Quantitation of spatial and temporal variability of biomarkers for Barrett's Esophagus

Summary

Chemoprevention and risk-stratification studies in Barrett's esophagus (BE) rely on biomarkers but the variability in their temporal and spatial expression is unknown. If such variability exists, it will impact sampling techniques and sample size calculations. Specimens from three levels of biopsies over two serial endoscopies in nondysplastic BE patients were analyzed for aneuploidy, proliferation markers (Ki67, Mcm2), and cell cycle markers (cyclin A and cyclin D1). A modification of the image cytometry technique, where cytokeratin staining automatically distinguished epithelial and stromal cells, measured aneuploidy on whole tissue sections. Other biomarkers were studied by immunohistochemistry. Coefficient of variability (SD/mean) was calculated; a value <10% indicated low variability. A total of 120 specimens (20 subjects each with three biopsy levels at two time points) from nondysplastic BE patients (71 ± 8.8 years, all Caucasian, 90% males, C5.1M7.5 ± 3.4 cm) were analyzed. The mean interval between endoscopies was 32.8 ± 8.4 months. Aneuploidy had a spatial variability of 6.8% at visit 1 (mean diploid index: 1.1 ± 0.09) and 7.9% at visit 2 (mean diploid index: 1.1 ± 0.06) and a temporal variability of 7.0–8.1% for the three levels. For other biomarkers, the spatial variability ranged from ∼5 to 30% at visit 1 and 11–92% at visit 2 and the temporal variability ranged from 0 to 77%. To conclude, of all the biomarkers, only aneuploidy had both spatial and temporal variability of <10%. Spatial and temporal variability were biomarker dependent and could be as high as 90% even without progression. These data will be useful to design chemoprevention and risk-stratification studies in BE.

ABBREVIATIONS

BE

Barrett's esophagus

DI

diploid index

INTRODUCTION

Esophageal adenocarcinoma (EAC) continues to be a major public health problem because of rapid increase and poor survival.¹ Barrett's esophagus (BE) is the premalignant lesion for the majority of patients with EAC with a 30–40-fold increase in risk compared to the general population.² Precision medicine is rapidly becoming a reality³ and is highly relevant to the management of BE where the majority of patients will not progress to EAC² but those who do have survival rates of < 20%.² Precision medicine heavily relies on accurate biomarker mapping for individualized recommendations.⁴ Biomarkers have been extensively studied in Barrett's esophagus for risk-stratification^5,6 and as endpoints in chemoprevention trials⁷ but most studies have not systematically quantified biomarker variability in esophageal space and time.⁸ A recent seminal article strongly argues to track BE biomarkers in esophageal space and in time for more robust risk-stratification.⁸ This is important because quantifying biomarker variability in BE is necessary to determine the robustness of biomarkers for risk-stratification and chemoprevention. The degree of variability will affect the sampling techniques in BE. If an expression of a biomarker is highly variable across the BE length, then the biomarker may need to be sampled at several locations within BE and an average or a worst value calculated for risk determination; or wide area sampling may be required, perhaps by novel cytology techniques. On the other hand, if a biomarker has minimal variability, only a few BE biopsies may need to be assayed, thereby, reducing costs.

We conducted a proof of principle study to methodically quantify the “natural” variability of time-tested biomarkers in BE, in absence of progression. Specifically, we wanted to determine how a biomarker expression varied along the length of BE (spatial variability) and changed over time (temporal variability) and whether variability was biomarker dependent. We chose different classes of biomarkers for this study based on their importance in risk-stratification and chemoprevention trials. We evaluated markers for DNA content (aneuploidy),^9–12 proliferation (Ki67^7,13,14 and Mcm2^15,16),^7,13 and cell cycle abnormalities (cyclin D1^17,18 and cyclin A^19,20). Aneuploidy i.e. abnormal DNA content has stood the test of time as a biomarker for Barrett's esophagus. Although conventionally measured by flow cytometry, image cytometry is more practical and has been shown to be reliable.^21,22 Typically, image cytometry relies on manually identifying epithelial and stromal cells for aneuploidy calculations. We improved upon this approach by using a simple step of cytokeratin staining to automatically identify epithelial and stromal cells for aneuploidy measurements. This also allowed global aneuploidy measurements of whole tissue sections, thus, minimizing selection bias. Proliferation and cell cycle markers have been shown to be useful biomarkers for BE neoplasia and have been widely used in BE chemoprevention trials.⁷

Aims: 1) To assess variability of aneuploidy, proliferation (Ki67 and MCM2), and cell cycle (cyclin D1 and cyclin A) markers within the BE segment and 2) To assess variability of the same biomarker panel over two-time periods (visits 1 and 2).

