Image-guided endoscopic surgery has become a common procedure in the abdominal and thoracic cavities, because it is less invasive than open surgery.1, 2, 3 It is recognized that such procedures may be improved by optically enhanced imaging using fluorescent probes. To improve their utility, fluorescence probes must be designed so as to yield a high target-to-background ratio that can be achieved in several ways. One approach is to utilize targeting ligands linked to fluorophores that bind to cancers, whereas unbound fluorophores are absorbed and removed from the field of view.
Prior reports have described a fluorescein-conjugated avidin (Av-FITC) and an avidin conjugated with rhodamine green (Av-RhodG), which were able to visualize submillimeter cancer implants in the peritoneal cavity with acceptable signal-to-background ratios.4 This conjugate relies on the affinity of avidin for lectin binding protein (synonyms of asialo receptors, -galactose receptors) on the cell surface of wide variety of cancers arising from the ovary, pancreas, colon, and stomach, which can spread in peritoneal space.5 Unbound avidin is rapidly removed from the peritoneal cavity by absorption and taken to the liver. Thus, high target-to-background ratios are achieved. However avidin, which is derived from a hen egg protein, is not clinically practical because of its well-known immunogenicity in humans.6, 7 As a clinically feasible alternative to avidin, which can still target the cell surface D-galactose receptor but is made from nonimmunogenic materials, with even better binding capability by multiplying binding sites but still maintaining a favorable characteristic of high isoelectric point, a serum albumin conjugated with 23 galactosamine and 2 rhodamine green molecules (GmSA-RhodG) was designed and synthesized by conjugation of D-galactosamines to carboxyl groups yet can be conjugated with rhodamine green (GmSA-RhodG) to amide groups on the albumin. The purpose of this study was to synthesize and evaluate GmSA-RhodG for its ability to bind ovarian cancer cells in vitro and compare the fluorescence intensity of bound GmSA-RhodG to bound Av-RhodG in vivo in a mouse model of ovarian cancer peritoneal metastases to develop a clinically feasible method for the D-galactose receptor-targeted molecular cancer imaging.
Materials and Methods
Synthesis of Avidin-Conjugated Green Fluorescence Dyes
Avidin was purchased from Pierce Biochemical Inc. (Milwaukee, Wisconsin), bovine serum albumin (BSA) and galactosamine-conjugated bovine serum albumin (GmSA), which contained 23 galactosamine molecules on a single albumin molecule, were purchased from Sigma Chemical (St. Louis, Missouri), amido-reactive RhodG was purchased from Molecular Probes Inc. (Eugene, Oregon). At room temperature, of GmSA, BSA, and avidin in of was incubated with of RhodG-succinoimidyl ester, respectively, in dimethyl sulfoxide (DMSO) for . The mixture was purified with Sephadex G50 (PD-10; GE Healthcare, Milwaukee, Wisconsin). GmSA-, BSA-, and avidin-conjugated RhodG samples (GmSA-RhodG, BSA-RhodG, and Av-RhodG, respectively) were kept at in the refrigerator as stock solutions.
The protein concentration of GmSA-RhodG, BSA-RhodG, and Av-RhodG samples was determined with Coomassie Plus protein assay kit (Pierce Chemical Company, Rockford, Illinois) by measuring the absorption at with a visible ultraviolet (UV-vis) system (8453 Value UV-Bis system, Agilent Technologies, Palo Alto, California) using GmSA, BSA, and avidin standard solutions of known concentrations (100, 200, and ). Then, the RhodG concentrations were measured by the absorption at , respectively, with a UV-vis system to confirm the number of fluorophore molecules conjugated with each avidin, BSA, or GmSA molecule. By changing the concentration of the avidin, BSA, or GmSA solution, the number of fluorophore molecules per avidin, BSA, or GmSA was adjusted to be 2.0.
An established ovarian cancer cell line SHIN3 Ref. 8 was used for in vitro fluorescence microscopy, flow cytometry, and in vivo optical imaging for intraperitoneal disseminated cancer implants. The cell lines were grown in RPMI 1640 medium (Gibco, Gaithersberg, Maryland) containing 10% fetal bovine serum (FBS) (Gibco), 0.03% L-glutamine at , 100 units/mL penicillin, and streptomycin in 5% .
