Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
Int J Radiat Biol. Author manuscript; available in PMC 2012 August 1.
Published in final edited form as:
PMCID: PMC3176460

The RABiT: A Rapid Automated Biodosimetry Tool For Radiological Triage. II. Technological Developments



Over the past five years the Center for Minimally Invasive Radiation Biodosimetry at Columbia University has developed the Rapid Automated Biodosimetry Tool (RABiT), a completely automated, ultra-high throughput biodosimetry workstation. This paper describes recent upgrades and reliability testing of the RABiT.

Materials and methods

The RABiT analyzes fingerstick-derived blood samples to estimate past radiation exposure or to identify individuals exposed above or below a cutoff dose. Through automated robotics, lymphocytes are extracted from fingerstick blood samples into filter-bottomed multi-well plates. Depending on the time since exposure, the RABiT scores either micronuclei or phosphorylation of the histone H2AX, in an automated robotic system, using filter-bottomed multi-well plates. Following lymphocyte culturing, fixation and staining, the filter bottoms are removed from the multi-well plates and sealed prior to automated high-speed imaging. Image analysis is performed online using dedicated image processing hardware. Both the sealed filters and the images are archived.


We have developed a new robotic system for lymphocyte processing, making use of an upgraded laser power and parallel processing of four capillaries at once. This system has allowed acceleration of lymphocyte isolation, the main bottleneck of the RABiT operation, from 12 to 2 sec/sample. Reliability tests have been performed on all robotic subsystems.


Parallel handling of multiple samples through the use of dedicated, purpose-built, robotics and high speed imaging allows analysis of up to 30,000 samples per day.

Keywords: Biodosimetry, Biomedical Automation, Radiation Triage, High Throughput


Following a large scale radiological event, it will become important to screen tens or hundreds of thousands of individuals for radiation exposure, both to identify those that would benefit from medical treatment and to alleviate the concerns of the “worried well” (Grace et al. 2010; Rea et al. 2010). Over the past five years the Columbia Center for Minimally Invasive Radiation Biodosimetry has developed the RABiT (Rapid Automated Biodosimetry Tool), a completely automated ultra-high throughput biodosimetry workstation (Salerno et al. 2007; Chen et al. 2009, 2010b; Garty et al. 2010; Lyulko et al. 2011; Turner et al. 2011). The RABiT analyzes fingerstick-derived blood samples to estimate past radiation exposure, or to identify individuals exposed above or below a cutoff dose.

The RABiT completely automates two well-established biodosimetry assays, the Cytokinesis Block Micronucleus assay (CBMN) (Fenech 2007; Lyulko et al. 2011) and the yield of phosphorylation of the histone H2AX (γ-H2AX assay) (Redon et al. 2009; Turner et al. 2011). Whilst the γ-H2AX assay provides a rapid, same-day, dose estimate and is thus, prima facie, the preferred assay (Figure 1a), due to re pair of the DNA (deoxyribonucleic acid) double strand breaks, it is typically useful for times only of the order of 24 to 48 h post exposure (Redon et al. 2009; Schmid et al. 2010), with the effective cut-off time depending on the dose and the method of analysis. The CBMN signal (Figure 1b), by contrast, has half life of around 1 year (da Cruz et al. 1994) but the assay currently requires up to three days for processing.

Figure 1
a), b) characteristics of the Histone H2AX phosphorilation (γ-H2AX) assay and the Cytokinesis Block Micronucleus assay (CBMN) assay. c) Concept of implementation of the two assays in the RABiT (in this case assuming a 36 hour useful life span ...

We have therefore designed the RABiT to be able to process either assay, with a rapid switchover between them. Our aim is that all samples collected within the first 24 to 48 hours post irradiation, will be processed using the γ-H2AX assay (Turner et al. 2011). After the γ-H2AX signal ceases to be useful, the RABiT will be switched over to the CBMN assay (Fenech 2007), and all subsequent samples processed using the CBMN assay (see Figure 1c). To establish the optimal time post exposure to switch between the two assays, we are currently characterizing the kinetics of γ-H2AX signal decay in the general population. We are also investigating the possibility of reducing the processing time of the CBMN assay.

The RABiT has been designed for operation as a permanent installation at a central location, with fingerstick blood samples collected in the field and transported to the RABiT (Garty et al. 2011). This allows a more robust installation, removes the need to design the system to withstand transport, and potentially allows use of the RABiT for other applications. Specifically, a long-term goal is that the RABiT may also be used for routine clinical applications in a hospital setting. Such dual usage has major advantages, in terms of both a) the ability of communities to acquire RABiT systems, and b) allowing the RABiT system to be in routine use, so that when urgently needed after a radiological event, the machine will be operational and staffed with experienced personnel. Two clinical applications that we foresee are high-throughput multiplexed cytogenetic screening such as for objective, fully-automated, resource-conserving, amniocentesis (Evans et al. 2006), with improved reproducibility over current approaches, and simultaneous profiling of multiple proteins with multiplexed immunoassay screening (Palzer et al. 2005). Initial tests of these endpoints are currently under way. Overall, a dual usage scenario will allow for continuous operation of the RABiT, at low capacity, ensuring appropriate maintenance and the availability of experienced personnel when needed after a radiological event.

This paper describes recent upgrades to the RABiT, resulting in increased throughput and reliability, and outlines the reliability tests that have been performed. Preliminary biodosimetry data collected using the RABiT are described elsewhere (Lyulko et al. 2011; Turner et al. 2011).

