Abstract
Reliable assessment of islet viability, mass, and purity must be met prior to transplanting an islet preparation into patients with type 1 diabetes. The standard method for quantifying human islet preparations is by direct microscopic analysis of dithizone-stained islet samples, but this technique may be susceptible to inter-/intraobserver variability, which may induce false positive/negative islet counts. Here we describe a simple, reliable, automated digital image analysis (ADIA) technique for accurately quantifying islets into total islet number, islet equivalent number (IEQ), and islet purity before islet transplantation. Islets were isolated and purified from n = 42 human pancreata according to the automated method of Ricordi et al. For each preparation, three islet samples were stained with dithizone and expressed as IEQ number. Islets were analyzed manually by microscopy or automatically quantified using Nikon's inverted Eclipse Ti microscope with built-in NIS-Elements Advanced Research (AR) software. The AIDA method significantly enhanced the number of islet preparations eligible for engraftment compared to the standard manual method (p < 0.001). Comparisons of individual methods showed good correlations between mean values of IEQ number (r2 = 0.91) and total islet number (r2 = 0.88) and thus increased to r2 = 0.93 when islet surface area was estimated comparatively with IEQ number. The ADIA method showed very high intraobserver reproducibility compared to the standard manual method (p < 0.001). However, islet purity was routinely estimated as significantly higher with the manual method versus the ADIA method (p < 0.001). The ADIA method also detected small islets between 10 and 50 μm in size. Automated digital image analysis utilizing the Nikon Instruments software is an unbiased, simple, and reliable teaching tool to comprehensively assess the individual size of each islet cell preparation prior to transplantation. Implementation of this technology to improve engraftment may help to advance the therapeutic efficacy and accessibility of islet transplantation across centers.
Keywords
Introduction
Islet transplantation has become an established biotherapy for patients with type 1 diabetes (27). The main objective of islet transplantation therapy is to infuse a sufficient amount of islets [>200,000 islet equivalents (IEQs)] to control blood glucose levels to achieve independence from exogenous insulin, thus avoiding hypoglycemic episodes (27). However, the overall process of islet isolation, purification, preservation, and quality control is subject to strict manufacturing regulations and, therefore, can hinder the clinical outcome of islet transplants (12,15). In particular, accurate islet mass quantification presents as the most crucial parameter for defining the dose and potency of the transplanted product when used as a therapeutic intervention for select patients with type 1 diabetes (7,11). Ricordi et al. also emphasized the need for consistency among different isolation centers when comparing islet preparations for clinical transplantation (22). In this context, new methodologies that can accurately standardize islet mass quantification across centers are a prerequisite for successful islet transplantation. Various different techniques were developed for evaluation of the islet yield over the years, but the standard method for quantifying isolated islets was established by Downing et al. in 1980 (2). In brief, a small fraction of the final islet preparation is dithizone stained to assess the purity of human pancreatic islet preparations. Dithizone provides a rapid method for dissociation of islet from acinar tissue by binding zinc ions present in the islet's β-cells, staining them red. Exocrine tissue also present in the preparations does not bind dithizone and is therefore not stained (22). As the majority of islets are not perfectly spherical, their estimated size is determined by mean diameter estimation using the established Ricordi algorithm (a range of 50 to >400 μm by dividing all islets into classes of 50-μm increments). However, islets <50 μm in size are not counted. Relative conversion factors make it possible to convert the total islet number of each diameter class into an IEQ number (one IEQ is equal to an islet of 150-μm diameter) (22). However, the skills required to carry out direct microscopic analysis are generally achieved through years of practice and are, therefore, difficult to teach. Moreover, quality assurance requires biyearly validation of personnel authorized for manual counting.
Computer-assisted digital image analysis (DIA) methods were developed with the intention to eliminate operator bias, previously achieved from conventional visual determination of isolated islet yields through a microscope, and to automate the islet counting process (3,5). Several reports indicate that quantification of isolated islets using DIA techniques has been shown to be rapid, consistent, and objective (3–5,9,19,20,28,29). However, there is still no widely accepted computerized method in place for quantification of isolated islets across centers. This could be due to the cost of some instruments and software used or, more likely, the complexity of software used to analyze images. Therefore, this study was undertaken to critically assess the performance of Nikon NIS-Elements Advanced Research (AR) software in regard to accuracy, reproducibility, and user simplicity to quantify human islet preparations and to compare these results to those obtained from the standard manual counting procedure.