Transfection of Red Fluorescence Protein (DsRed2) to the SHIN3 Cell
The red fluorescence protein (RFP DsRed2)–expressing plasmid was purchased from Clontech Laboratories Inc. (Mountain View, California). The plasmid was transfected into the SHIN3 cells to validate the results with the targeted fluorophores (vide infra). The transfection of RFP was performed with the electroporation method using Gene Plus II (Bio-Rad Laboratories, Hercules, California). Briefly, of DsRed2-express plasmid was mixed with 2 million SHIN3 cells in of the cell culture media (RPMI 1640 with 10% FCS). Then the cell suspension was put in a pulse cuvette (Bio-Rad Laboratories) and 250-V pulses were delivered after 950 cycles.
SHIN3 cells were plated on a cover glass bottom culture well and incubated for . BSA-RhodG, GmSA-RhodG, or Av-RhodG was added to the medium and the cells were incubated for 1 or . Cells were washed one time with phosphate-buffered solution (PBS) and fluorescence microscopy was performed using an Olympus BX51 microscope (Olympus America Inc., Melville, New York) equipped with the following filters: excitation wavelength 470 to and emission wavelength 515-nm long pass. Transmitted light differential interference contrast (DIC) images were also acquired.
One-color flow cytometry was performed for the assessment of the fluorescing capability of Av-RhodG, GmSA-RhodG, and BSA-RhodG in SHIN3 cancer cells. SHIN3 cells were plated on a 12-chamber culture well and incubated for . Av-RhodG, GmSA-RhodG, or BSA-Rhod was added to the medium and the cells were incubated for 1, 3, and . For the assessment of competitive binding of GmSA-RhodG and avidin-RhodG, SHIN3 cancer cells were incubated with a large amount of excess GmSA for . At after incubation with or without GmSA, Av-RhodG was added to a concentration of and incubated for an additional . Cells were washed once with PBS and trypsinized, and then flow cytometry was performed. A 488-nm argon ion laser was employed for excitation. Signals from cells were collected using a 530/30-nm band pass filter. Cells were analyzed in a FACScan cytometer (Becton Dickinson, Franklin Lakes, New Jersey), and all data were analyzed using CellQuest software (Becton Dickinson). The fluorescence intensity of each fluorophore was termed the mean fluorescence index (MFI). A regression line was calculated from the data sets of incubation time and common logarithmic value of MFI, and then plotted as a function of incubation time using Microsoft Excel 2003 (Microsoft, Redmond, Washington). The MFI value at each time point and the slope of the regression line were compared among the three fluorophores.
Measurement of Fluorescence Intensity of Fluorophore Conjugates In Vitro
To compare the fluorescing capability of Av-RhodG, GmSA-RhodG, and BSA-RhodG in vitro, fluorescence intensity and emission spectra of Av-RhodG, GmSA-RhodG, and BSA-RhodG were measured by the In-Vivo Imaging System (CRI Inc., Woburn, Massachusetts) in arbitrary units [(a.u.)]. Av-RhodG, GmSA-RhodG, and BSA-RhodG ( PBS) were put in a nonfluorescent 96-well plate (Costar, Corning Incorporated, Corning, New York) and spectral fluorescence imaging was performed. A bandpass filter from 445 to and a long pass filter over were used for emission and excitation light, respectively. The tunable filter was automatically stepped in 10-nm increments from 500 to while the camera captured images at each wavelength interval with constant exposure. Spectral unmixing algorithms were applied to create the unmixed image of the three probes. A region of interest (ROI) as large as each well was drawn to determine the emission spectra and the fluorescence intensity of three probes using commercial software (Maestro).
Animal Model of Peritoneal Metastases
All procedures were carried out in compliance with the Guide for the Care and Use of Laboratory Animal Resources (1996), National Research Council, and approved by the National Cancer Institute Animal Care and Use Committee. The intraperitoneal tumor xenografts were established by intraperitoneal injection of SHIN3 or SHIN3 transfected with RFP DsRed2 cells suspended in of PBS in female nude mice (National Cancer Institute Animal Production Facility, Frederick, Maryland). Experiments with tumor-bearing mice were performed at 14 days after injection of the cells.
