Uveal Melanoma: Mean of the Longest Nucleoli Measured on Silver-Stained Sections

Abstract

purpose. To summarize and compare the various histologic methodologies for using nucleoli to assess the malignant potential of uveal melanoma.

methods. This was an observational series of 100 samples of uveal melanoma in which histologic sections were studied. The cases were selected so that approximately half (n = 49) of the tumors were from patients who had died of metastatic malignant melanoma. The 51 remaining tumors were from patients who had survived at least 9 years without development of metastasis. Central sections from the uveal melanomas were stained using the colloidal silver nitrate stain for nucleolar organizing regions (AgNOR). These were compared with an adjacent hematoxylin and eosin (H&E)–stained section. A light microscope with a micrometer inset into the eyepiece (×10) was used at a final magnification of ×1000 under oil immersion to measure the length of the nucleolus along the longest axis and the width perpendicular to that axis. From at least twenty cells selected from random fields throughout the tumor, the mean of the 10 longest and widest nucleoli (MLN) was calculated. Seven samples had to be discarded because the nucleoli were unmeasurable.

results. T-tests and Cox proportional hazard regression analysis indicated that the MLN of nucleolar length as measured on AgNOR-stained slides was as significant as cell type but was more significant than other histopathologic prognosticating variables measured and evaluated in this study. These prognosticators included tumor size, calculated as the largest tumor dimension; MLN width; and MLN length, as measured on H&E-stained sections.

conclusions. It has previously been demonstrated that AgNOR-stained nucleoli, unlike H&E-stained nucleoli, can be captured and measured by an automated image analyzer with prognostically significant results. This new method of simple oil-immersion measurements of the longest AgNOR-stained nucleoli length in microscopic sections of uveal melanoma provides an inexpensive and highly significant method for predicting outcome in patients with uveal melanoma. Because of the high contrast with the background, the silver-stained nucleoli clearly define the nucleolar boundaries, rendering them readily discernible and allowing greater ease and speed of measurement when compared with H&E-stained nucleoli. The method of random sampling that was used was comparable with linear sampling in predicting outcome. Highly necrotic tumors, however, had to be excluded from the study because of loss of nucleolar morphology.

The use of morphologic features of uveal melanoma as important prognostic predictors of patient survival was first established by Callender. ¹ Wilder and Paul ² found that pathologists using Callender’s classification at the Armed Forces Institute of Pathology (AFIP) could correctly predict the outcome in 73% of cases based on 10-year survival data. Callender’s classification is highly subjective, because it is based on multiple features without any ranking of their importance. Therefore, it is difficult to use this classification to categorize cells that have some features of epithelioid-type cells and some features of spindle-type cells. Because of this problem, ophthalmic pathologists at different institutions have had great difficulty in reproducing the results achieved at the AFIP. ³  

Gamel and McLean ⁴ and McLean et al. ⁵ focused attention on the large size of the nucleolus as the feature of Callender’s epithelioid cells that has the greatest prognostic significance. Thereafter, numerous techniques (Table 1) were developed and refined to objectively measure the mean of the 10 largest nucleoli (MLN). ^{6 7 8 9 10} Predictive results of high correlation between nucleolar diameter and malignant potential were delivered with the studies of both Huntington et al. ⁶ and McCurdy et al. ⁷ However, Coleman et al. ⁸ and Pe’er et al. ⁹ found no association between their measurements of MLN, made by using computer-assisted automated image capture and analysis (AICA), and patients’ outcomes. 

McLean et al. ¹⁰ criticized the AICA studies by Coleman et al. ⁸ and Pe’er et al. ⁹ because they used monochromatic images of hematoxylin and eosin (H&E)–stained sections. With monochromatic images, the nucleolus could not be distinguished from the nuclear chromatin that was usually condensed around the nucleolus. Their method differed from the method of Huntington et al. ⁶ and McCurdy et al. ⁷ who measured only the eosinophilic central mass of the nucleolus and excluded the surrounding hematoxylin-stained material. 

