Dual-energy micro-CT of the rodent lung

Abstract

The purpose of this work is to investigate the use of dual-energy micro-computed tomography (CT) for the estimation of vascular, tissue, and air fractions in rodent lungs using a postreconstruction three material decomposition method. Using simulations, we have estimated the accuracy limits of the decomposition for realistic micro-CT noise levels. Next, we performed experiments involving ex vivo lung imaging in which intact rat lungs were carefully removed from the thorax, injected with an iodine-based contrast agent, and then inflated with different volumes of air (n = 2). Finally, we performed in vivo imaging studies in C57BL/6 mice (n = 5) using fast prospective respiratory gating in end inspiration and end expiration for three different levels of positive end expiratory pressure (PEEP). Before imaging, mice were injected with a liposomal blood pool contrast agent. The three-dimensional air, tissue, and blood fraction maps were computed and analyzed. The results indicate that separation and volume estimation of the three material components of the lungs are possible. The mean accuracy values for air, blood, and tissue were 93, 93, and 90%, respectively. The absolute accuracy in determining all fraction materials was 91.6%. The coefficient of variation was small (2.5%) indicating good repeatability. The minimum difference that we could detect in material fractions was 15%. As expected, an increase in PEEP levels for the living mouse resulted in statistically significant increases in air fractions at end expiration but no significant changes at end inspiration. Our method has applicability in preclinical pulmonary studies where changes in lung structure and gas volume as a result of lung injury, environmental exposures, or drug bioactivity would have important physiological implications.

computed tomography (CT) is one of the most used imaging modalities for the evaluation of thoracic disorders. With the recently developed dual-energy CT technique (DECT), the clinical utility of CT in the pulmonary diseases could expand even more. DECT allows analysis of the chemical composition of tissues by means of dual-energy data acquisition and tissue decomposition (14). CT data analysis with DECT is based on differences in X-ray absorption of heavy elements with energy. In clinical use, many DECT studies are already focusing on the thorax (9, 18, 19, 23–27). In the thorax, the three materials most frequently analyzed are iodine, air, and soft tissue. Recently, researchers (26) have evaluated the feasibility of using DECT for pulmonary perfusion and ventilation in the lungs using both iodinated contrast agents and xenon to provide distributions representing the local perfusion and ventilation. More interestingly, a recent study (8) has abandoned xenon for ventilation imaging in CT because the use of xenon gas is logistically demanding and requires the use of special inhalators and pressurized gas bottles. Consequently, the authors have investigated the feasibility of using DECT for evaluation of blood and air distribution in differentiation of pathological conditions in the human lung. The study found important relationships between blood and air distributions needed for the interpretation of dual-energy CT imaging of the lungs.

Similar dual-energy micro-CT studies of the lungs have never been reported in rodents, the preferred live animals for medical research, for which there are widely available reagents and genetic models of various disease. Lung imaging in rodents by micro-CT is extremely challenging because sophisticated instruments and imaging methods are required to cope with the demands of higher resolution and physiologic motion (13). Our group has been actively engaged in designing pulmonary gating techniques for micro-CT (4) as well as in developing methods for pulmonary perfusion imaging (5, 6). We have also built a dual source micro-CT system (2) that enables not only faster dynamic imaging but also dual-energy micro-CT studies.

This work specifically investigates the use of dual-energy micro-CT to provide the volumetric air, blood, and tissue fractions in the rodent lung. Such a method could facilitate preclinical studies in cardiopulmonary models of disease or help in testing new therapeutics.

MATERIALS AND METHODS

Dual-energy decomposition.