In Vivo Spectral Fluorescence Imaging
Av-RhodG, GmSA-RhodG, and BSA-RhodG were diluted in PBS and injected into the peritoneal cavities of mice with peritoneally disseminated cancer implants. Each set of three mice (one for each fluorophore) was repeated in triplicate. Three hours after intraperitoneal injection, all three mice were sacrificed with carbon dioxide inhalation. Immediately after sacrifice, the abdominal cavity was exposed and the three mice were placed side by side on a nonfluorescent plate to compare the fluorescence intensity of the tumors. Spectral fluorescence images were obtained using Maestro. Whole abdominal images as well as closeup images of the peritoneal membrane were obtained. A bandpass filter from 445 to and a long pass filter over were used for emission and excitation light, respectively. The tunable filter was automatically stepped in 10-nm increments from 500 to while the camera captured images at each wavelength interval with constant exposure. The spectral fluorescence images consisting of autofluorescence spectra and the spectra from the probes were obtained, and then, unmixed based on their spectral patterns using commercial software (Maestro).
Using the unmixed fluorescence image of the three peritoneal membranes, fluorescence intensity of the cancer implants was semiquantitatively compared among the three probes. A ROI as large as the peritoneal membrane was drawn inside the bowel, and a histogram (number of pixels at specific fluorescence intensity) was created using ImageJ software (http://rsb.info.nih.gov/ij/plugins/mri-analysis.html). Then, a threshold was set in the fluorescence intensity above which a pixel is counted. The total number of pixels within the threshold range was calculated at a threshold value ofis the fluorescence intensity in arbitrary units, is the number of pixels at the fluorescence intensity of , is the threshold value, and is the total umber of pixels within the threshold range . The common logarithm (log) values of were calculated and plotted as a function of . The regression line was calculated from these data sets ( and ) using Microsoft Excel 2003. For comparison of the fluorescent intensity or the “brightness,” the slope of regression line was compared among the three fluorophores.
Assessment of the Sensitivity and Specificity of GmSA-RhodG for the Detection of Peritoneal Cancer Foci
The sensitivity and specificity of GmSA-RhodG imaging for the detection of peritoneal disseminated cancer foci were studied using three tumor-bearing mice. The intraperitoneal tumor xenografts were established 14 days after intraperitoneal injection of RFP-transfected SHIN3 cancer cells suspended in of PBS in female nude mice (National Cancer Institute Animal Production Facility). Three hours after intraperitoneal injection of GmSA-RhodG diluted in PBS, spectral fluorescence images of the peritoneal membranes were obtained by Maestro. For each mouse, two different parts of the peritoneal membranes were randomly selected and spread out on a nonfluorescent plate and closeup spectral fluorescence imaging was performed. ROIs were drawn both within the nodules depicted by RFP spectral unmixed images (standard reference for cancer foci) and in the surrounding adjacent areas (standard reference for noncancerous foci), and the average fluorescence intensity of each ROI was calculated both on the RFP and the GmSA-RhodG spectral unmixed images using commercial software (Maestro version 2). The number of ROIs drawn in the noncancerous areas was the same as that drawn in the cancer foci. All visible nodules with short-axis diameters and average fluorescence intensities (a.u.) on RFP spectral unmixed images were included for analysis, because the minimum possible diameter for ROI drawing was and the minimum visible fluorescence intensity was 2 (a.u.). Then, additional ROIs were drawn in the nodules depicted only by the spectral unmixed GmSA-RhodG images to count the number of false positive lesions. The average fluorescence intensities of false positive foci were calculated both on the RFP and the GmSA-RhodG unmixed images. Positivity for GmSA-RhodG was defined as average fluorescence intensity (a.u.), whereas for a negative for GmSA-RhodG was defined as average fluorescence intensity (a.u.) on the spectral unmixed GmSA-RhodG images. The numbers of foci positive for both GmSA-RhodG and RFP, negative for both GmSA-RhodG and RFP, and positive only for GmSA-RhodG or RFP were counted. Sensitivity of GmSA-RhodG for the detection of peritoneal cancer foci was defined as the number of peritoneal foci positive for both GmSA-RhodG and RFP divided by number of peritoneal foci positive for RFP. Specificity of GmSA-RhodG was defined as the number of peritoneal foci negative for both RFP and GmSA-RhodG divided by number of peritoneal foci negative for RFP.