METHODS

Study Cohort and sample procurement

BE patients were selected from a well-established patient database.^23–25 Those BE patients who underwent at least two endoscopies one year apart were included. The specific inclusion criteria were a) patients aged 18–75 years, b) presence of intestinal metaplasia without dysplasia, to control for the influence of dysplasia on biomarker variability, c) BE segment ≥4 cm in length to allow at least three sets of biopsies from different levels within BE, and d) at least 6 months of acid suppression. Exclusion criteria included a history of esophageal surgery and presence of erosive esophagitis in the BE segment at the time of biopsies. Standard biopsy protocols of four-quadrant biopsies every 1–2 cm along the length of the esophagus were followed. Three levels of BE segment were chosen for analysis: distal (closest to the gastroesophageal junction), proximal (closest to the squamocolumnar junction), and intermediate (any biopsy level between distal and proximal). The protocol was approved by the local Institutional Review Board. All patients signed written informed consent.

Sample processing

From formalin-fixed tissues, consecutive 4 μm thin sections were prepared; the first section was used for histologic evaluation (H&E) and adjacent sections were used for immunohistochemistry and aneuploidy analysis as described below. All biopsies underwent a dedicated pathologic review by an experienced GI pathologist to rule out dysplasia according to previously established criteria.²⁶ All biopsies with dysplasia were reviewed by a second GI pathologist. Only biopsies that had intestinal metaplasia without dysplasia were included.

Whole section aneuploidy calculations

We devised a novel approach to automate calculations of whole section aneuploidy by image analysis using a laser scanning cytometer (iCYS, CompuCyte Corporation, Cambridge, MA).^27,28 The validated approach of image cytometry for aneuploidy measurements in Barrett's esophagus^21,22 was modified by introducing an additional step of cytokeratin staining to distinguish epithelial from stromal cells. Typically, the DNA content of epithelial cells is normalized against that of stromal cells by manually sampling a few hundred epithelial and stromal cells. Because the cytometer includes multiple lasers, it allowed us to combine stains for DNA content and cellular structures into a single analysis. We were able to automatically detect fluorescently labeled cytokeratin antibody with one laser and the DAPI nuclear stain intensity with the other laser. After staining, contours were drawn to outline the cells based on the fluorescence intensity (Fig. 1). Once the contours were drawn to identify individual cells, the automated software provided high-resolution histograms. Previous studies have shown that the DNA content can be precisely measured with a coefficient of variation of 2%.^34,35

Fig. 1.

Panel A demonstrates automated selection by the cytometer of the tissue sections on a biopsy slide from a patient with Barrett's esophagus, outlined in yellow. Panel B shows the laser scanning images of the nuclei using DAPI staining and panel C shows the scanned outlines of cell membranes using cytokeratin antibody. The lowest panels show representative histograms of whole section aneuploidy measurements in biopsies from a patient with intestinal metaplasia (panel D) and in biopsies from a patient with high-grade dysplasia (panel E). Panel E shows a significantly high proportion of cells with DNA content > 2N (rightward shift) consistent with a diagnosis of high-grade dysplasia.

Briefly, 4 μm sections were deparaffinized and were stained with pan Cytokeratin mouse monoclonal antibody (Biocare Medical) at 1:50 for 60 minutes. Secondary antibody was PE Goat anti-Mouse Ig. Nuclei were stained with DAPI at 5 μg/ml for 10 minutes. Epithelial cells were cytokeratin-positive and stromal cells were cytokeratin-negative. For analysis, the manufacturer's protocol was followed. The DNA content was measured based on the intensity of DAPI nuclear staining. Two cell populations were identified: DAPI+, cytokeratin+ (epithelial cells) and DAPI+, cytokeratin- (stromal cells). The diploid index (DI) was expressed as the ratio of DAPI intensity between the whole populations of epithelial and stromal cells, thus minimizing selection bias. Histograms of diploid index and cell counts were generated (Fig. 1).