In principle, dual-energy CT can only accurately decompose a mixture into two materials. To decompose a mixture into three constitutive materials using dual-energy CT measurements, a third constituent must be provided to solve for three unknowns with only two spectral measurements. One solution is to assume that the sum of the volumes of three constituent materials is equivalent to the volume of the mixture (volume or mass conservation). We used a simple postreconstruction dual-energy CT decomposition method similar to (11) to estimate the volume fractions of blood (via iodinated contrast agent), air, and soft tissue in the rodent lung. Two lung CT data sets, each acquired with a different energy [E1 and E2 i.e., 40 and 80 peak kilovolts (kVp) for our studies], are used to find the solution to a system with three unknowns i.e., the fractions F_air, F_tissue, and iodine F_blood according to the following equations:

$$\begin{array}{c}{F}_{air\hspace{0.17em}}C{T}_{air,E1\hspace{0.17em}}+F_{tissue\hspace{0.17em}}C{T}_{tissue,E1\hspace{0.17em}}+F_{blood\hspace{0.17em}}C{T}_{blood,E1\hspace{0.17em}}=\hspace{0.17em}C{T}_{E1}\\ F_{air\hspace{0.17em}}C{T}_{air,E2\hspace{0.17em}}+\hspace{0.17em}{F}_{tissue\hspace{0.17em}}C{T}_{tissue,E2\hspace{0.17em}}+\hspace{0.17em}{F}_{blood\hspace{0.17em}}C{T}_{blood,E2\hspace{0.17em}}=\hspace{0.17em}C{T}_{E2}\\ F_{air\hspace{0.17em}}+\hspace{0.17em}{F}_{tissue\hspace{0.17em}}+\hspace{0.17em}{F}_{blood\hspace{0.17em}}=\hspace{0.17em}1\end{array}$$

The values for CT_air,E1, CT_tissue,E1, CT_blood,E1, and CT_air,E2, CT_tissue,E2, CT_blood,E2 are mean CT numbers measured in regions of interests with 100% content of these materials at the two energies and are used to create a sensitivity matrix. The material fractions are obtained using the inverse of the sensitivity matrix as in the next equation:

$$\left[\begin{array}{c}{F}_{air}\\ F_{tissue}\\ F_{blood}\end{array}\right]={\left[\begin{array}{ccc}C{T}_{air,E1}& C{T}_{tissue,E1}& C{T}_{blood,E1}\\ C{T}_{air,E2}& C{T}_{tissue,E2}& C{T}_{blood,E2}\\ 1& 1& 1\end{array}\right]}^{-1}\left[\begin{array}{c}C{T}_{E1}\\ C{T}_{E2}\\ 1\end{array}\right]$$

The pipeline for decomposition is presented in Fig. 1. Before decomposition, a series of processing steps were applied to the 40- and 80-kVp sets and included three-dimensional (3D) bilateral filtering (BF; Ref. 28) and affine registration. Similar to the work of Tomasi et al. (28), 3D BF is applied to each volume to reduce noise while preserving edges. BF is a nonlinear filter that replaces the intensity of each voxel with a weighted average of neighboring voxels, assigning the highest weights to voxels that are close to the voxel being corrected both in space (domain) and intensity (range). In our implementation of BF, the domain was a 3 × 3 × 3 cube (x, y, z) and the range was a Gaussian with a mean equal to the intensity of the voxel being corrected and a standard deviation of 136.59 Hounsfield units (HU). A second iteration of BF was then applied with an SD of 58.99 HU to further suppress noise. The filtration protocol was identical for both 40- and 80-kVp data. The two energy sets were acquired simultaneously but by two separate orthogonal imaging chains using a dual-source micro-CT system (2). A geometric calibration that we have previously developed (16) ensures that the two sets are reconstructed in the same system of reference. However, since every voxel should be matched in the dual-energy sets, a postreconstruction dual-energy method requires a perfect registration. We therefore corrected any remaining errors of geometric calibration by performing an affine registration between 40 and 80 kVp using an open source, ITK-based toolkit called ANTs [Advanced Normalization Tools (1); http://picsl.upenn.edu/ANTS/]. To restrict analysis to the lung parenchyma only, we used the correlation based segmentation tools built into Avizo (specifically, “Correlation Histogram”) to extract the bulk of the lungs. This preliminary lung mask was then filled in using morphological closing operations. This binary mask was required to isolate the lungs for measuring material fractions and volumes. The mask also sped up the dual-energy decomposition by restricting the decomposition to voxels within the mask. The decomposition by matrix inversion was followed by correction of invalid values using a lsqlin function available in MATLAB (MathWorks, Natick, MA) that constrains possible fractions to [0,1] while also imposing that their sum to be 1. The outputs of the processing pipeline are 3D material fraction volumes corresponding to air, blood, and tissue. A combined visualization of these maps was achieved by mapping each component as green for air, red for blood, and blue for tissue using scripts available in ImageJ (rsbweb.nih.gov/ij/).