Optical Characteristics of Av-RhodG, GmSA-RhodG, and BSA-RhodG
Prior to in vivo study, preinjection fluorescence intensities as well as the emission spectra of Av-RhodG, GmSA-RhodG, and BSA-RhodG in PBS were assessed. The fluorescence intensities of Av-RhodG, GmSA-RhodG, and BSA-RhodG were 181, 211, and 180 (a.u.), respectively (Fig. 1 ). The emission spectra of these three probes were comparable with the same emission peak at (Fig. 1). GmSA-RhodG showed the highest and BSA-RhodG showed the lowest fluorescence intensity. The difference of fluorescence intensity between BSA-RhodG and GmSA-RhodG was within 15% of GmSA-RhodG fluorescence intensity.
Flow Cytometry Analysis Demonstrates that GmSA-RhodG Cellular Uptake is Significantly Higher than Av-RhodG
At 1, 3, and after incubation with Av-RhodG, GmSA-RhodG, or BSA-RhodG, one-color flow cytometry was performed. GmSA-RhodG showed a significant rightward shift ( order shift) as compared with SHIN3 control cells at 3 and after incubation, while BSA-RhodG showed a minimal shift at all time points [Fig. 2a ]. The percent of positive cells (M1) in Av-RhodG, GmSA-RhodG, and BSA-RhodG were 9.3, 96.4, and 2.1% for ; 15.3, 99.8, and for , 39.4, 99.8, and 2.8% for , respectively. The rate of uptake, as shown by the slope of regression lines calculated from the data sets of incubation times and common logarithm values of the MFI, were 0.043, 0.147, and 0.004 (a.u./h) for Av-RhodG, GmSA-RhodG, and BSA-RhodG, respectively. The absolute MFI values at all time points and the increase rates of MFI uptake values were consistently higher for GmSA-RhodG than Av-RhodG, whereas BSA-RhodG showed little or no increase in MFI values. These results indicate that the GmSA-RhodG more effectively accumulates within SHIN3 cancer cells and that the accumulation of GmSA-RhodG is mediated by galactosylation on albumin because the chemically similar BSA-RhodG had substantially reduced uptake.
Competitive Inhibition of Binding between GmSA-RhodG and Av-RhodG
To assess whether Av-RhodG and GmSA-RhodG compete for the same receptor, the D-galactose (asialoglycoprotein) receptor on the SHIN3 cancer cell, the competitive inhibition of Av-RhodG binding and internalization was evaluated by adding a large amount of GmSA prior to incubation with Av-RhodG. Flow cytometry analysis demonstrated that in the absence of GmSA, the MFI of Av-RhodG in SHIN3 cancer cells increased from 4.9 to 77.5 (a.u.) during 3-h incubation, whereas, when GmSA was administered prior to Av-RhodG, the MFI only increased from 4.9 to 13.0 (a.u.) (83% reduction) [Fig. 2b]. The percent of positive cells was 99.5% for those without preincubation with GmSA, and the percentage of positive cells was 55.5% (44% reduction) for those with preincubation with GmSA. These results indicate that Av-RhodG and GmSA-RhodG compete for the same receptor of SHIN3 cancer cell.
Fluorescence Microscopy of Internalized GmSA-RhodG and Av-RhodG in SHIN3 Ovarian Cancer Cells
To study the cellular location of each probe, fluorescence microscopy and DIC imaging were performed 1 and after incubation with Av-RhodG, GmSA-RhodG, or BSA-RhodG. Fluorescence microscopy of SHIN3 cells cultured with GmSA-RhodG and Av-RhodG showed a large number of fluorescent dots within the cytoplasm as early as after incubation (Fig. 3 ). The number of fluorescent dots increased over 3-h incubation with GmSA-RhodG and Av-RhodG; however, the number of GmSA-RhodG dots was larger and more apparent than Av-RhodG dots both at 1 and after incubation. BSA-RhodG showed no fluorescent dots within the cells at any time. These results indicate that both GmSA-RhodG and Av-RhodG are internalized within SHIN3 cancer cells and are likely located within the endoplasmic vesicles or lysosomes,4 but GmSA-RhodG are more effectively internalized into the cells based on the number and size of internalized dots.