In addition to the patients with nondysplastic BE described earlier, we also analyzed specimens with high-grade dysplasia in BE as positive controls to test the validity of our modified cytokeratin-assisted image cytometry technique.

Immunohistochemical staining

Paraffin blocks were sectioned at 4 μm, mounted on Superfrost + slides and baked in a 65°C oven for 1 hour. After deparaffinization, heat-treated slides were labeled with MCM2 1:200, Cyclin D1 1:100, Cyclin A 1:50 (Lab Vision, Fremont, CA) and Ki-67 (MIB-1) 1:200 (Dako, Carpinteria, CA) in citrate buffer pH 6.0 for 5 minutes using the Biocare Medical Decloaking Chamber (Biocare Medical, Concord, CA). After pressure returned to zero, the slide jars were removed from the Chamber, cooled for 10 minutes, and then transferred to tris-buffered saline with Tween. Staining was performed using the Dako Autostainer per company protocol. Endogenous peroxidase was blocked using 3% hydrogen peroxide for 10 minutes, rinsed, followed by primary antibody incubation and detection. All markers were detected using Envision + HRP and DAB+ chromogen per manufacturer's protocol (Dako, Carpinteria, CA). Slides were counterstained with Mayer's Hematoxylin, dehydrated, and coverslipped using permanent mounting media. Normal juvenile tonsil was used as a positive control for all antibodies. Negative controls were included by omission of the primary antibody. The expression of Ki67, Mcm2, cyclin D1, and cyclin A was calculated by a gastrointestinal pathologist with expertise in immunohistochemistry.

Data collection and statistical analysis

Aneuploidy was expressed as DI. For other markers such as proliferation markers, Ki67 and Mcm2, and cell cycle markers, cyclin A and cyclin D1, an expert GI pathologist scanned the entire slide to determine the frequency of staining and a labeling index was calculated—the number of labeled cell nuclei (i.e. numerator) was divided by the total number of cells (labeled and unlabeled) (i.e. denominator) and expressed as a percentage. Only diffuse nuclear staining was considered positive for all of the markers. The differences in expression at different levels of Barrett's esophagus (spatial variability) and over time (temporal variability) were analyzed by calculating the coefficient of variation (CV)(standard deviation/mean). A priori, we stringently defined low variability as CV < 10% as a variation of 10% or less is considered acceptable for biomarker measurements.

RESULTS

Study population

A total of 120 biopsy specimens that sampled distal-, mid-, and proximal-BE from 20 patients at two visits were selected from an established database of patients with BE. All were Caucasian males with a mean age of 71 ± 9 years and a mean BMI of 29 ± 6.0. The mean BE length was 7.4 ± 3.4 cm (Prague C5.1M7.4) and the mean hiatus hernia length was 3.9 ± 1.7 cm. All were on stable acid suppression for at least 6 months and none of the patients had evidence of erosive esophagitis on endoscopy. The mean interval between the initial (visit 1) and the follow up (visit 2) endoscopy was 32.8 ± 8.4 months consistent with the prevailing guidelines on BE surveillance. None of the patients developed dysplasia over a mean follow up of 6.4 ± 2.2 years.

Spatial and temporal variability of biomarkers

Aneuploidy

High resolution histograms for aneuploidy (Fig. 1) could be obtained from 112/120 (93%) specimens. The histograms clearly separated the specimens with intestinal metaplasia (n = 5) from those with high-grade dysplasia in BE (positive controls, n = 5) (Fig. 1). The mean DI in the biopsies with intestinal metaplasia was 1.08 ± 0.07 compared to 2.28 ± 0.30 in those with HGD (P = 0.007). These results suggest that our modification of the image cytometry is effective.

After initial validation, biomarker variability was measured on 120 independent specimens as described earlier. The overall mean DI was 1.1 ± 0.09 for visit 1 and 1.1 ± 0.06 for visit 2. The spatial variability ranged from 6.8 to 7.9% and the temporal variability ranged from 7 to 8.1% (Table 1).

Table 1.

