological stages (pT2, pT3a, and pT3b), and recurrence status (free, biochemical, and cancer). n
RESULTS Multispectral imaging revealed AR expression in the nucleus, cytoplasm, and membrane compartments of PCa, BPT, and CRecur patients. AR exits in the cytoplasm, bound to heat shock proteins. Upon ligand interaction, the AR homodimerizes, undergoes phosphorylation and translocation to the nucleus, where it binds to androgen response elements and induces transcription of genes. inForm™ analysis comparing AR proteins levels in PCa and BPT patients revealed elevated nuclear expression and lower expression in the cytoplasm. However, AR protein level in all the cellular compartments was lower in the PCa compared to the BPT patients. In addition, analysis of nAR and mAR levels in patients with cancer recurrence (CRecur) was significantly lower compared to recurrence-free (RF) patients. No significant difference in AR levels was observed between RF patients versus biochemical recurrence (BRecur), or between BRecurr and CRecur patients.
279
APPENDIX 2
Figure 1 Images from four example TMA cores. (A) RGB representations of multispectral images. (B) Unmixed DAPI component images. (C) Unmixed androgen component images. (D) Unmixed background and autofluorescence component images.
280
Figure 2 RGB representations from four example TMA cores displaying inForm™ epithelial tissue segmentation and epithelial cell segmentation overlays. (A) The tissue was segmented into three classes: epithelial (the target tissue); stroma; and background. (B) The individual cells inside the target epithelial tissue were segmented into nuclei, cytoplasm, and membrane.
Appendix 2
Subcellular AR Expression 6
Subcellular AR Expression 3.5
(A)
Subcellular AR Expression 3.4
(B)
(C)
3.2 5
3.0
3.0
2.8
4
2.6
2.5 3
nAR_A555 Mean
nAR_A555 Mes
cAR_A555 Mes
1
mAR_A555 Mes
8PT
GS5
GS6
GS7
GS8
GS9
nAR_A555 Mean
2.2
2.0 2
2.4
cAR_A555 Mean
mAR_A555 Mean
1.5 BPT
None
Biochemical
Cancer
cAR_A555 Mean
2.0 1.8
mAR_A555 Mean
BPT
P12
P13a
P13b
Figure 3 Correlation of subcellular androgen receptor expression with: (A) Gleason scores; (B) prostate cancer outcomes in terms of recurrence; and (C) prostate cancer pathological stages. These correlations are made from expression in the nucleus (nAR), cytoplasm (cAR), and membrane (mAR) of epithelial cells. Each graph displays the correlation with benign prostatic tissue (BPT) as a control.
Conclusion Automated multispectral imaging of tissue microarrays allows ability to identify multiple proteins and pathways involved in the progression of prostate cancer. inForm™ image analysis revealed the translocation of AR to the nucleus from the cytoplasm that is consistent with previous studies (Waltering et al., 2009; Qiu et al., 2008). In our study low AR expression was observed in patients with metastatic PCa that is consistent with data reported by Takeda et al. 1996 and Segawa et al., 2001. Although the changes in AR levels in all the three compartments were not significantly correlated with PCa stages or
Gleason Score, they are consistent with other reports. This could be due to the heterogeneity of AR immunostaining within a tumor. Thus, this application demonstrates the ability of automated multispectral imaging system in aiding pathologists, as well as researchers to identify multiple biomarkers involved in the prognosis of prostate cancer. In addition, inForm™ software with its machine-learning abilities is capable of segmenting tissue based on morphological features, performing cell segmentation, and biomarker expression on a per cell basis on large data sets for quantitative analysis.
281