Comparison of In Vivo Spectral Fluorescence Imaging of Ovarian Cancer Implants with Av-RhodG, GmSA-RhodG, and BSA-RhodG
To compare the fluorescence intensities of the cancer implants in vivo, spectral fluorescence imaging was performed after an intraperitoneal injection of Av-RhodG, GmSA-RhodG, or BSA-RhodG. Three mice injected with each of the three probes were placed side by side on a nonfluorescent plate, and spectral fluorescence images of the abdominal cavities and closeup images of the peritoneal membranes were obtained. Both GmSA-RhodG and Av-RhodG clearly depicted the aggregated tumor foci as well as submillimeter implants in vivo (Fig. 4 ). However, BSA-RhodG failed to depict the tumor foci due to the insufficient fluorescence. The visual assessment of spectral fluorescence images revealed that the signal intensity from the cancer implants exposed to GmSA-RhodG was higher than that of Av-RhodG.
To objectively analyze the fluorescence intensities of the peritoneal implants among the three optical probes, a ROI encompassing the peritoneal membrane and circumscribed by the colon was drawn on the unmixed RhodG fluorescence image [Fig. 5a ] and a histogram depicting the distribution of pixel intensities was created [Fig. 5b]. The dynamic range of signal intensity in the unmixed fluorescence image was set from 1 to 256 (a.u.) and the threshold value was changed from 31 to 241 in increments of 10, because the background signals, such as the normal peritoneal membrane excluding tumors and the nonfluorescent plate, were mostly less than 31 (a.u.). Then, the total number of pixels within the threshold range was calculated as a function of threshold , and a regression line was calculated in each ROI [Fig. 5c]. The slopes of Av-RhodG, GmSA-RhodG, and BSA-RhodG were , , and , respectively. These results indicate that the in vivo fluorescence of the GmSA-RhodG is higher than Av-RhodG, and the galactosylation of albumin is essential for binding ovarian cancer cells.
Sensitivity and Specificity of GmSA-RhodG In Vivo Spectral Fluorescence Imaging to Detect RFP-Transfected Peritoneal Cancer Foci
To investigate the effectiveness of GmSA-RhodG, sensitivity and specificity of spectral unmixed GmSA-RhodG imaging for peritoneal cancer foci was studied using dual color in vivo spectral fluorescence imaging of RFP DsRed2 and RhodG [Fig. 6a ]. A total of 283 peritoneal cancer foci in 3 mice were identified by the spectral unmixed RFP images [Fig. 6b]. An additional 283 corresponding ROIs were created in the noncancerous areas. No foci showed GmSA-RhodG fluorescence intensities (a.u.) among the 283 foci that were negative for RFP. However, another three foci were detected only by GmSA-RhodG images with no corresponding RFP focus (false positive foci). Thus, the spectral fluorescence imaging of GmSA-RhodG had a sensitivity of 100% (283/283) and a specificity of 99% (283/286).
Prior reports have focused on the ability of avidin-fluorophore conjugates to target lectin binding proteins, also known as asialo receptors or -galactose receptors, which are overexpressed on the cell surface of many cancers including ovarian cancer. Avidin is also readily absorbed through the peritoneal membrane, transported to the liver, and trapped by D-galactose receptors on hepatocytes,4, 5, 9 well away from the imaging field. In this manner, high target-to-background ratios can be obtained in peritoneal implants. However, avidin is highly immunogenic and, therefore, not suitable as a clinical agent.
A galactosyl serum albumin (GSA), which is conjugated galactose molecules to the amino groups of albumin with a short spacer, has been reported to have affinity for the D-galactose receptors.10 This GSA has already been approved for human use when radiolabeled with technetium-99m using a diethylenetriamine pentaacetic acid and is commercially available in Japan and Europe as means of noninvasively assessing regional liver function by imaging the binding of -labeled GSA to asialo receptors on hepatocytes.11, 12 No immunological reaction against the GSA has been reported so far as expected. Similar to avidin, unbound GSA is absorbed rapidly through the peritoneum into circulation and then is quickly trapped by the asialo receptors on the hepatocytes of the liver, resulting in decreased background fluorescence in the peritoneal cavity.10 However, galactose conjugation to amino groups lowered the isoelectric point of GSA resulting in less accumulation in the peritoneal tumors than avidin.10 In this study, to enhance preferable chemical characteristics of avidin by multiplying binding sites and maintaining its high isoelectric point characteristics, conversion of serum albumin to GmSA involves galactosamine-conjugation using 23 D-galactosamines reacted with carboxyl groups rather than amino groups on an albumin molecule. Therefore, this GmSA is not only a viable clinical alternative to avidin but may also possess advantages over avidin as well as the conventional version of GSA.