System Overview

An overview of the RABiT is shown in Figure 2. Capillary-blood samples arrive at the RABiT in a holder, described in the materials and Methods section, figure 4). After the holders are placed by the RABiT operator in centrifuge buckets on the input stage, no further human intervention is required. A SCARA robot (Selective Compliant Articulated Robot Arm – Stäubuli Robotics, Santa Rosa, CA, USA) loads the buckets into a modified centrifuge (Chen et al. 2010b) based on an Eppendorf 5810R (Eppendorf North America, Hauppauge, NY, USA) for lymphocyte enrichment. The buckets, containing capillaries with a separated lymphocyte band, are then transferred by the service robot to a lymphocyte harvest station where each capillary is identified, cut and the lymphocytes dispensed to a multi-well plate. When all 96 micro-wells are loaded, the multi-well plate is transferred to a Sciclone ALH3000 liquid handling system (Caliper Life Science, Hopkinton, MA, USA) where reagents can be added or removed as needed. An LTX220 robotic incubator (Liconic Instruments, Woburn, MA, USA) is available for lymphocyte culturing in the CBMN assay. After the assay is finished, the multi-well plates, containing fixed, stained lymphocytes, are moved to the transfer-to-substrate system (TTS) (Chen et al. 2010b) where the filter bottoms are removed and sealed between two layers of transparent tape. The lymphocytes are then imaged at 40x using a custom built imaging system (IS), based on Nikon infinity optics components (shown below in Figure 8, further details provided in the Materials and Methods section).

Figure 2
Photo of the RABiT, showing the various subsections. The Lymphocyte Harvest station is shown in close up in figure 5 below. The imaging system is shown in Figure 8 below.
Figure 4
Sample collection kit developed for the RABiT. Shown are the capillary holder (full in insert) with a prototype padded shipping container and ice pack, required phlebotomy supplies and the data collection card with an attached barcoded capillary.
Figure 8
Photos of the RABiT imaging system. a) Top view showing the illumination light path as well as the three imaging light paths, separated by dichroic mirrors. Each imaging path passes through a 4x relay lens and an image intensifier (II) to a CMOS camera. ...

The current (2011) version of the RABiT consists of three tiers of hardware (Figure 3) interconnected using six communications protocols. The top tier - central master controller (MC - based on the Real Time Application Interface Linux kernel) - is used to control, monitor, and record the operation of the system. The middle tier includes all subsystems which have their own controllers, and the actuators directly controlled by the central master controller.

Figure 3
Hardware structure of the RABiT. The top tier (Main Computer, MC) controls the overall operation of the RABiT. Two additional computers control the cell harvesting system (CH, dealing with the robotics and CHP dealing with the capillary imaging and barcode ...

The bottom tier encompasses peripherals under the control of the Second Tier systems (SCARA robot, cell harvesting computer, TTS and IS). Digital cameras use the Camera Link serial communication protocol to transfer pictures to their server computers. Two image storage devices, specifically a 15 Giga Byte Emprise 3000 hard disk array (Xiotech Corporation, Eden Prairie, MN, USA) and T120 tape library (Spectra Logic corporation, Boulder, CO, USA), use extremely high-speed (4 Gigabit per second) fiber channel links. Large packets of data between the central master computer and subsystem controllers are transferred using the User Datagram Protocol (UDP), while small packets of data are transferred using the Recommended Standard 232 (RS232) protocol.

Three speed bottlenecks were identified in earlier versions of the RABiT:

  1. Sample collection – this is the first stage, performed in the field, where the individuals need to be identified, their blood drawn and packed for shipping to the RABiT.
  2. Lymphocyte harvesting – this is the second stage of processing in the RABiT, following a centrifugation step. In this stage individual capillaries are extracted from the centrifuge bucket, their barcodes are read, the lymphocyte band is located and the capillary is cut. The bottom of the capillary, containing red blood cells, is discarded, the lymphocytes dispensed to a multi-well plate and the (empty) capillary discarded as well.
  3. Imaging – in this final stage of sample processing several tens of images from each sample need to be grabbed, analyzed and stored.

Common to all of these stages is that handling of individual samples limits throughput. By contrast, during centrifugation, 384 capillaries are centrifuged simultaneously; during liquid handling, transfer to substrate and while transporting samples between the various stages of the RABiT, 96 samples are handled simultaneously. Described below are our solutions and initial benchmarking data for these three bottlenecks.

Materials and Methods

Sample Collection

In order to facilitate blood collection we have developed a sample collection kit (Figure 4), consisting of lancets, bar-coded, heparin-coated capillary tubes with matched personal data cards and patient-tracking wristbands, alcohol wipes and sample holders for filled capillaries. The kit is designed around the 10–15 sample/hour throughput of a single minimally-trained collector, who would use 3–4 such kits in an 8-hour shift. See (Garty et al. 2011) in this issue for a discussion of how sample collection for the RABiT can be merged into the emergency planning scenarios of a large metropolitan area.

Each individual blood sample is collected into a 100 µl barcoded Polyvinyl Chloride (PVC) capillary tube (Safe-T-Fill, RAM Scientific, Yonkers, NY, USA). The main advantage of plastic capillary tubes is that they are safer to handle than glass (Jagger & Deitchman 1998). PVC capillaries are also easy to cut and laser-etch, allowing bar-coding and rapid cutting, both of which are required for use in the RABiT.

A capillary holder was designed to serve the dual purpose of facilitating sample collection and shipping as well as an insert for loading multiple capillaries into the RABiT. The holder consists of a plastic block with 32 holes, preloaded with sealing putty (Fisherbrand Hemato-Seal Capillary Tube Sealant, Fisher Scientific, Pittsburgh, PA, USA) and separation medium (Histopaque-1083, Sigma Aldrich, St. Louis, MO, USA).

After about 30 µl of blood is loaded into the capillary, the capillary is inserted into the holder, trapping an air bubble between the blood and separation medium, thus preventing their mixing during shipping. The sealing putty is compressed into and around the capillary ensuring a seal, requiring a weak force (less than 1 Newton) to extract the capillary from the holder. This prevents the capillary from falling out even if the holder is inverted and vigorously shaken, but still allows the RABiT robot to extract the capillary from the holder. This procedure allows the sample to be collected by an individual with minimal training, while maintaining the required layering of the blood and separation medium and preventing contamination. Since only minimally trained personnel are required, many sample collectors may be engaged and work in parallel, increasing sample collection throughput.