Spatial and temporal variability of Aneuploidy

	Visit 1	Visit 2	CV for Visit 1 vs. Visit 2
Distal	1.13 ± 0.17	1.07 ± 0.11	8.1 ± 7.6%
Mid	1.07 ± 0.07	1.1 ± 0.1	7.8 ± 5.0%
Proximal	1.06 ± 0.1	1.06 ± 0.07	7.0 ± 5.9%
P value	NS	NS	NS
CV*	7.9 ± 7.8%	6.8 ± 4.6%	–

^*coefficient of variation between levels

Proliferation and cell cycle markers These markers were studied by immunohistochemistry (Fig. 2). Ki67 had a spatial variation of ∼30% at visit 1 and 25% at visit 2. The spatial variation for other markers ranged from ∼5 to 92%. Of the two proliferation markers, MCM2 had lower spatial variation and of the two cell cycle markers, cyclin A had lower spatial variation (Table 2). The temporal variability was calculated over a mean interval of ∼36 months. The temporal variation in biomarker expression between visits 1and 2 ranged from 0 to 77% for various biomarkers (Table 2). The temporal variation was highest at the distal level, closest to the gastroesophageal junction, for three (Ki67, MCM2, and cyclin D1) of the four markers with a CV ranging from 60 to 70%.

Fig. 2.

The above figure demonstrates H&E and parallel IHC sections from a representative patient with Barrett's esophagus. Proximal (adjacent to the squamocolumnar junction), distal (adjacent to the gastro esophageal Junction), and mid (a level in between) refers to the level of biopsies within the BE segment. Ki67 and MCM2 were used as markers of proliferation and cyclin D and cyclin A as markers for cell cycle abnormalities.

Table 2.

Spatial and temporal variability of proliferation and cell cycle markers

	Ki67			MCM2			Cyclin D1			Cyclin A
Level	Visit 1	Visit 2	CV Visit 1	Visit 1	Visit 2	CV Visit 1	Visit 1	Visit 2	CV Visit 1	Visit 1	Visit 2	CV Visit 1
			vs. Visit 2			vs. visit 2			vs. visit 2			vs. visit 2
Distal	6.8%	2.3%	77.7%	2.0%	5.2%	71.7%	1.2%	3.0%	69.7%	1.1%	1.1%	0%
Mid	4.0%	3.8%	5.3%	2.1%	4.5%	61.3%	0.8%	0.9%	11.9%	1.2%	1.7%	32.6%
Proximal	4.4%	3.5%	22.3%	2.2%	4.2%	54.3%	1.0%	0.5%	57.2%	1.3%	1.6%	20.5%
P value	NS	0.01	–	NS	NS	–	NS	0.03	–	NS	NS	–
CV*	29.9%	25%		4.7%	11%		20%	91.5%		8.3%	21.9%

^*CV, coefficient of variation calculated between levels for each visit

Summarizing the results for both spatial and temporal variation (in absence of progression), besides aneuploidy, none of the other markers had both spatial and temporal variability of <10% (Tables 1 and 2).

DISCUSSION

We systematically evaluated expression of multiple biomarkers (aneuploidy, proliferation markers, and cell cycle markers) at three levels within BE (distal, mid, and proximal) to quantitate spatial variability and across two time points, approximately 3 years apart, to quantitate temporal variability. None of the patients progressed histologically over the duration of follow-up. We selected aneuploidy because aneuploidy is widely recognized as an important marker for Barrett's progression.^9–12 We decided to select proliferation and cell cycle markers as they have been successfully used in prior chemoprevention trials^18,29,30 and have been shown to be important risk biomarkers in BE.^14,15,17 We also selected two markers for each of the cancer promoting processes (proliferation and cell cycle abnormalities) to test if the variability was similar for different biomarkers evaluating the same process. We made several novel observations. First, aneuploidy could be calculated in an automated manner from the entire tissue section by using whole-section-cytokeratin-assisted image cytometry, an improvement over the validated and practical technique of image cytometry.^21,22 This can avoid the need for manual calculations for image cytometry. Second, spatial and temporal variability were biomarker dependent and could be as high as 90%, even without progression. Third, the variability had a different degree even for the biomarkers evaluating the same biological process. For instance, Ki67 and Mcm2 both evaluate proliferation but had different degrees of variability. Lastly, of all the biomarkers, only aneuploidy had both spatial and temporal variability of <10%.

Our results argue that if aneuploidy were to be used for BE chemoprevention or progression trials, then a few representative biopsies may suffice but if other markers such as Ki67, etc. were to be used, then an average value of expression in multiple systematically obtained biopsies may need to be calculated to reflect the true expression in BE. Furthermore, every biomarker in BE needs to be independently examined for quantification of variability prior to use as a chemoprevention endpoint or in a clinical assay.