McLean et al. ¹⁰ argued that the size of the nucleolus measured in sections that were stained with silver nitrate for nucleolar organizing regions (AgNOR stain) should correlate with the size of the eosin-stained nucleolus in H&E-stained sections. They found that nucleoli on AgNOR-stained sections were measurable with AICA of monochromatic images, and several of the measurements were superior predictors of the patient’s outcome when compared with MLN measured by the method of McCurdy et al. ⁷  

McLean et al. ¹⁰ indicated that computer-assisted AICA could be used to measure MLN on silver-stained sections, but their study also left a number of unanswered questions. These included: Was the better prediction of outcome observed with AICA of silver-stained sections than with the method of McCurdy et al. ⁷ a valid finding? If valid, what was the reason for its success? In this study we demonstrated that light micrometer random-field measurements of nucleolar area on silver-stained sections provide more prognostic information on uveal melanomas than these measurements made on H&E-stained sections. 

Materials and Methods

Case Selection and Slide Preparation

The study protocol adhered to the tenets set forth by the Declaration of Helsinki. Unstained and H&E-stained central 6-μm sections were obtained from 100 patients with uveal melanoma in the Registry of Ophthalmic Pathology, AFIP, Washington, DC. Cases were selected so that approximately half (n = 49) of the patients had died of metastatic malignant melanoma (median, 3.4 years; range, 0.2–16.9). All 51 of the remaining patients survived at least 9 years without development of metastasis (median, 14.2 years; range, 9.0–29.3). These intraocular melanomas had been previously classified (by IWM) according to the modified Callender’s classification consisting of two groups (tumors composed only of spindle cells and tumors that contained epithelioid cells) and measured for largest tumor dimension (LTD) on either the gross specimen or microscopic section. ¹¹ In all cases, MLN had been measured on H&E-stained sections in John W. Gamel’s laboratory (Department of Ophthalmology, University of Louisville) using the method of McCurdy et al., ⁷ and these measurements are designated MLN (McCurdy) in this report. 

The unstained slides were heated in a 60°C oven for 20 minutes to promote adhesion. The sections were deparaffinized and bleached. One-step AgNOR ⁴ staining was performed using two solutions. The first solution was 40 ml of 2% gelatin (Bloom type A; Fisher Scientific, Fair Lawn, NJ) and 0.88% formic acid. The second solution was 80 ml of distilled water in which 40 g of silver nitrate was dissolved. In the dark, the two solutions were mixed, poured into the slide dish to cover the sections, and used to stain them for 30 minutes. The sections were then washed in distilled water, dehydrated, and coverslipped. 

Image Analysis

A standard light microscope with a reticle inset into the eyepiece (×10) and a ×100 oil-immersion objective at a final magnification of ×1000 was used to measure the nucleoli. Measurements were made first on the 100 silver-stained sections and then on the 100 H&E-stained sections. The length of the nucleolus was measured by turning the eyepiece micrometer to measure along the longest nucleolar axis. The width was measured perpendicular to the longest axis. The whole tumor section was scanned, and the observer (AM) selected for measurement only those nucleoli judged to be the largest in each field. Measurements on 20 of these nucleoli were retained per tumor, but only the longest and widest 10 were used for final computation of MLN. The observer had no knowledge of the patient’s outcome. Seven samples were discarded because the nucleoli were unmeasurable, either on the AgNOR-stained slides (six samples) or on H&E-stained slides (one additional sample). 

Statistical Analysis

Descriptive statistics and t-tests were performed using standard algorithms. Univariate Cox regression analysis was used to determine the relative prognostic value of the different measurements of MLN, cell type, and tumor size (LTD). Cox regression was chosen for two reasons. First, it is the most frequently used method of analyzing follow-up data with varying lengths of follow-up times, and, second, in all previous studies of MLN, ^{6 7 8 9} except in the study by McLean et al., ¹⁰ Cox regression Pearson correlation was used to examine the amount of association between the variables. 

After performing Cox regression, we converted each of the variables to a two-group categorical variable by using the cut point closest to the median, so that for each variable, standard measures of the reliability (sensitivity, specificity, and correct prediction) could be calculated. The proportion correctly predicted for two groups is the equivalent of the receiver operating characteristic area reported by McLean et al. ¹⁰ All analyses were performed by computer (Prodas Statistical Software; Conceptual Software, Houston, TX), with P < 0.05 regarded as significant. 