Lymphocyte Harvesting Station

In earlier prototypes of the RABiT (Salerno et al. 2007; Chen et al. 2009, 2010b; Garty et al. 2010), each capillary tube was picked out of the bucket using the SCARA robot, then presented to the barcode reader, then to the imaging camera, and finally to the laser for cutting. The capillary was then located over the appropriate well of the multi-well plate, the lymphocytes dispensed, and the capillary discarded. As this was done sequentially, on individual capillary tubes, this procedure was inefficient and delayed performing other tasks by the SCARA robot, namely, transferring the buckets to and from the centrifuge and transferring plates to and from the liquid handling system. Using this sequential capillary handling approach, which took ~12 sec per capillary, the maximum throughput was 6,000 samples per 24 hour day, where roughly 20 hours are spent by the SCARA robot moving capillaries, with the rest of the time spent by the SCARA robot moving buckets and plates.

To overcome this bottleneck, a new scheme for harvesting lymphocytes in fewer than 2 sec/sample was developed, based on the use of a dedicated robot, independent of the SCARA robot. This new scheme also involved an upgrade to the capillary-cutting laser and the development of a new capillary handling station (see Figure 5). This new station processes four capillaries at once and up to 96 capillaries without requiring assistance from the SCARA robot, freeing up the SCARA robot to handle plates and centrifuge buckets exclusively.

Figure 5
The newly designed lymphocyte harvesting station. Showing the incoming bucket and plate queue, the XYZ stage, with the capillary-filled bucket and plate currently being processed. One capillary is in each of the stations: A – Lymphocyte dispense, ...

The new capillary handling station contains:

  1. A Cartesian three-dimensional XYZ-stage (Mitsumi, Tokyo, Japan) serving as a movable platform for a) a centrifuge bucket, containing capillaries b) a micro-well plate, c) a disposing gripper and a trash chute. In this station, lymphocytes are dispensed into a multi-well plate, the empty capillary is discarded and a new capillary is loaded.
  2. A Charge Coupled Device (CCD) camera (CV-M4+CL, JAI, San Jose, CA, USA) which images the capillary to locate the boundary between red blood cells and separation medium.
  3. A Hawkeye barcode reader (BR) (Siemens AG, Munich, Germany), which reads the bar code on each capillary.
  4. A cutting station where an Osprey 2 Watt Ultraviolet (UV) laser (Quantronix Lasers, East Setauket, NY, USA), is used to cut the capillaries, 9 mm above the boundary identified in D.
  5. A motorized carousel (ϕ-stage, Mitsumi, Tokyo, Japan), with four rotating arms and four capillary grippers, rotates 90 degrees anticlockwise at each step to transfer capillaries to the four work stations (A–D above)

Table I presents the timing of the parallel processes performed in the cell harvesting station. In the new scheme, the SCARA robot interacts with the lymphocyte harvest station only by loading empty multi-well plates and removing full ones and by loading full centrifuge buckets and removing the empty ones. This is required once every 96 capillaries (approximately every 3.2 minutes).

Timing of new lymphocyte harvest system

Capillary Gripper Upgrade

The upgrade to the cell harvesting station also includes a new capillary gripper design, providing better reliability. The capillary gripper is used to manipulate individual capillaries, extracting them from the centrifuge bucket, presenting to the camera, barcode reader and laser cutting system and finally dispensing the lymphocytes into multi-well plates. The details of the design of the original capillary gripper were previously described by (Salerno et al. 2007). In the original RABiT system, the capillary gripper was mounted on the end arm of the SCARA robot, whilst in the current design four capillary grippers are mounted on the rotating arms of the ϕ stage.

We have modified the capillary gripper from its original design (Figure 6a) by introducing two additional rubber O-rings in the bore of the gripper (Figure 6b). This allows the capillaries to be grabbed with sufficient force to pull them reliably from the bucket inserts. High picking depth (the length of the tube actually in the gripper) leads to a high success rate, but also a higher rate of wear on the O-rings and the plunger (which seals the top of the capillary). The resulting design involved a trade-off between reliability and wear. Additionally, the shape of the bore was changed from cylindrical to conical. This relaxes the required tolerances on griper alignment with respect to the capillary, as well as the tolerances on the capillary diameter.

Figure 6
Design drawing of the capillary gripper for a) the original RABiT (Salerno et al. 2007) and b) the modified design.

In addition to increasing gripper reliability, we have developed an online system for diagnosing capillary picking errors. This system is based on a force/torque sensor, integrated into the gripper, and a Support Vector Machine (SVM) (Burges 1998) algorithm for error detection. This system is described in more detail in (Chen et al. 2010a).

Barcode Reading

Prior to cutting the capillaries and dispensing the lymphocytes into one micro-well of a multi-well plate, the barcode on each capillary is read and matched with the barcode and the position of the well on the multi-well plate. Successful reading of this barcode is crucial for automated sample tracking. We have developed a procedure for marking the capillaries using a UV laser similar to the one used for cutting the capillaries, but operated at lower power. A barcode etching spanning roughly 120° of the capillary’s circumference is etched using a laser power of 0.7 W. The capillary is then rotated and the process repeated to imprint a barcode on the entire circumference of the tube. Table II shows the various barcodes that we tested for marking the capillaries. Eventually we chose barcode 128C, which is a very high-density barcode symbology, with a smallest element size of 0.2 mm.

Comparison of Barcodes Tested

Improvement on laser cutting

In addition to the implementation of parallel sample processing, we have increased the speed and reliability of cutting the capillaries. The increase in laser cutting speed was achieved by upgrading the Osprey UV laser (355 nm) used for cutting the capillaries from 1.1 W to 2 W.

To increase reliability, the beam is now swept across the diameter of the capillary, as shown in Figure 7, in order to minimize issues resulting from small misalignments of the capillary position and the vibration caused during picking and disposing processes. Once the laser receives the cut command from the central master computer, it sweeps the UV beam with a constant linear speed across the capillary tube. At the same time, the capillary gripper rotates with a constant speed in the direction opposite to the direction of the beam sweeping. Optimization of the two speeds is described in the results section.

