Blood Collection

Blood samples (0.5–1 ml) were collected into heparinized vacutainer tubes (BD, Franklin Lakes, NJ) from 8 healthy volunteers (6 females and 4 males; age: 24–50) following informed consent (IRB protocol #AAAF2671). For each donor, 20 μl aliquots were pipetted into 200 μl 2D-barcoded tubes (Matrix Storage Tubes; Thermo Fisher Scientific Inc., Waltham, MA), placed into a compatible 96-tube rack (8×12) and covered with the supplied top (Thermo Fisher Scientific Inc.). For each donor, duplicate samples were prepared.

Irradiation

The covered tubes in the racks were transported to a Gammacell 40 ¹³⁷Cesium irradiator (Atomic Energy of Canada, Ltd., Canada) and each tube exposed to 0 (control), 1.0, 2.0, 3.0, 4.0, 5.0 or 6.0 Gy of gamma rays at a dose rate of 0.82 Gy/min. After irradiation, 200 μl of PB-MAX™ karyotyping media (Life Technologies, Grand Island, NY) was added to the 20 μl human whole blood samples. This differs from the proposed use in an actual emergency scenario where exposure would occur in vivo and samples collected into tubes preloaded with anticoagulant and medium.

Automation of Sample Processing

For the automation of sample processing we used a cell::explorer system [PerkinElmer, Waltham, MA (Fig. 2)]. Like most HTS/HCS systems the cell::explorer consists of sub-stations surrounding an anthropomorphic robotic arm [Denso Inc., Long Beach, CA (Fig. 2A)]. The stations used in this work were: Liquid handling: pipetting steps were performed on JANUS® platform [PerkinElmer (Fig. 2B)] with 96 channel multiple dispensing head, aspiration was done using an EL406™ microplate washer [BioTek, Winooski, VT (Fig. 2C)], liquid dispensing was done by FlexDrop [PerkinElmer (Fig. 2D)]. Automated incubator: an STX500 automated incubator [Liconic US Inc., Woburn, MA (Fig. 2E)], allows maintaining of up to 500 96-well plates in a humidified environment at 37°C, 5% CO₂. Centrifuge: sedimentation of cells was performed using a V-Spin 96-well plate centrifuge [Agilent Technologies, Santa Clara, CA (Fig. 2F)]. Assay: the following assay (shown schematically in Fig. 3) was programmed using PerkinElmer's plate::works automation control and scheduling software and performed entirely within the cell::explorer:

FIG. 2.

Layout of the cell::explorer™ at the Columbia Genome Center used as RABiT-II, showing the various subsystems: anthropomorphic robotic arm (A), the JANUS® liquid-handling system (B), EL406™ microplate washer (C), the FlexDrop dispenser (D), STX500 automated incubator (E), V-spin™ automated centrifuge (F).

FIG. 3.

Scheme of the automated peripheral human blood sample harvesting in cell::explorer after 70 h of cell culturing.

The rack of tubes containing blood and PB-MAX™ karyotyping media was placed into an incubator 37°C, 5% CO₂. After 44 h of incubation, cytochalasin-B (Sigma-Aldrich, St. Louis, MO) was added to cultures to block cytokinesis of proliferating lymphocytes at a final concentration of 6 μg/ml. Cells were gently mixed five times by aspiration and dispensing in the JANUS workstation, to break up pellets. Samples were placed back into a 37°C, 5% CO₂ incubator for an additional 26 h. After completion of cell culturing, the rack containing whole blood samples in 2D-barcoded tubes was removed from the automated incubator, and 200 μl of each sample was transferred into 96-well microplate (Greiner Bio-One, Austria) and processed as follows. The microplate was centrifuged at 1,200 rpm for 2 min and media above the settled cell pellet was aspirated. The following steps included addition on FlexDrop dispenser of 200 μl of hypotonic solution (0.075 M potassium chloride), 200 μl of fixative (3:1 methanol:acetic acid) followed by mixing three times on JANUS and centrifugation after each dispense step according to the scheme shown in Fig. 3. Next, samples were transferred to two imaging glass bottom 96-well plates (Brooks Automation Inc., Chelmsford, MA) preloaded with 300 μl of fixative (10:1 methanol:acetic acid), centrifuged, aspirated and allowed to dry for 10 min before staining. Finally, 200 μl of PBS containing 1.5 μg/ml DAPI (4′,6-diamidino-2-phenylindole) (Thermo Fisher Scientific Inc.) was added. Microplates were sealed and stored at 4°C until imaging and analysis.

Imaging and Image Analysis

Prepared imaging microplates were scanned on a commercial General Electric IN Cell Analyzer 2000 system (22). This automated imager is equipped with a large sensor CCD camera (2,048 × 2,048 pixel resolution) that is capable of whole-well imaging in 96- or 384-well microplates at high sensitivity. The IN Cell Analyzer 2000 is equipped with automation for filter, polychroic and objective changing for rapid multi-color imaging.

For this work, eighty-one 20× fields, covering an area of 60 mm² were captured per well (Fig. 4A). Image analysis was performed using FluorQuantMN, a custom-designed software, written in Visual C++ (Microsoft Inc, Redmond, WA) using the OpenCV image analysis libraries. Image analysis is performed by locating all nuclei in the image by first binarizing the log-transformed image using an adaptive thresholding algorithm, which assigns each pixel a value of 1 if its value is significantly larger than pixels in its neighborhood and zero otherwise. Nuclei are then located as binary large objects (BLOBs), using the algorithm of Suzuki and Abe (23) (Fig. 4C). Nuclei are correlated to cells based on their proximity and the number of nuclei of different sizes within a cell is scored. Nuclei with serrated or abnormal morphology as well as overlapping or clumped cells are rejected from the analysis. After initial optimization of the software parameters for a specific cell preparation and staining protocol, the software analyzes sets of images, without human intervention, reporting the number of cells with a given number of main nuclei and micronuclei.