Results

The descriptive statistics (Table 2) indicate that there is good agreement between our measurements made on H&E-stained sections and the measurements made previously by the method of McCurdy et al. ⁷ For both the survivor and nonsurvivor groups, the MLN (McCurdy) is within the 95% confidence interval of the average of the mean of the MLN length (MLN L; H&E) and the mean of MLN width (MLN W; H&E). This difference is −0.11 μm for the survivor group and 0.06 μm for the nonsurvivor group. With silver-stained sections, the mean length and width of the nucleoli are significantly larger than those measured with H&E-stained sections only in the nonsurvivor group. The differences in mean nucleolar sizes for the surviving patients are 0.01 μm for length and −0.06 μm for width. The differences in sizes for the patients who died are 0.44 μm for length and 0.2 μm for width. The t-tests (Table 2) , univariate Cox analyses, and the sensitivity and specificity testing (Table 3) all indicate the superiority of MLN L (AgNOR). MLN L (AgNOR) was a better predictor of the patient’s outcome than MLN L (H&E), MLN (McCurdy), and MLN W, measured on both silver and H&E stained sections, and LTD. It should be noted that all these variables are associated (P < 0.05) with mortality from intraocular melanoma but only cell type (determined by IWM) was of prognostic value comparable with MLN L (AgNOR). 

Pearson correlation (Table 4) among the prognostic variables was generally high. Values of approximately 0.9 were obtained between length and width of the nucleolus, which were often measured on the same nucleolus. Differences ranging from 0.5 to 0.7 were obtained between different measures of MLN, from 0.4 to 0.6 between cell type and measures of MLN, and from 0.1 to 0.2 between LTD and measures of MLN. 

Discussion

Our study on a different sample of tumors demonstrates that the higher association with patient outcome observed by McLean et al. ¹⁰ between their measurements of MLN compared with those obtained using the method of McCurdy et al. ⁷ was a valid finding. We did not use AICA technology and achieved the same high predictive value for MLN. This indicates that the improved prediction of patient outcome was most probably due to the use of silver-stained (AgNOR) sections by McLean et al. ¹⁰ and not to AICA technology. Our findings of comparable nucleolar sizes and predictive values for MLN as measured by the method of McCurdy et al. ⁷ and MLN as measured by our method on H&E-stained slides eliminate the possibility that the improved prediction was due to the use of different sampling methods. We believe that the AgNOR stain provides greater prognostic value, because it more precisely defines the boundaries of the nucleolus than does eosin in the H&E stained sections. Silver staining affects measurements of nucleoli from tumors that metastasize more than from those that are not fatal and affects the length more than the width. Because of these findings, we also believe that the AgNOR stain facilitates the detection of irregularly shaped nucleoli and that these nucleoli are probably associated with tumors that metastasize and are fatal. 

It is likely that the eosin and AgNOR stains do not stain the entire nucleolus. Neither of these should stain the granular component that is present in the outermost coating of the nucleolus. ¹² This coating of granular material and any condensed nuclear chromatin around the nucleolus should stain with hematoxylin on H&E-stained sections. Segmentation of the nucleolus using AICA on monochromatic images would not distinguish hematoxylin from eosin staining. If the amount of this staining were not important in predicting the outcome of patients, then this could be an explanation for the failures of Coleman et al. ⁸ and Pe’er et al. ⁹ to find a significant association between their measurements of MLN and patient outcome. 

 Presented in part at the annual meeting of the Association for Research in Vision and Ophthalmology, Fort Lauderdale, Florida, May 2000.

 The opinions and assertions contained herein are the private views of the authors and are not to be construed as official or as reflecting the views of the Department of the Air Force, Department of the Army, Department of the Navy, or Department of Defense.

 Submitted for publication June 5, 2000; revised August 30 and December 7, 2000; accepted January 8, 2001.

 Commercial relationships policy: N.

Table 1.

 

 

Summary of Various Methods Used to Measure the Mean of the Ten Largest Nucleoli

Table 1.