Figure 7
Close-up of the laser cutting station. The Rotating Drive Motor (RDM) is used for spinning the capillary to ensure uniform cutting.

We have also introduced an assisting gripper; it clamps on to the bottom part of the capillary at the end of the laser cutting process and facilitates detaching and discarding the bottom part of the capillary, which contains unwanted red blood cells.

Imaging System

High speed imaging is central to the RABiT, and the imaging system has a number of innovative features. Despite much current focus on high-throughput high-content imaging systems (Gough & Johnston 2007; Trask et al. 2009), no commercial system is fast enough with adequate spatial resolution to allow the throughput necessary for the RABiT system. In particular, most current imaging systems acquire images sequentially, using mechanical means to move the sample as well as mechanized cube changers. This results in long lag times (50–500 msec) between image grabs.

We have developed a novel automated imaging system (Figure 8) based on Nikon CFI60 infinity optics components. The use of infinity optics components (Sluder & Nordberg 2007) allows inserting a scan head (SH) and multiple dichroic mirrors and filters in the” infinity space” between the objective (Nikon CFI60 10× Plan Apochromat, Morrell Instruments, Melville, NY, USA) and the tube lens (Edmund optics, Barrington, NJ, USA). An additional Nikon 4× relay lens (Morrell Instruments), mounted on each camera, allows achieving a 40x magnified image of the lymphocytes.

The RABiT imaging system makes use of three new techniques, briefly described here, all of which are critical for speeding up the automated imaging in the RABiT:

  1. Light steering: Within one micro-well, the RABiT images adjacent 200×200 µm frames by steering light from off-optical-axis frames into the cameras using fast galvanometric mirrors (SH) (HurryScan II, ScanLab AG, Munich, Germany). This results in typical transit times between adjacent frames of less than 1 msec - two orders of magnitude faster than would be possible with a purely mechanical stage. The light steering technique is facilitated by the splitting of the magnification (10× in the objective and an additional 4× in the relay lens) as the 10× objective is optimized for aberration-free imaging of a larger area of the sample than would be possible with a 40× objective. As an example, Figure 9 shows a 4 mm2 area containing Hoechst-stained cells, plated on the filter bottom of a multiwell plate and imaged using light steering. The scan head is limited to deflections of 1 mm, as described in the discussion. To achieve larger motions (e.g. moving between adjacent samples) a mechanical stage consisting of an LXR404 slide (X axis - 150mm travel), an MX80 slide (Y axis - 150 mm travel) and a ZP200 Vertical wedge (Z axis - 25 mm travel) (Parker-Hannifin Corp, Rohnert Park, CA, USA) is used.
    Figure 9
    Image of a 2 mm × 2 mm region of a filter containing stained lymphocytes, stitched from 100 individual images. The edge of the 9 mm diameter filter is marked by a dashed line.
  2. Multiple cameras in parallel: The RABiT imaging system uses multiple fast Complementary Metal-Oxide Semiconductor (CMOS) cameras (MV D1024-160 CL, Photonfocus AG, Lachen, Switzerland), coupled to image intensifiers (II - VS-1485, Videoscope International, Dulles, VA, USA) capturing images, produced by different fluorophores and separated using Brightline dichroic mirrors and filters (Semrock, Inc, Rochester, NY, USA). For the γ-H2AX assay, this allows simultaneous imaging of the nuclei using Hoechst 33342 (Invitrogen, Carlsbad, CA, USA) and γ-H2AX foci using Alexa Fluor 555 (Invitrogen), while for the CBMN assay, the nuclear stain Hoechst 33342 and the cytoplasmic stain CellMask Orange (Invitrogen) are used. Given that the staining protocols for both bioassays are optimized to use equivalent illumination and filters, this allows the RABiT to be rapidly switched from the γ-H2AX to the CBMN assay in less than 1 hr, by just loading new reagents and using alternate software routines.
    To facilitate high speed imaging, the RABiT uses an intense UV light source based on a high intensity Light Emitting Diode (LED) (Hamamatsu Photonics, Bridgewater, NJ, USA) to excite the nuclear stain (5mW on the sample); this allows a faster image grab in the nuclear channel. For cytoplasm/γ-H2AX imaging the green line from an Excite 120PC mercury lamp (Exfo life Science, Toronto, ON, Canada) is used, which provides 60mW on the sample. The beams from the two light sources are joined using a dichroic mirror (Semrock, Inc), as shown in Figure 8c. Since each sample is illuminated for only a few seconds, photobleaching is not a concern.
  3. Single-Step Autofocus: Automated focusing is usually a relatively slow iterative process, involving moving the objective towards and away from the stage until best focus is obtained. A faster approach that we have developed is to use a cylindrical lens – a circular bead imaged in focus through a cylindrical lens will appear circular, but when out of focus, it will appear as an ellipse whose aspect ratio defines the distance to the focal plane. Based on the measured aspect ratio, the objective lens is moved using an OP100 piezoelectric Z stage (PZS - Mad City Labs, Madison, WI, USA).
  4. Image analysis. To accelerate image processing and to minimize dead time due to the control and monitoring software, the imaging system software is run on a dedicated high speed Dual Quad core computer (Micro Disc Inc, Yardley, PA, USA), running Linux (Red Hat, Raleigh, NC, USA). Most of the image processing steps, for the nuclear and γ-H2AX/cytoplasm images, are performed on a Matrox Helios xCL (Matrox Imaging Systems Ltd, Dorval, QC, Canada) frame grabber with on-board pixel processing. A Solios xCL framegrabber (Matrox Imaging Systems Ltd.) with an on-board Field-Programmable Gate Array (FPGA) is used for asynchronous processing of the focus information.