Pepe et al., in their review of phases of biomarker development emphasized the importance of defining biomarker variability.³¹ Degree of variability in a biomarker expression could determine its usefulness for clinical application.³² Chemoprevention studies widely use biomarkers as surrogate endpoints to perform rapid, short-term trials to discover efficacious chemopreventive agents to be further tested in phase 3 studies. However, inspite of being the central endpoints, biomarkers have not been widely examined for their biologic variability in a disease process. Chemoprevention continues to be an important goal in BE,⁷ the only known premalignant condition for the fast increasing esophageal adenocarcinoma. The current study was designed to be a proof of principle study to evaluate patients with BE for variability in commonly used biomarkers for chemoprevention trials and to determine if this variability is biomarker dependent.

We have devised a new technique that can automatically calculate whole tissue section aneuploidy in Barrett's esophagus. Aneuploidy has stood the test of time as a strong predictor of progression in Barrett's esophagus.^9–12 Conventionally, aneuploidy has been measured by flow cytometry that requires special methods for sample collection and processing and is not practical.³³ Image cytometry to calculate aneuploidy has been a significant improvement because it can be performed on fixed samples and can be applied in clinical practice.^21,22 In head-to-head comparisons with flow cytometry, the results of image cytometry were 93% identical.³⁴ We further improved upon the technique of image cytometry by adding a step of epithelial staining to differentiate between the epithelial and stromal cells. This has multiple advantages: 1) it obviates the need for manually counting few hundred epithelial and stromal cells as is typically done, 2) it samples the whole tissue section for presence of aneuploidy, thus removing selection bias, and 3) it can make the entire process an automated laboratory procedure to make real progress toward using aneuploidy as a biomarker. Other investigators have also developed methods to measure whole tissue section aneuploidy using the advanced techniques of tissue slide scanning methodology and histological-image processing.³⁵ Our technique was more automated because the staining could readily separate the epithelial and stromal cells and did not need to identify specific regions of interest by a pathologist. Future comparison between the two methods could determine which of these methods is more suitable for clinical practice.

Previously, Eisen et al. qualitatively evaluated the remapping of Barrett's esophagus in 18 patients by using an adapted upper endoscope. They reported that 85% of sites with abnormal Ki67 expression were reproducibly identified at the subsequent endoscopy.³⁶ The second study compared 700 biopsy specimens from 22 patients and was able to accurately relocalize areas with abnormal biomarkers on endoscopic remapping.³⁷ However, both these studies focused on the qualitative (yes/no) rather than the quantitative expression of these biomarkers and also did not address variability across time. In chemopreventive trials, the quantitative variability of the marker(s) to be used is very relevant since a chemopreventive agent is unlikely to alter the degree of expression rather than completely switching the marker ON or OFF. Recently, Reid and colleagues proposed that the spatial and temporal variability of somatic chromosomal content could be a useful biomarker for progression of BE to cancer with an intervention window of ∼2 years.¹¹ Thus, understanding the degree of expected variability in a biomarker could help in the development of serial biomarker assays that could confidently detect early cancer. A biomarker could have good discriminatory ability to differentiate disease from normal but if the degree of expression depends on where the biopsies are obtained, it may render that biomarker less useful. This is especially relevant for BE since BE lengths may vary from as short as 1 cm to as long as 15 cm. Moreover, BE patients undergo surveillance at regular intervals, and hence the spontaneous variability of biomarker expression over time in the same patient also becomes important.

Our study does have limitations. We examined only longer segments of BE. Biomarker variability may be less of an issue with shorter segments of BE. Given that length is an important predictor of progression, our results are still relevant to the biomarker-guided personalized management of BE. We did not include patients with dysplasia but we did this because we wanted to measure the natural variability of biomarkers in BE and presence of dysplasia could confound our results by affecting biomarker expression. We did not compare our modification of image cytometry with flow cytometry but previous studies that performed direct comparisons of image cytometry and flow cytometry showed good agreement.³⁴

In summary, we have performed a comprehensive study evaluating spatial and temporal variability of potentially useful biomarkers in BE and found the variability to be biomarker dependent. Based on our results, aneuploidy was the most stable biomarker in BE and should be further pursued in biomarker studies. Quantification of biomarker variability needs to be an important component of biomarker studies prior to their use as endpoints in chemoprevention trials and clinical assays for risk prediction.