Materials and Methods
Islet Isolation
Human pancreata [mean donor age: 48 ± 12 years, body mass index (BMI): 28 ± 7, n = 42; 13 male, 29 female] were harvested from adult, brain-deceased donors in agreement with the French Regulations and with our Institutional Ethical Committee. Pancreatic islets were isolated according to a slightly modified version (6) of the automated method of Ricordi et al. (22). Purification was achieved with Biocolle (Biochrom Ltd., Cambridge, UK) continuous density gradients using a COBE 2991 cell separator (Terumo BCT, Lakewood, CO, USA) (10). After gradient purification, tissue was collected from the topmost islet-rich layer to obtain the purest fraction.
Standard Counting Method
For each of the 42 preparations, three islet samples (25 μl taken from a total volume of 50 ml suspension for each of the 42 preparations) were centered on a 35-mm petri dish (Becton Dickinson, Le Pont-De-Claix, France) and stained with dithizone (Sigma, Saint Quentin Fallavier, France) [25 mg dissolved in dimethyl sulfoxide (DMSO; Sigma) and subsequently diluted in 50 ml]. Islets were spread out in a standardized square using a bent insulin needle (27 gauge; Becton Dickinson, Le Pont-De-Claix, France). The islets on the eyepiece were divided according to their estimated diameter into categories under the Nikon microscope (Tokyo, Japan) using a calibrated grid (Optimask, Morangis, France). The number of islets in each category was multiplied by a factor that converts the number of islets to IEQ. The total sum of equivalents represented the islet volume in one sample. The final yield was evaluated as the mean volume of three samples recounted on the whole suspension volume (22). The approximate size of nonspherical islets was determined by estimating their mean diameter. Calculating the percentage of the total dithizone-stained islets to the surrounding unstained exocrine tissue assessed purity. Multiple operators carried out standard manual analysis. To convert islet number into IEQ number (for manual and automated counting), the Ricordi algorithm was used. This consists of classifying and counting islets according to their size using 50-μm diameter range increments (22,23). For manual counting purposes and the AIDA method, islets or islet fragments smaller than 50-μm diameter were excluded from the analysis in order to compare both methods.
Automated Digital Image Analysis (Nikon Method)
The same 42 samples that were manually analyzed were also subjected to ADIA using the Nikon NIS-Elements Advanced Research version 3.2 software (Nikon Instruments, Tokyo, Japan) according to the manufacturer's guidelines. Briefly, images were obtained with a DS-Ri1 digital camera connected to Nikon's inverted Eclipse Ti microscope equipped with an automatic scanning table (Nikon France S.A., Champigny sur Marne, France). A total scan, one large image assembled from 25–42 images (depending on the size of the drop) of the region of interest was taken using a 4× magnification at a resolution of 1280 × 1024 (RGB 8 bit). This takes approximately 2 to 5 min; therefore, it is very useful for current good manufacturing practice (GMP), where a quick, easily reproducible method is desired. The aim was to achieve a comprehensive scanning of the entire sample while obtaining a balance between high-resolution imaging (for quantification accuracy) and tissue density. Dithizone-stained islet characterization was acquired by red and green channel subtraction (red–green) of the large RGB image. The generated image consists of high gray pixel levels (corresponding to the red color of islets) and low gray pixel levels (corresponding to exocrine tissue and remaining background). This step increases the discrimination ability of analysis software by using a single parametric threshold (monochromatic image) and by enhancing the signal/noise ratio. The software enables the threshold to be manually refined by the user in order to detect the maximum amount of positive stained tissue.
To assess purity, exocrine tissue was identified using the more classical method of white/gray color picking (red–green–blue threshold or hue–saturation–intensity threshold) on the original large RGB image. During the procedure, several automatic algorithm-treated images were obtained. The user can manually set thresholds or edit detection layers to add areas of interest, to remove artifacts, and to gather or separate closed detected objects (31). After user validation, equivalent (Eq) diameter for each islet was automatically calculated by object measurement and the Ricordi algorithm (22), and purity was obtained by a ratio of the full area of field measurement of islets and exocrine detection (sum of islets and exocrine, i.e., total tissue). Moreover, this technology also has the ability to exclude islet fragments smaller than 10-μm diameter in size. The data generated were transferred to a Microsoft Excel spreadsheet (Redmond, WA, USA) for calculations of purity, total islet number, and IEQ number. Purity was calculated by using the following formula: purity = islet area/(islet area + exocrine area).