Image analysis software

The image analysis software for the RABiT was written in the C programming language and utilizes the image processing and image analysis capabilities of the Matrox Imaging Library (MIL 9.0; Matrox Imaging Systems Ltd) with part of the analysis performed in the frame grabber itself. As noted above image analysis is performed based on a two-color image, with one color corresponding to the nuclear image and the second corresponding to either the cytoplasmic image (for CBMN analysis) or γ-H2AX fluorescence. When analyzing micronuclei, this approach provides more reliable identification of binucleated cells, compared to staining the nuclei alone. When analyzing γ-H2AX fluorescence this approach allows scoring cells with no fluorescence (unirradiated cells or cells in which the damage has been repaired) and also allows elimination of spontaneous fluorescence (e.g. dirt on the sample) which is not correlated with a nucleus.

The algorithm used for analyzing γ-H2AX fluorescence, described in detail in (Turner et al. 2011) is based on integration of the total fluorescence of AlexaFluor 555 stained antibodies within the boundaries of each nucleus. This is a simpler, faster and more reliable (at high doses) (Rogakou et al. 1998; Andrievski & Wilkins 2009) approach than the more common foci-counting, which requires high-resolution 3D imaging (Böcker & Iliakis 2006; Hou et al. 2009; Roch-Lefèvre et al. 2010) and underestimates doses above about 2 Gy due to focus overlap (MacPhail et al. 2003; Böcker & Iliakis 2006). Calibrated orange fluorescent beads (Spherotech,Inc., Lake Forest, IL, USA) are used to monitor variations in illumination intensity that are related to the 2,000 hour lifetime of the mercury lamp and correct the measured fluorescence accordingly.

Our automated algorithm for scoring micronuclei follows the standard criteria for scoring micronuclei in binucleated (once-divided) cells, as described by Fenech et al (Fenech et al. 2003; Fenech 2007). The actual algorithm used is described in ref (Lyulko et al. 2011). Briefly, nuclei are imaged as before, each nucleus is associated with a single cell (imaged in orange) and the sizes of all nuclei within a single cell compared to determine the number of binucleated cells (two equally sized nuclei) and the number of micronuclei.


The focus of this paper is to describe the technical upgrades to the RABiT and recent reliability tests performed on it. Validation of the biodosimetry assays and first dose-response curves measured using the two assays are described in (Lyulko et al. 2011) (CBMN) and (Turner et al. 2011) (γ-H2AX).

No Blood Leakage

When processing capillaries in the RABiT, there is a potential for leakage during centrifugation, where the blood in the capillary experiences acceleration of 3000 g. We have monitored all capillaries run through the RABiT and in about 5% there was minor penetration of Red Blood Cells into the sealing putty, but no leakage of lymphocyte separation media or lymphocytes out of the capillary.

Additionally, there is a possibility for leakage following the cutting stage, when the bottom of the capillary is open. To prevent leakage at this stage, the capillary gripper contains a plunger (see Figure 6) which seals the top of the capillary. Surface tension forces then prevent leakage from the capillary until the plunger is removed, and pressurized air is inserted into the top of the capillary, when the capillary is above the appropriate well. No leakage was seen using this system.

Reliability of Capillary picking

Figure 10 shows the force required to pull a capillary out of the centrifuge bucket in the lymphocyte harvest module. Table III shows the capillary picking success rate for the original (Figure 6a) and modified (Figure 6b) gripper, for various speeds of the handling robot and for various picking depths.

Figure 10
Force required to extract capillaries from the centrifuge bucket. The error bars correspond to the standard error ±√[p x (1-p)/N] , where p is the is the value of each bar in the histogram and N=41 is the number of trials.
Success Rate of Picking Capillaries

Reliability of Capillary Cutting

For a given laser power (1.9 W) and cutting time (1.7 sec) we have tested various laser sweeping and capillary rotation speeds (see Figure 7). Table IV shows the success rate for cutting 32 capillaries. Based on these measurements, we have selected operational settings of 55 rpm and 0.9 mm/s. Overall, using the higher laser power, the swept beam, and the assisting gripper, we have reduced the cutting time from 3.5 sec to 1.7 sec while improving the success rate to >99%.

Success Rate of Laser Cutting

Sample Imaging - Focus system

The focusing system uses fluorescent spherical beads interspersed with the lymphocytes. In order to calibrate the autofocus system, we used 10 µm crimson fluorescent beads (Invitrogen). Images were taken at 1 µm intervals over a range of 100 µm around the focal plane. For example, Figure 11 shows single 10 µm beads imaged through a 1000 mm cylindrical lens (Thorlabs, Newton, NJ, USA) at 20 µm above focus (Figure 11a), at focus (Figure 11b), and at 20 µm below focus (Figure 11c). Figure 12 shows the measured aspect of a bead as the objective lens is moved across focus as well as the apparent size of a lymphocyte nucleus co-planar with the bead and imaged simultaneously.

Figure 11
Images of 10 μm beads through a cylindrical lens. The Z-values correspond to the position of the objective lens, mounted on a piezo-actuator.
Figure 12
Measured aspect ratio of 10 μm fluorescent beads as a function of objective lens position (line). The values shown correspond to the vertical extent of a bead image divided by its horizontal extent. As a comparison, the symbols show the apparent ...

Sample motion during Imaging

Figure 13 shows the measured time to perform a given deflection. A deflection of 200 µm (360 µsec) corresponds to motion from one field of view to the adjacent one. A deflection of 1 mm (~800µsec) corresponds to “resetting” the scan head after 100 images were taken.

Figure 13
Settling time of the scan head as a function of deflection distance along the two axes.

The mechanical stage used to augment the scan head for deflections above ±1mm requires 50 msec to perform a 9 mm motion (the distance between two adjacent wells in a multi-well plate).

Changing of the lens position is done via a fast piezo-actuator. Figure 14 shows the time required for a 100 µm vertical motion of the objective lens. 12.6 msec are required to reach 90% of the required motion. This settling time does not depend on the magnitude of the motion. We have seen that typical deflections required for maintaining focus when imaging a filter bottom of a multi-well plate are 12 µm.

Figure 14
Settling time for the piezoelectric Z stage (PZS) following a 100 μm motion of the objective lens. The upper curve corresponds to the requested position whilst the lower curve corresponds to the actual position. 1 vertical division corresponds ...