Fig. 1.

Processing pipeline for dual-energy decomposition into 3 material fractions corresponding to air, tissue and blood (i.e., via iodinated contrast agent). Before decomposition, the 2 original sets acquired at 80 and 40 kVp are filtered via bilateral filtration and registered in ANTS to ensure superior performance of our approach. A binary mask of the lung parenchyma is used to restrict decomposition. Outputs of our pipeline are three-dimensional (3D) fraction maps that can be combined as red (R), green (G), and blue (B) components for visualization.

Simulations.

The accuracy of decomposition was evaluated through simulations using a phantom with three materials (air, blood, and tissue). The phantom contained random triangular structures obtained using the delaunay function in MATLAB (see Fig. 2). Each triangle was assigned a set of random material concentrations of iodine solutions, soft tissue (i.e., water), and air, and these fractions sum to 1 at every voxel. Two homogeneous disk structures containing only iodine and only soft tissue were also included. These were included in the simulations to provide a standard for the accuracy of the decomposition. The background was considered air. The iodine concentration in unmixed regions was 50 mg/ml. The phantom was used to create projections via a process similar to an X-ray beam irradiation that included X-ray tubes with a tungsten-based spectra, an aluminum filter (0.7 mm), and a Gd₂O₂S scintillator. These simulations used the Spektr package (22). A total of 300 projections were used for each peak kilovolt scanning. Poisson noise was added to the projections to match the noise levels of typical micro-CT data (63.35 HU in the 40-kVp image and 43.14 HU in the 80-kVp image) after reconstruction using filtered back projection with a Ram-Lak filter. The selection of the two scanning voltages maximizes the performance of dual-energy decomposition. For the dual source micro-CT system, 40 kVp is the lowest voltage allowed by the micro-CT system's X-ray generator. This voltage provides minimum enhancement while values in the range of 70 or 80 kVp provide maximum enhancement when imaging iodine solutions (3). BF was also applied. The phantom did not require registration. The decomposition produced the fraction maps that were next compared with the known fractions to calculate errors. We have also plotted the true vs. the estimated fractions of each component and computed correlation coefficients. To assess results in a quantitative manner in simulations, we used absolute errors measured between the true and computed material fractions for each pixel. We present plots of histograms of the absolute errors for all three material fractions, individually and combined. The accuracy was computed as one minus the modulus of the mean absolute error. We have run the simulations 10 times with different phantoms to assess repeatability.

Fig. 2.

Simulation protocol and results. Phantom made of 278 random triangular structures with components of air, blood containing iodine, and tissue was virtually irradiated to create noisy projections used to reconstruct computed tomography (CT) images at 40 and 80 peak kilovolts (kVp). CT images were preprocessed to lower their noise using bilateral filtering (BF). Next, the resulting images were decomposed to estimate the 3 material fractions. Errors of the decomposition were analyzed.

Ex vivo lung experiments.

All animals were handled in accordance with good animal practice as defined by the relevant national and/or local animal welfare bodies, and all animal work was approved by the Institutional Animal Care and Use Committee of Duke University Medical Center. Following simulations, we performed two experiments using intact lungs carefully extracted from the thorax of Sprague-Dawley rats. The lung vasculature was filled with iodinated contrast agent (Isovue 370; Bracco Diagnostics, Princeton, NJ) solution. Four different levels of lung inflations were created by injecting volume increments of air. Initially, the lungs were deflated, and we then injected 1.5 ml of air. Next, we added 1 ml of air, two times. At each level of lung inflation (deflated, 1.5, 2.5, and 3.5 ml), dual-energy scanning and decomposition was performed. In each scan, two sets of projections were acquired, one set at 40 kVp, 250 mA, 10-ms exposure, and another set at 80 kVp, 160 mA, 10-ms exposure per projection. Each set consisted of 300 projections. A single scan required ∼5 min to complete. Since each experiment consisted of four scans for different levels of lung inflation, the total experiment time was ∼20 min. The fraction images were used to compute global lung volumes of air, tissue, and blood by summing the fractions over the whole lung parenchyma (provided by semiautomatic segmentation) and multiplication with the voxel size. All tomographic data have been reconstructed with an isotropic voxel size of 88 μm. The ex vivo lung case allowed for a controlled experiment in which both visual inspection and quantitative evaluation are possible.