The GmSA used in this study was composed of 1 molecule of BSA with 23 attached galactose molecules. The multivalency of binding sites on a binding reagent can enhance the binding affinity and minimize the binding off-rate.13 In contrast, avidin has only four sugar chains terminated with galactose molecules on the polypeptide chains of similar molecular weight to albumin. GmSA-RhodG showed enhanced accumulation within SHIN3 ovarian cancer cells compared with Av-RhodG, probably because of the multivalency of GmSA compared to avidin. It is theoretically possible that GmSA could be conjugated with up to 91 molecules of galactosamine and this could further increase the binding of GmSA to cancer cells.14
Compared to GmSA-RhodG, BSA-RhodG, which lacked galactosamine but was otherwise a similar chemical conjugate, showed minimal accumulation within SHIN3 cells. This is strong evidence that conjugation of galactosamine is responsible for the tumor specificity and that the binding mechanism involves interactions of conjugated galactosamine and the D-galactose receptor. We were able to confirm the sensitivity and specificity of GmSA-RhodG for ovarian cancer implants (for lesions at least in size) using a RFP-transfected SHIN3 cell line. All lesions seen by RFP were also seen by GmSA-RhodG indicating a 100% sensitivity for lesions or greater. Three foci identified by GmSA-RhodG were not confirmed by RFP and were considered false positives resulting in a 99% specificity. It is possible, that these lesions were, in fact, false negatives for RFP. For instance, using the spectral imaging method, both fluorophores were excited with a blue light (445 to ), so that RhodG was excited with better efficiency resulting in higher light output than RFP. In addition, the production of RFP is not uniform among the DsRed2-transfected SHIN3 cells (data not shown), probably because of the cell cycling and the speed of proliferation. Thus, the strength of the emission light from DsRed2-transfected SHIN3 tumors showed some variations at different locations even with similar sizes. Therefore, it is possible that GmSA-RhodG identified lesions that had undetectable emissions from RFP due to their small size. We attempted to equalize the emissions from the two fluorophores using two different excitation filters (445 to for RhodG and 503 to for RFP) to excite RhoG and RFP with similar efficiency without using the spectral technique; however, different inflection ratios of these two filters did not allow us to achieve perfect coregistration of the two images especially in tiny tumors. The clear distinction between GmSA-RhodG–positive and GmSA-RhodG–negative foci as shown in Fig. 6 may indicate the high specificity of GmSA-RhodG. Although the high sensitivity and/or specificity can be determined for or larger tumors, even smaller tumors are readily detectable with this method. Lesions as small as were detectable but it is technically impossible with current image acquisition methods using Maestro to determine either the limits of size detection or the sensitivity and specificity for these nearly microscopic cancer clusters.
In view of the emphasis on near-infrared probes for in vivo imaging, it may be paradoxical that green fluorophores were used in this study. In general, green fluorophores (wavelengths 490 to ) are not ideal for in vivo imaging because of their short penetration within tissue. However, when the task is identifying surface lesions as is the case during surgery and endoscopy, green fluorophores are highly satisfactory.4 Green fluorophores not only provide sufficiently different spectra that they can usually be discriminated from background autofluorescence in the visible red spectrum but they generally have a high quantum yield resulting in very bright images compared to near-infrared probes, which are widely used for the in vivo fluorescence imaging from outside of the body.15, 16
Newly designed and synthesized GmSA-RhodG is able to identify tiny foci of peritoneal metastases due to ovarian cancer by binding to D-galactose receptors on the cell membrane. GmSA-RhodG is readily internalized within cancer cells and continues demonstrate high intensity fluorescence for at least after incubation with the agent. Using a RFP counterstain, GmSA-RhodG demonstrated a 100% sensitivity for peritoneal metastases or greater in size and had a 99% specificity. This agent, which should not elicit immunogenic responses based on prior human experience of a similar reagent, may be suitable for clinical trials in women with ovarian cancer who are undergoing cytoreductive surgery.
This research was supported by the Intramural Research Program of the National Institutes of Health, National Cancer Institute, Center for Cancer Research and funded in part with federal funds from the National Cancer Institute, National Institutes of Health, under Contract No. N01-CO-12400.