A full benchmarking of the RABiT imaging system, including image acquisition and image analysis is currently under way.

Sample imaging

Recently, we have used the RABiT system to measure γ-H2AX fluorescence labeling in lymphocyte nuclei from blood samples collected from four healthy female donors, irradiated ex vivo with a range of γ-ray doses between 0 and 8 Gy (Turner et al. 2011). Figure 15 shows sample images of lymphocytes, irradiated to 0, 2 and 5 Gy and processed using the RABiT γ-H2AX assay (See (Turner et al. 2011) for exact protocol), in filter bottom multiwell plates and then imaged using the RABiT imaging system. The figure shows representative γ-H2AX fluorescent images captured, in the RABiT, as 12-bit monochrome images (using green illumination and viewing through a 585nm filter). The images also show an overlaid outline of the nuclear image (as generated by the RABiT image analysis routines.

Figure 15
Images of irradiated lymphocytes (0 Gy, 2 Gy and 5 Gy), processed using the RABiT γ-H2AX assay protocol, as described in (Turner et al. 2011), and imaged using the RABiT imaging system. The outlines around each nucleus were generated from the ...


The Rapid Automated BIodosimetry Tool (RABiT) is a fully automated, ultra-high throughput robotically-based biodosimetry workstation. It screens fingerstick-derived blood samples, either to estimate past radiation dose, or to sort individuals exposed above or below a cutoff dose. The RABiT automates two mature, but currently manual, biodosimetry assays (CBMN and γ-H2AX), both of which have been shown to be highly radiation specific. In designing the RABiT we have taken a comprehensive approach. In additional to the robotic sample handling and analysis, we have developed and tested a sample collection kit, as well as transport procedures that are compatible with high throughput collection in a disaster environment, while maintaining sample integrity. In order to upgrade the RABiT’s throughput from a few thousand samples per day, we have located three bottlenecks in the RABiT sample flow, namely, sample collection, lymphocyte isolation and imaging.

This paper describes our approach for accelerating these three processes. We also describe the reliability tests performed on the capillary handling and tracking. Biological tests of the automated lymphocyte processing in multiwell plates are given elsewhere (Lyulko et al. 2011; Turner et al. 2011).

Sample integrity

A major concern in RABiT design was to avoid blood leakage from the capillaries both during shipping and during processing in the RABiT. This is important to prevent cross contamination and loss of samples, as well as being a safety issue for the RABiT operator and maintenance personnel. In theory, because the bottom of the capillary is sealed, the blood and separation medium cannot leak out of the top during shipping due to surface tension. Our tests have confirmed this both in the lab and in the field.

To demonstrate this, we have tested leakage during blood shipping (Garty et al. 2010), by collecting 32 (unirradiated) blood samples and shipping them, using a commercial carrier from Arizona to New York. None of the samples leaked during shipping and all samples maintained the required layering of blood and separation medium, allowing efficient separation of lymphocytes from red blood cells. The lymphocytes were then cultured to division and formed binucleated cells.

Another aspect of sample integrity is that of the viability of lymphocytes during shipping, in particularly for the CBMN assay where the lymphocytes must be healthy enough to divide. Additionally, the yield of phosphorylated H2AX was seen in many studies to decrease with time (Redon et al. 2009; Redon et al. 2010) due to repair of the DNA double strand breaks. While this cannot be prevented prior to the blood draw, it can be greatly reduced by storing the blood at low temperatures (Moroni et al. 2008). We therefore envision shipping the blood samples chilled, in containers similar to those used for shipping vaccines (Kendal et al. 1997). A comprehensive study, comparing different shipping modalities is currently under way. The logistics of sample collection and shipping are discussed elsewhere in this issue (Garty et al. 2011).

A second major concern is sample tracking. Each capillary is marked with a barcode, corresponding to a specific individual’s contact info. The reliable reading of this barcode, prior to cutting the capillary, is therefore of utmost importance. In the RABiT, we have achieved a success rate of >95% for reading the bar codes (Chen et al. 2009). The misread barcodes are due to improper angular orientation of the capillary – the barcode reader is facing the seam between two barcode etchings. This inaccuracy will be resolved by rotating the capillary during the barcode reading. The filter-bottomed multi-well plates and the sealed imaging substrates are also marked with a barcode and barcode readers are installed at the entrances to all RABiT subsystems.

Lymphocyte harvesting

In a previous work (Garty et al. 2010) we have reported on an efficient separation of the lymphocytes within the capillary. In order to transfer these lymphocytes to a multi-well plate, we must first successfully extract the capillary from the centrifuge bucket, then cut the capillary and finally dispense the lymphocytes into the well.

The large variation seen in the force required for extracting the capillary from its holder (Figure 10) stems from variations in both capillary diameter (±30 µm, due to the extrusion process) and the amount of putty used to seal the capillary bottom (loaded manually). The force required to extract the capillary is directly related to the area of the capillary over which the putty is bound. Nevertheless, in no case did we see that the binding was too strong for the gripper arm to successfully extract the capillary.

The lymphocyte harvesting station has been changed significantly from the one described previously (Chen et al. 2009, 2010b; Garty et al. 2010). It now consists of a standalone robot whose sole purpose is to handle the capillaries. This enabled processing of 4 capillaries in parallel as well as freeing up the SCARA robot to perform other tasks.

Although the new robot is slower than the SCARA robot at picking capillaries, the reliability of capillary picking has been significantly improved, with 1–5% of capillaries not picked on the first attempt (compared to almost a half with the old design). This is typically due to small misalignmnets between the capillary and the gripper. An online rapid failure detection system (Chen et al. 2010a) was designed to trap this error and perform a second attempt at picking the capillary in case of failure. Since the capillary picking time is still shorter than the cutting time, this has no impact on throughput.

An improvement in both cutting speed and reliability has also been shown in this work.