FIG. 4.

Imaging and image analysis. Panel A: A set of 81 images captured by IN CELL Analyzer using 20× objective lens representing one well of 96-well imaging plate with fixed and DAPI-stained sample. Panel B: Cropped image from a well of 96-well imaging plate with a cell with 2 nuclei and micronucleus (yellow arrow) and panel C: the same cell (green circle) with 2 nuclei (magenta circles with green border) and micronucleus (green dot), as delineated by FluorQuantMN.

Sample Collection Kit

The sample collection kit described previously for the RABiT system (9, 13) was centered on capillary tubes, which are not appropriate for culturing blood. This kit has been updated to include components compatible with a commercial platform (Fig. 5). The custom engraved capillary tubes (9) were replaced with off-the shelf 2D-barcoded tubes organized in the form of 96-well plate ANSI/SLAS format. This use of a standard ANSI/SLAS platform at the stage of blood sample collection allows simplified introduction of the blood samples directly into a commercial robotic system, eliminating the need for a complex customized cell harvesting module (13). In the field, sample collection is performed using a commercial high-flow fingerstick. A measured amount of blood (20 μl) is metered using a micro syringe and loaded into a 2D-barcoded blood culture tube, which has been preloaded with an appropriate amount of cell culture medium. The tubes (96) are then loaded into an ANSI/SLAS footprint rack and transported to the RABiT-II system in a temperature controlled box.

FIG. 5.

Kit for blood sample collection developed for the implementation of the CBMN assay of peripheral human blood lymphocytes on commercial biotech robotic systems.

Sample Preparation

Automated preparation of 96 CBMN assay samples using RABiT-II approach was completed within 1 h starting from the end of cell culturing and ending with sample staining in two 96-well imaging microplates. It is considerably faster in comparison with one day needed for manual preparation of samples using the traditional method based on 5–10 ml blood cultures (21). The throughput of sample preparation for CBMN assay using RABiT-II approach corresponds to more than 2,000 samples per 24 h.

Imaging and Scoring

For high-speed triage purposes IAEA recommends scoring 200 binucleated cells (20), although it was demonstrated that 100 binucleated cells is sufficient to produce accurate dose estimations between 0 and 4.0 Gy (24). Our previously developed protocol (21), based on a 50 μl sample of blood, yielded a dose-dependent average number of binucleated cells between 300 to 1,100 per sample (1,875 mm²) using commercial Metafer 4 software. In contrast, the available area in a microwell of standard plate is much smaller and even a square-profile well only affords about 60 mm² of surface. This requires increasing cell density while reducing the number of cells per sample. Cell distribution within the well was kept as uniform as possible by optimizing the parameters of the liquid dispensing (25).

In our RABiT-II assay we reduced the initial blood volume to 20 μl and split the fixed cells between two imaging microplates (Fig. 3). Our software was able to identify on average 500 binucleated cells per sample with more than 100 binucleated cells identified in 95 of the 96 samples analyzed. That corresponds to the average number of detected binucleated cells from our previous work (21), taking into account the difference in the initial volume of blood. One sample (Donor 1, 4.0 Gy) yielded only 69 binucleated cells; however, 285 binucleated lymphocytes were detected by FluorQuantMN software in the duplicate sample. The number of detected binucleated cells in a single well varied from 29 (Donor 1, 4.0 Gy) to 668 (Donor 8, dose 1.0 Gy) with average number 250 binucleated lymphocytes. Dose dependence for all 8 volunteers is shown in Fig. 6.

FIG. 6.

Dose-response curves of the MN yields induced in human lymphocytes (from 8 donors) exposed to gamma rays produced by using automated RABiT-II approach. The error bars represent the upper and lower 95% confidence intervals.

Calibration Curve

A calibration curve (Fig. 7) was generated based on the data from all donors by pooling the yields for each dose. Between 5,000 and 10,000 binucleated cells per dose and 48,083 cells in total were used to generate this calibration curve. Comparison of the calibration curve (Fig. 7) with the corresponding curve for manual preparation of samples and image analysis of slides using commercial Metafer 4 software (21) shows that the decreased level of micronuclei was detected by the FluorQuantMN software. This could be explained by the low level of false micronuclei (less than 3%) detected by the FluorQuantMN software. It should be noted that the absolute yields of micronuclei depicted in this curve are somewhat lower than the typical values for manual scoring. This is because the imaging and image analysis can miss: 1. Very small micronuclei, which end up as 1–2 pixel BLOBs and are ignored; 2. Micronuclei partially overlapping a nucleus, which a human scorer would identify but an automated system cannot; and 3. Micronuclei, which are out of the imaging plane and would be seen by a human scorer manually adjusting the focus knob, or by using a time consuming z-stack.

FIG. 7.

Fitted linear-quadratic dose-response calibration curve of the MN yields in human lymphocytes for pooled data from 8 healthy donor using our RABiT-II system.

These issues can in principle be alleviated by allowing a human scorer to review the images and override the identification of micronuclei and binucleated cells generated by the software (as is done in the Metasystems platform) but this would defeat the purpose of a fully automated scoring system. Rather, as long as a significant dose response can be obtained with a fully automatic system, the difference in scoring efficiency between the manual and automated systems is largely only of academic interest.