Graphical analyses were performed using PRISM 5.0a for Mac OS X (Insight, Velizy, France). Results are expressed as mean ± SEM. Correlation was tested by determining the Pearson correlation coefficient r. Correlation was considered as good for a coefficient of determination r2 more than 0.8. Linear regression was performed to calculate β coefficients (slopes) and develop regression equations. This analysis was conducted with SPSS 18.0 software (SPSS, Inc., Chicago, IL, USA). Differences were considered significant at p < 0.05.
Results
Validation of Methodologies for Quantifying Dithizone-Stained Human Islets
First, we compared direct manual counting from n = 42 dithizone-stained human islet preparations to the counts obtained from the ADIA method (employing the Nikon NIS-Elements AR software, version 3.2). The mean of total islet equivalent numbers were 104,151 ± 76,434 IEQ for standard manual and 119,723 ± 72,515 IEQ for the ADIA method, shown (Fig. 1). The main criterion for grafting in Lille is to infuse a sufficient amount of islets (>200,000 IEQ). Interestingly, the AIDA method significantly enhanced the number of islet preparations eligible for engraftment compared to the standard manual method (p < 0.001) (Fig. 1).
Comparison of standard manual and automated digital image analysis (ADIA) of dithizone-stained human islets. N = 42 dithizone-stained islet preparations were manually counted by direct microscopic analysis or the ADIA method using Nikon's inverted Eclipse Ti microscope with built-in NIS-element AR software (computerized). Each point represents the mean of three separate counts. Preparations above 200,000 islet equivalents (IEQs) are eligible for grafting into type 1 diabetic patients (gray dotted line).
We next selected one illustrative sample of a purified human islet preparation stained with dithizone (red) and exocrine tissue (white) to elucidate the key functions of the software (Fig. 2A). High-resolution imaging enables the user to visualize their rotund, often irregular structure, while built-in macros can facilitate the accurate size of each islet to be accurately calculated (Fig. 2B). Furthermore, it is important to mention that islets often move during the counting process, forming clumps or halos around the periphery of the islet, which could be mistakenly counted as two islets. The software can correct this by using the separate function in the threshold dialog known as watershed segmentation. Once the user is satisfied with the threshold settings, the software selects the dithizone-stained islets overlayed with a binary layer using the user-selected threshold settings, highlighting them in green, and accurately calculates islets into total islet number and IEQ number based on size (Fig. 2B). For instance, with the standard manual method, the size of an islet with a diameter of approximately 175.2 μm can only be estimated, not measured, as its diameter lies between 150 and 200 μm.
Computerized scanning of an entire purified human islet preparation produced from an assembly of 25 high-resolution (4×) single images. This technique allows for the quantitative analysis of an entire drop of islets stained with dithizone (red) and exocrine tissue (white) to elucidate the key functions of the software. A photographic image of human islets after purification used automatic counting is illustrated in (A). Upon automated acquisition using the Nikon and NIS-element AR software, the islets appear green in color, enclosed by a red perimeter (B), and the diameter of each islet is calculated, as represented in (B). Scale bar: 500 μm.
Consistent with previous reports, including the one published by Niclauss et al. (19), comparison of standard manual and ADIA method in n = 42 samples showed good correlations between mean values of IEQ number (r2 = 0.91, β coefficient = 0.907) (Fig. 3A) and total islet number (r2 = 0.88, β coefficient = 0.774) (Fig. 3B). The following equations for IEQ number and the total islet number were developed by linear regression: IEQ number (standard manual) = 0.907 × IEQ number (ADIA) - 4457.47; Total islet number (standard manual) = 0.774 × Total islet number (ADIA) + 10757.24.
Islet surface area was estimated comparatively with IEQ number. Each point represents the mean of three separate counts for each islet preparation (n = 42) for IEQ (A) or islet number (B). In (A) and (C), the number of islets was normalized to islet equivalents (IEQs); one IEQ is defined as an islet with a diameter of 150 μm. All islet preparations indicated in gray were transplanted in patients with type 1 diabetes, the minimum required being 200,000 IEQ. (A) r2 = 0.91; (B) r2 = 0.88; (C) r2 = 0.93; and (D) r2 = 0.83. Regression equations and their slopes are shown in (A) and (B).
The greys dots represent the islet preparations, which have been grafted. By the ADIA method, this correlation was increased to (r2 = 0.93) when islet surface area was estimated comparatively with IEQ number (Fig. 3C). There was also a good correlation between islet surface area versus islet number (r2 = 0.83) (Fig. 3D).