As described in detail in (Turner et al. 2011) the lymphocytes released from the capillary into the multi-well plate showed: (i) sufficient lymphocyte number to perform the assay (ii) adequate dispersal and (iii) good cellular morphology. In particular we have seen no difference in lymphocyte morphology when comparing lymphocytes from capillaries cut with a blade to those from capillaries cut using the laser. This confirms our previous calculations that the laser does not significantly heat the lymphocytes.


A key component of the RABiT’s imaging system is the simultaneous acquisition of images using different stains and multiple cameras. As both image analysis algorithms (γ-H2AX and CBMN) require the two stain images to be matched, proper alignment of the two cameras is vital. We routinely use multispeck fluorescent beads (Invitrogen), to measure alignment of the RABiT cameras. At the beginning of each session of imaging, we grab images of the beads and verify alignment. Using this method, alignment of the two images with a precision of 5 pixels (roughly 1 µm) is typically achieved.


A major rate-limiting step in high-throughput imaging systems is the focusing process. In order to get good image quality, typical microscope objective lenses have a shallow depth of field and may therefore be sensitive to the roughness of the sample being imaged. This is noticeable when imaging cells on a microscope slide which may be slightly tilted and becomes a serious problem for imaging rough substrates, such as the filter bottoms of multi-well plates. The typical solution in automated imaging systems is to grab a series of images at different focal planes and select the best one. This is obviously not compatible with high throughputs.

Our solution, of interspersing fluorescent beads within the sample and imaging them through a cylindrical lens, allows making rapid on-line focus corrections. The aspect ratio of the beads is continuously monitored. If a bead appear as a vertical (horizontal) ellipse, the objective can be moved up (down) rapidly before the next image is taken. Furthermore, the value of the aspect ratio (see figure 12) can be used to determine the extent of the motion. The range of accuracy of this method is, in our case, about ±30 µm, which is larger than the ~10µm variations we have seen when imaging actual filters. It should be noted that the emission spectrum of the crimson fluorescent beads (Invitrogen) used for monitoring the focus quality is sufficiently separated from the emission spectrum of the nuclear, cytoplasmic and AF555 stains used and therefore bead images do not appear in the other cameras.

Sample motion during imaging

As noted above, mechanical motion in an automated microscope system typically takes tens of msec. We have overcome this by performing 99% of sample motion using a galvanometric scan head (namely imaging a 10×10 array of frames, using the scan head, for every motion using the mechanical stage). We have limited the deflection to ±1mm or less on both axes. At larger deflections, due to the short working distance of the objective used (1.2 mm; Nikon S Fluor 10×; Morell Instruments), the light is collected into the objective lens at an angle approaching 45°, resulting in coma aberrations. This limit was seen both in simulations and in the actual RABiT imaging system.


Over the past five years the Columbia Center for Minimally Invasive Radiation Biodosimetry has developed the RABiT, a completely automated, ultra-high-throughput biodosimetry workstation. The RABiT analyzes fingerstick-derived blood samples to estimate past radiation exposure or to identify individuals exposed above or below a cutoff dose

Here we have described ongoing upgrades made to the RABiT, with the goal of increasing its throughput from 6,000 samples to 30,000 samples per day, while maintaining or improving reliability. Specifically, we have streamlined and, wherever possible, parallelized all stages of the RABiT where individual samples are handled (sample collection, lymphocyte harvest and the imaging system). The current RABiT (2010) is undergoing extensive benchmarking and reliability tests, and a large scale demographic study is in preparation for establishing RABiT calibration curves for both assays.


This work was supported by grant number U19 AI067773 to the Center for High-Throughput Minimally Invasive Radiation Biodosimetry from the National Institute of Allergy and Infectious Diseases / National Institutes of Health.


Declaration of Interest

The authors report no declarations of interest.