In vivo experiments in mice.

A small number of C57BL/6 mice (n = 5) were used for proof-of-concept in vivo experiments. These mice were anesthetized with ketamine and xylazine (dose of 100 mg/kg ketamine and xylazine 5 mg/kg) administered intraperitoneally for induction. Additional doses were given at ∼20 min, intravascular with 50% of initial dose via tail vein. The animals were not removed from the scanner to administer this maintenance sedation. Mice were injected via tail vein with a liposomal blood contrast agent (dose: 0.4 ml/25 g mouse) with a concentration of iodine of 120 mg/ml (10). This contrast injection ensured a prolonged blood enhancement of ≥400 HU and as high as 650 HU. The half-life of the liposomal blood pool contrast agent is ∼40 h (10); thus there is a very limited variation in the enhancement during a study. For example, over a 30-min study with six scans, the mean blood measured in the aorta was 639 ± 11 HU (mean ± SD) with a coefficient of variation of only 1.7%. The mice were intubated and mechanically ventilated with tidal volume of 0.4 ml.

Micro-CT sets were acquired using a fast prospective gating strategy (12) at the same imaging parameters as in the ex vivo experiments. In essence, fast prospective gating allows for the acquisition of two projections at each angle, one at end inspiration and one at end expiration. Since we use a dual source micro-CT system, the 40- and 80-kVp projections were acquired simultaneously. An in vivo scan corresponding to each PEEP level acquired at end inspiration and end expiration required 10 min. The total scan time to acquire all sets for three different PEEP levels was 30 min. A total of 12 sets were reconstructed (two energies of 80 and 40 kVp, each in end inspiration and expiration and three levels of PEEP: 0, 1, and 2 cm water). The radiation dose associated with the vivo mouse studies was 0.8 Gy. Changes in PEEP were achieved using an external line connecting the ventilator valve to a water vial. Our hypothesis for these experiments was that our imaging method would be sensitive enough to detect the expected small changes in air fraction and total lung volume.

Statistical analysis.

Statistical analysis was performed for in vivo experimental data by univariate ANOVA using MATLAB to compare the changes of the end expiration or end inspiration over different PEEP levels. Data are presented as means ± SD. A P value <0.05 was considered statistically significant.

RESULTS

The value of BF was confirmed both by simulations and its application on real data. The noise was reduced from 65 HU in the 40-kVp image to 24 HU and from 43 HU in the 80-kVp image to 13 HU after BF. Figure 3 presents the results of the decomposition in a simulated phantom with and without BF. Plots of true vs. estimates of each material fraction are shown in Fig. 3, A and C. The correlation coefficients for air, blood, and tissue have increased from [0.91, 0.91, 0.83] before BF to [0.91, 0.93, 0.88] after BF. The regression lines show a bias since they do not start from the origin, and this bias was found to be associated with the tomographic reconstruction errors. The slopes of the regression lines corresponding to air, blood and tissue were [0.75, 0.88, 0.67] before and [0.70, 0.75, 0.64] after BF. As indicated by these results, there is a decrease in the slope values and an increase in the correlation coefficients after BF. This is expected because the BF is just a special type of smoothing filter, which will push the few extreme values towards the center in the plots (Fig. 3, A and B). We also present the histograms of the absolute errors (Fig. 3, B and D). Note the fact that these histograms are centered around zero. To verify that BF introduces no bias, we used a t-test and compared the distributions of differences at all pixels and all materials between the true phantom values and the estimated values, with and without BF. The means of two distributions were not statistically significant. However, the effect of the BF can be seen in the clear reduction of SD of these histograms of the absolute errors. When using BF, the SD of the absolute errors for all combined fractions was 5%. Therefore, according the Rose criterion that specifies that two different structures can be discriminated if their difference in value are at least three times the noise given by SD (20), the minimum difference that we could detect is on the order of 15%. The results of running the simulations 10 times with a different phantom before BF provided the following mean accuracy values: air (87%), blood (86%), and tissue (76%). The absolute accuracy in determining all fraction materials combined without BF was 83%. These values increased after BF to air (93%), blood (93%), tissue (90%), and all combined (91.6%). The coefficient of variation over the 10 runs with different phantoms was 2.5% therefore these accuracy values were highly reproducible.