Purity Assessment: Composition of Islet Preparations
Islet purity could be visually assessed by staff and stored for image archiving for quality control documentation, staff training, and donor verification purposes with the ADIA method (Fig. 4A). Islet purity was routinely estimated as significantly higher with the manual method versus the ADIA method (p < 0.001) (Fig. 4B), thus ascertaining that the ADIA method can reduce variation between individual users compared to manual evaluation. In fact, there was no correlation between the two methods (r2 = 0.34) (Fig. 4B).
Purity assessment: standard method versus the automated method. In (A) using the ADIA method, acinar tissue fragments are identified and outlined in green. Thus, after calculating the area of each cell population, the purity is estimated by the formula: purity = islet area/(islet area + exocrine area). In (B) the correlation between data obtained with standard counting (manual) and the ADIA (computerized) method, r2 = 0.34, n = 42. The inset in the upper right hand corner in (A) corresponds to the islet preparation depicted by the red square in (B).
Intraobserver Variability
One of the strongest arguments for implementation of the ADIA method for counting isolated islet tissue for research and clinical applications is to demonstrate that the ADIA method will produce more reliable, reproducible, or even more objective results than the standard manual method. To address this, the reproducibility of automated counting on different users (highly experienced vs. inexperienced islet counters) was assessed. Three different operators, using the ADIA method, counted 14 different islet preparations independently. If we compare user 1 (highly experienced user) versus user 2 or 3 (inexperienced users), the ADIA method showed coefficients of determination (r2) between 0.96 and 0.97, respectively (Fig. 5). The ADIA method showed very high intraobserver reproducibility compared to the standard manual method (p < 0.001).
Correlation between different users using the Nikon NIS-element AR software. An experienced user (user 1) is compared with two inexperienced users (users 2–3). Each user has independently counted three samples from each islet preparation (n = 14) using the Nikon NIS-element AR software. Individual values of IEQ number calculated using the software are compared between users. The computerized method showed coefficients of determination r2 between 0.96 (experienced user) and 0.97 (inexperienced users), respectively.
The Nikon NIS-Elements AR Software Can Detect Islets Between 10 and 50 μm in Size
Islets are categorized according to their diameters within 50-μm increments. Generally, one IEQ corresponds to an islet with a diameter of 150 μm. Values here are shown as mean ± SEM when comparing the standard manual with the ADIA method (n = 42) (Fig. 6). The manual technique eliminates small islets (with a diameter less than 50 μm), whereas the ADIA method can be programmed to count all islets but excludes islet fragments smaller than 10-μm diameter in size.
Automated method detects islets between 10 and 50 μm in size. Standard manual (Manual) and the ADIA (Automated) method were carried out to count n = 42 islet preparations. Using the standard manual method, the user does not count small islets with an estimated diameter of less than 50 μm. The ADIA method using the Nikon NIS-element AR software enables the user to count all islets that are >10 μm in size. Values here are shown as mean ± SEM when comparing the standard manual with the ADIA method (n = 42).
Discussion
From a clinical point of view, the most problematic part of islet transplantation is the islet isolation procedure from a donated pancreas. Without a doubt, even in the leading transplantation centers worldwide, a transplantable yield of isolated islets is obtained in <50% of processed pancreata (8,18). It is therefore evidently clear that producing high-quality functional islet preparations consistently across transplantation centers is a complex process, which requires considerable expertise. The origin and condition of a pancreas and size of human islets, as well as the method of islet isolation, can greatly affect the yield and viability of human islets for either transplant or research purposes. In this regard, previous reports showed that pancreata from obese donors gave higher islet yield than those from lean donors (1,13,18). Other studies have shown that donors of age 50 years or older gave good islet yields (1,17), but these islets responded poorly to glucose (30). Indeed, Lehmann et al. reported that size does matter and that small human islets were more superior in function than large human islets (14). Regardless, continuous improvements in islet isolation and quantitation have been made and are still ongoing. This study was conducted to comparatively assess the performance of the Nikon ADIA method to the standard manual method for counting isolated islets from human pancreata [mean donor age: 48 ± 12 years, body mass index (BMI): 28 ± 7, n = 42].