  • Andrievski A, Wilkins RC. The response of γ-H2AX in human lymphocytes and lymphocytes subsets measured in whole blood cultures. International Journal of Radiation Biology. 2009;85:369–376. [PubMed]
  • Böcker W, Iliakis G. Computational methods for analysis of foci: Validation for radiation-induced γ-H2AX foci in human cells. Radiation Research. 2006;165:113–124. [PubMed]
  • Burges CJC. A tutorial on support vector machines for pattern recognition. Data Mining and Knowledge Discovery. 1998;2:121–167.
  • Chen Y, Wang H, Zhang J, Pile JA, Garty G, Simaan N, Yao YL, Brenner DJ. Automated recognition of robotic manipulation: Failure modes in high-throughput biodosimetry tool. Engineering Applications of Artificial Intelligence Submitted. 2010a
  • Chen Y, Zhang J, Wang H, Garty G, Xu Y, Lyulko OV, Turner HC, Randers-Pehrson G, Simaan N, Yao YL, Brenner DJ. Design and preliminary validation of a rapid automated biosodimetry tool for high throughput radiological triage. Proceedings of the American Society of Mechanical Engineers. 2009;3:61–67.
  • Chen Y, Zhang J, Wang H, Garty G, Xu Y, Lyulko OV, Turner HC, Randers-Pehrson G, Simaan N, Yao YL, Brenner DJ. Development of a robotically-based automated biodosimetry tool for high-throughput radiological triage. International Journal of Biomechatronics and Biomedical Robotics. 2010b;1:115–125.
  • da Cruz AD, McArthur AG, Silva CC, Curado MP, Glickman BW. Human micronucleus counts are correlated with age, smoking, and cesium-137 dose in the Goiania (Brazil) radiological accident. Mutation Research/Environmental Mutagenesis and Related Subjects. 1994;313:57–68. [PubMed]
  • Evans MI, Sharp M, Tepperberg J, Kilpatrick MW, Tsipouras P, Tafas T. Automated microscopy of amniotic fluid cells: Detection of FISH signals using the FastFISH imaging system. Fetal Diagnosis and Therapy. 2006;21:523–527. [PubMed]
  • Fenech M. Cytokinesis-block micronucleus cytome assay. Nature Protocols. 2007;2:1084–1104. [PubMed]
  • Fenech M, Chang WP, Kirsch-Volders M, Holland N, Bonassi S, Zeiger E. HUMN project: Detailed description of the scoring criteria for the cytokinesis-block micronucleus assay using isolated human lymphocyte cultures. Mutation Research. 2003;534:65–75. [PubMed]
  • Garty G, Chen Y, Salerno A, Turner H, Zhang J, Lyulko OV, Xu Y, Wang H, Simaan N, Randers-Pehrson G, Yao YL, Amundson SA, Brenner DJ. The RABiT: A rapid automated biodosimetry tool for radiological triage. Health Physics. 2010;98:209–217. [PMC free article] [PubMed]
  • Garty G, Karam PA, Brenner DJ. Infrastructure to support high-throughput biodosimetry screening after a radiological event. International Journal of Radiation Biology. 2011;87:754–765. [PMC free article] [PubMed]
  • Gough AH, Johnston PA. Requirements, features, and performance of high content screening platforms. Methods in Molecular Biology. 2007;356:41–61. [PubMed]
  • Grace MB, Moyer BR, Prasher J, Cliffer KD, Ramakrishnan N, Kaminski J, Coleman CN, Manning RG, Maidment BW, Hatchett R. Rapid radiation dose assessment for radiological public health emergencies: Role of NIAID and BARDA. Health Physics. 2010;98:172–178. [PubMed]
  • Hou Y-N, Lavaf A, Huang D, Peters S, Huq R, Friedrich V, Rosenstein BS, Kao J. Development of an automated γ-H2AX immunocytochemistry assay. Radiation Research. 2009;171:360–367. [PubMed]
  • Jagger J, Deitchman S. Hazards of glass capillary tubes to health care workers. Journal of the American Medical Association. 1998;280:31–31. [PubMed]
  • Kendal AP, Snyder R, Garrison PJ. Validation of cold chain procedures suitable for distribution of vaccines by public health programs in the USA. Vaccine. 1997;15:1459–1465. [PubMed]
  • Lyulko OV, Garty G, Randers-Pehrson G, Turner H, Brenner DJ. Fast image analysis for the micronucleus assay in a fully automated high-throughput biodosimetry system. Radiation Research Submitted. 2011 [PubMed]
  • MacPhail S, Banath J, Yu T, Chu E, Lambur H, Olive P. Expression of phosphorylated histone H2AX in cultured cell lines following exposure to x-rays. International Journal of Radiation Biology. 2003;79:351–358. [PubMed]
  • Moroni MM, Krasnopolsky K, Subramanian U, Martin PR, Doherty KM, Prasanna PGS. Does cell culture type and blood transport temperature affect dicentric yield and radiation dose assessment? [cited 2011 February 14];Journal of Medical, Chemical, Biological and Radiological Defense. 2008 6 Available from
  • Palzer S, Bailey T, Hartnett C, Grant A, Tsang M, Kalyuzhny AE. Simultaneous detection of multiple cytokines in elispot assays. Methods in Molecular Biology. 2005;302:273–288. [PubMed]
  • Rea ME, Gougelet RM, Nicolalde RJ, Geiling JA, Swartz HM. Proposed triage categories for large-scale radiation incidents using high-accuracy biodosimetry methods. Health Physics. 2010;98:136–144. [PubMed]
  • Redon CE, Dickey JS, Bonner WM, Sedelnikova OA. γ-H2AX as a biomarker of DNA damage induced by ionizing radiation in human peripheral blood lymphocytes and artificial skin. Advances in Space Research. 2009;43:1171–1178. [PMC free article] [PubMed]
  • Redon CE, Nakamura AJ, Gouliaeva K, Rahman A, Blakely WF, Bonner WM. The use of gamma-H2AX as a biodosimeter for total-body radiation exposure in non-human primates. PLoS ONE. 2010;5:e15544. [PMC free article] [PubMed]
  • Roch-Lefèvre S, Mandina T, Voisin P, Gaëtan G, Mesa JEG, Valente M, Bonnesoeur P, García O, Voisin P, Roy L. Quantification of γ-H2AX foci in human lymphocytes: A method for biological dosimetry after ionizing radiation exposure. Radiation Research. 2010;174:185–194. [PubMed]
  • Rogakou EP, Pilch DR, Orr AH, Ivanova VS, Bonner WM. DNA double-stranded breaks induce histone H2AX phosphorylation on serine 139. Journal of Biological Chemistry. 1998;273:5858–5868. [PubMed]
  • Salerno A, Zhang J, Bhatla A, Lyulko OV, Dutta A, Garty G, Simaan N, Pehrson GR, Yao YL, Brenner DJ, Nie J. Design considerations for a minimally invasive high-throughput automation system for radiation biodosimetry; IEEE International Conference on Automation Science and Engineering, CASE 2007; 2007. pp. 846–852.
  • Schmid TE, Dollinger G, Beisker W, Hable V, Greubel C, Auer S, Mittag A, Tarnok A, Friedl AA, Molls M, Roper B. Differences in the kinetics of gamma-H2AX fluorescence decay after exposure to low and high let radiation. International Journal of Radiation Biology. 2010;86:682–691. [PubMed]
  • Sluder G, Nordberg JJ. Microscope basics. In: Greenfield S, David EW, editors. Methods in cell biology. Academic Press; 2007. pp. 1–10. [PubMed]
  • Trask OJ, Nickischer D, Burton A, Williams RG, Kandasamy RA, Johnston PA. High-throughput automated confocal microscopy imaging screen of a kinase-focused library to identify p38 mitogen-activated protein kinase inhibitors using the GE incell 3000 analyzer. Methods in Molecular Biology. 2009;565:159–186. [PubMed]
  • Turner HC, Brenner DJ, Chen Y, Bertucci A, Zhang J, Wang H, Lyulko OV, Xu Y, Schaefer J, Simaan N, Randers-Pehrson G, Yao YL, Garty G. Adaptation of the g-H2AX assay for automated processing in human lymphocytes. Radiation Research. 2011;175:282–290. PMC3121903. [PMC free article] [PubMed]