Fig. 3.

Plots of true vs. estimated fractions for air, blood, tissue in all triangles of the simulated phantom, and their regression lines (A and C). Histograms of absolute errors for each material individually and all combined are shown by B and D. We present data before (A and B) and after BF (C and D). Combined material fractions uses all errors for the 3 materials in the histogram plot and essentially is equivalent to an average of these errors.

BF applied predecomposition increased performance both in simulations and the animal studies. As visually confirmed by Fig. 4, the result of the 3D BF is a decrease in noise by 62% at 40 kVp and 54% at 80 kVp without losing edges (see subtraction image).

Fig. 4.

Performance of our decomposition has been increased by applying 3D BF. Same axial micro-CT slice is shown before (A) and after (B) BF. As shown by the difference image (C), the noise has been reduced while edges have been relatively well preserved. Some errors appear around the heart and large airways. We believe these errors were introduced by the tomographic reconstruction.

Figure 5 presents results obtained during the ex vivo rat lung imaging studies. Visually, the results confirm the enlargement of the lung due to air injections (Fig. 5A). Note the changes in air fractions that reflect changes associated with the four different levels of lung inflation. As shown by the plot in Fig. 4B, the estimated volumes of air track the volumes of air added to the lungs for the last three increments. The first increase from deflated lung by injecting 1.5 ml is less accurate. As expected, the soft tissue and blood volumes have very limited variations over the course of the experiment. This is expected since no iodine or soft tissue were added or subtracted.

Fig. 5.

A: original CT images (80 kVp) and decomposed fraction maps for an ex vivo lung experiment with different levels of lung inflation. B: whole lung volume estimation of air, tissue, and blood over different inflation levels during experiments in (n = 2) animals. There are relatively flat values for blood and tissue components.

Finally, some representative images from our murine in vivo experiments are shown in Fig. 6. Both end inspiration and expiration images are shown for three levels of PEEP (0, 1, and 2 cmH₂O). Experimentally, an increase in PEEP should increase the air volume mainly at end expiration, but it may also increase the air volume at peak inspiration depending on where the lung is set on its pressure-volume curve. The increased inflation due mostly to the air component with added PEEP is illustrated visually by the 3D volume rendering in Fig. 7 (see arrow). Our statistical analysis detected significant differences (P < 0.05) in end expiration for the three PEEP levels (Fig. 8), although some changes also occurred at end inspiration that were not statistically significant (see Table 1 for P values). The air volume at end inspiration has been used to provide an in vivo repeatability measure. As indicated in Table 1, the mean values of these volumes over the three PEEP levels were not statistically different (P = 0.84). The mean coefficient of variation (σ/μ) for these three PEEP levels was 8%. In Fig. 6, lower levels of air fractions are seen in end expiration images as increased blue color and more soft tissue fraction (blue) and less air (green) are present. The increase in PEEP reduces the differences as more air is present in the lungs at end expiration (compare arrows between 0 and 2 cmH₂O). The total lung volumes for different PEEP levels were only significantly different at end expiration (P = 0.04). Small but significant changes were also found for tissue at end inspiration (see Table 1). The subtraction of the air fractions between inspiration and expiration provides another physiologically relevant measure i.e., the tidal volume. For our in vivo mice experiments, the mean tidal volumes actually fell from 0.22 ml for 0 cmH₂0 to 0.17 ml for 1 cmH₂O PEEP and to 0.15 ml for 2 cmH₂O PEEP.