 

Summary of Various Methods Used to Measure the Mean of the Ten Largest Nucleoli

Study	Method	Stain	Cases (n)	Field Selection	Orientation of Measured Diameter	Type of Measurement
Huntington et al. 6	Filar*	H&E	100	Random	Longest axis	Length
McCurdy et al. 7	Filar	H&E	100	5 mm Central linear strip	Random	Width
Pe’er et al. 9	AICA	H&E	234	5 mm Central linear strip	Longest axis	Length
Coleman et al. 8	AICA	H&E	94	5 mm Central linear strip	Longest and perpendicular axes	Length and width
McLean et al. 10	AICA	AgNOR	63	Random	Longest and perpendicular axes	Length and width

*  Filar micrometer with light microscope.

Table 2.

 

 

Descriptive Statistics and t-Test Results of Prognostic Variables

Table 2.

 

Descriptive Statistics and t-Test Results of Prognostic Variables

Variable	Survival Group	n	Mean Size*	Minimum Size*	Maximum Size*	t-Test Value	Probability > t
MLN L (AgNOR)	Survivors	46	2.62 ± 0.12	1.10	5.30	5.27	<0.0001
	Nonsurvivors	48	3.50 ± 0.12	2.00	5.50
MLN L (H&E)	Survivors	48	2.61 ± 0.12	2.00	5.20	2.97	0.004
	Nonsurvivors	48	3.06 ± 0.09	2.00	4.40
MLN (McCurdy)	Survivors	51	2.53 ± 0.08	1.57	4.84	2.67	0.009
	Nonsurvivors	49	2.85 ± 0.09	1.03	4.41
MLN W (AgNOR)	Survivors	46	2.26 ± 0.10	1.00	4.80	3.59	0.001
	Nonsurvivors	48	2.75 ± 0.09	2.00	5.00
MLN W (H&E)	Survivors	48	2.32 ± 0.09	1.00	4.30	1.99	0.05
	Nonsurvivors	48	2.55 ± 0.07	1.40	3.40
Cell type	Survivors	51	1.35 ± 0.07	1.00	2.00	5.94	<0.0001
	Nonsurvivors	49	1.86 ± 0.05	1.00	2.00
LTD	Survivors	51	11.82 ± 0.47	5.00	20.00	2.92	0.004
	Nonsurvivors	49	13.86 ± 0.51	5.00	21.00

*  Units for the different measurements of MLN are micrometers. Cell type is coded as 1 for tumors composed purely of spindle cells and 2 for tumors that contain epithelioid cells. LTD is measured in millimeters. Mean size is shown as mean ± SE.

Table 3.

 

 

Cox Regression Analysis Performed on Continuous Variables, Where Applicable

Table 3.

 

Cox Regression Analysis Performed on Continuous Variables, Where Applicable

Variable	χ2 of Cox	β Cox Regression Coefficient	Specificity	Sensitivity	% Correct
MLN L (AgNOR)	27.8	0.83	80	69	74
MLN L (H&E)	12.1	0.61	67	67	67
MLN (McCurdy)	8.5	0.62	68	69	69
MLN W (AgNOR)	13.0	0.67	70	73	71
MLN W (H&E)	5.6	0.54	65	67	66
Cell type	28.2	1.90	65	86	75
Tumor size (LTD)	8.0	0.12	59	63	61

 Specificity and sensitivity are calculated with the cutpoint closest to the median value. Specificity is the percentage of those who survived who were predicted to survive. Sensitivity is the percentage of those who died of metastasis who were not predicted to survive. Units of the Cox regression coefficients for the different measurements of MLN are micrometers⁻¹. Cell type is coded as 1 for tumors composed purely of spindle cells and 2 for tumors that contained epithelioid cells and LTD is millimeters⁻¹.

Table 4.

 

 

Pearson Correlation Matrix for the Prognostic Variables

Table 4.

 

Pearson Correlation Matrix for the Prognostic Variables

	MLN (McCurdy)	MLN L (AgNOR)	MLN L (H&E)	MLN W (AgNOR)	MLN W (H&E)	Cell Type	LTD
MLN (McCurdy)	1.00
MLN L (AgNOR)	0.68	1.00
MLN L (H&E)	0.59	0.55	1.00
MLN W (AgNOR)	0.70	0.92	0.53	1.00
MLN W (H&E)	0.61	0.58	0.90	0.59	1.00
Cell type	0.51	0.57	0.43	0.52	0.45	1.00
LTD	0.22	0.21	0.09	0.18	0.11	0.12	1.00