Herein, the AIDA method significantly enhanced the number of islet preparations eligible for engraftment compared to the standard manual method (p < 0.001) (Fig. 1) by providing accurate quantification of islet mass, the most crucial parameter for successful islet transplantation (24). Our method of counting islets is based on similar techniques outlined previously (19,28,29) with the additional parameters of measuring islet surface area and counting small islets (>10 and <50 μm). For both the manual counting method and the AIDA method, islets or islet fragments smaller than 50-μm diameter were excluded from the analysis in order to compare both methods. However, the AIDA has the ability to accurately count small islets (>10 and <50 μm) if desired. In fact, by the ADIA method, this correlation was increased to (r2 = 0.93) when islet surface area was estimated comparatively with IEQ number (Fig. 3C). Notably, 50% of islet preparations that were considered insufficient in number for clinical transplantation (>200,000 IEQ) with the standard manual counting method were indeed sufficient in number for transplantation based on the ADIA method (p < 0.001) (Fig. 1).
Although obtaining a high yield of islets from each islet isolation procedure is of paramount importance, so is the attainment of better quality islets with preserved β-cell function and viability. A lot of centers do not count islets smaller than 50 μm because they believe their contribution is not significant. However, studies in rodents and humans have shown that small islets are superior in function and viability than large islets when transplanted into diabetic animals or patients (14,16). In this study, we have utilized the Nikon software's ability to count islets up to a diameter ≤50 μm but to exclude islets ≤10 μm. This parameter increased the total islet yield by approximately 70 more islet counts compared to the manual method (Fig. 6).
Since the beginning of islet transplantation, there has been a huge emphasis on the importance of islet purity (30). In this study, the average percentage of purity was less than that of the standard method (Fig. 4) but still was within range (>40%). A possible explanation for this is that the ADIA method/NIS-element software has the ability to monitor the impurity profile of an islet preparation objectively, whereas independent users may be more subjective in their analysis. Niclauss et al. have shown that the percentage of purity was very similar between the manual method and their computerized analysis, which was performed using the MetaMorph and ImageJ software to automatically quantify purity and size of islets (19). However, in agreement with our findings, Pisania et al. found that islet purity by conventional dithizone staining was substantially higher with up to 20–30% overestimation when compared to the values obtained by both light and electron microscopy (21). Of note, it is also important to emphasize that the ADIA method/Nikon software can also detect small exocrine tissue (<10 μM), thereby calculating stringent values for the percentage of islet purity.
One of the major concerns of the “Edmonton Protocol” for islet transplantation published in 2000 was an emphasis on the need for successive transplants of highly purified islets (averaging 24% β-cell purity) and the close correlation between the numbers of islets transplanted and the success of the procedure (25,26), despite the expected loss of islet mass during the isolation process. Since then, recent studies have shown that autotransplantation of nonpurified islets presently offers the most successful insulin independence rates within the field of islet transplantation (32), suggesting that acinar and ductal contamination is well tolerated.
Introducing a new technique to replace an existing one is often a complex process; therefore, it is important to quantify the functional user requirements that the software delivers. As the ADIA method was superior to the standard manual counting method, we next assessed the reproducibility and ease of quantification with three different users. Indeed, comparing a highly experienced user versus two inexperienced users, the ADIA method showed coefficients of determination (r2) between 0.96 and 0.97, respectively (Fig. 5), with perceived ease of use. These findings suggest that the ADIA/Nikon technique could be applied to standardize islet mass quantification across centers worldwide.
In conclusion, we demonstrate that the ADIA method utilizing the Nikon software is an unbiased, simple, and reliable teaching tool to comprehensively assess the individual size of each islet cell preparation prior to transplantation, thus increasing the number of islet preparations eligible for engraftment compared to the standard manual method. Moreover, the Nikon software facilitated image archiving for quality control documentation, staff training, and donor verification purposes. Implementation of this technology to improve engraftment may help to advance the therapeutic efficacy, accessibility, and standardization of islet transplantation across centers.
Footnotes
Acknowledgments
This work was supported in part by grants from Conseil Regional Nord Pas de Calais (EGID European Genomics Institute of Diabetes), Contrat Plan Etat Region (CPER), INSERM, European Consortium for Islet Transplantation funded by Juvenile Diabetes Research Foundation International, and core facilities including the Biotherapies Platform provided by or funded by IFR 114 Institut de Medecine Predictive et de Recherche Therapeutique, and the University Hospital of Lille. The authors would like to thank Agence de la Biomedicine and Violeta Raverdy for pancreas coordination and our pancreas harvesting team for their technical contribution. We also thank Dr. Rouskas, INSERM U859, Biotherapy for Diabetes, University of Lille 2, for his help with statistical analysis. The authors would also like to thank Laura R. Sysko, of Nikon Instruments, Inc., Melville, NY, USA, for her assistance in writing the AIDA method. The authors declare no conflicts of interest.