Fig. 6.

Axial and coronal micro-CT slices in a mouse during in vivo experiments at end expiration, end inspiration and with overlaid decomposition information (green for air, red for blood, and blue for soft tissue). Three levels of positive end expiratory pressure (PEEP; 0, 1, and 2 cmH₂O) are shown. Note the changes between end inspiration and end expiration at PEEP 0 cmH₂O. Increases in PEEP levels make these changes less obvious (compare arrows between the 0 cmH₂O PEEP with the 2 cmH₂O PEEP).

Fig. 7.

3D volume rendering of the lungs with combined visualization. Both inspiration (insp) and expiration (exp) images are shown for all 3 PEEP levels. Note the inflation of the lungs with added PEEP (see arrows).

Fig. 8.

Data are means ± SD for the 3 lung material fractions (A) and absolute volumes obtained via multiplications between the fractions and lung volume (B). *Significance levels (P < 0.05) for air in end expiration and tissue in end inspiration.

Table 1.

P values in ANOVA single factor analysis for different PEEP values

P Values	Inspiration	Expiration
Air	0.8452	0.0172
Blood	0.5172	0.0945
Tissue	0.0145	0.0536

Each test was performed as for end expiration or end inspiration case. PEEP, positive end expiratory pressure. Note P < 0.05 values for the air component at end expiration and for the tissue component at end inspiration.

DISCUSSION

Historically, ventilation-perfusion imaging of the lungs has been used mainly in combined imaging modalities in pulmonary scintigraphy, in which two examinations are performed separately for evaluation of the distribution of perfusion and ventilation in vivo. Established imaging methods rely mainly on rare elements for evaluation of the ventilation, for example, Kr in ventilation scintigraphy, hyperpolarized He in magnetic resonance (7), or Xe in CT (26). Micro-CT has been also used with xenon imaging to provide ventilation information (17, 21) in rodents. The use of gases complicates the setup during imaging and reduces the throughput. When the pulmonary parenchyma is imaged by means of CT with intravenous application of a contrast medium, the final density of the pulmonary parenchyma depends on the iodine and air tissue distribution; therefore, it is dependent on substances the distribution of which in the pulmonary parenchyma is expressed by thoracic gas and pulmonary vascular perfusion.

Our present study has focused on the application of dual-energy micro-CT to determine 3D air tissue and blood distributions. Our three material decomposition is based on the assumption of no variations in density for the three materials over the course of the study. This assumption may not be completely valid for air in the lungs, but we believe that variations in air density would be very limited and, therefore, would not affect the validity of our analysis. The use of a blood pool contrast agent facilitates the perfusion imaging and reflects in fact pulmonary blood volume information. A sensitivity matrix can be created even without a special need for calibration vials. Blood and air values are easily sampled in regions containing 100% blood (such as large vessels) and air. In previous work (10), we determined that postcontrast injection the muscle (i.e., tissue) increases its CT number by only 4% relative to blood enhancement. Although small, we propose that this value could be subtracted from the muscle value to eliminate any bias. With very fast imaging as done with clinical CT systems, the same methodology could be applied with conventional contrast agents. Our results suggest that such a method based on dual-energy micro-CT can be applied as an imaging tool for lung research with estimated limits of detectability of changes of material fractions on the order 15%. Postprocessing via BF has shown that the performance can be improved with noise reduction. A perfect match between the two energy 3D sets is also essential, and it has been provided via registration.

Via controlled perturbations, we have shown both ex vivo and in vivo that we can reliably sense differences associated with these added perturbations. PEEP experiments in particular validated our method in vivo in mice. While our air fractions distribution do not provide ventilation information per se, as also shown by Ferda et al. (8) in the clinical domain, these images can be used to localize suspected regions of gas exchange impairment by lung disease or the effects of therapeutic agents.

The radiation dose associated with our in vivo studies was relatively high (0.8 Gy) and could become a limiting factor for longitudinal studies. However, our in vivo studies were designed for repeated end inspiration and end expiration acquisitions. In many situations, only one of these respiratory phases would be required, resulting in half the radiation dose per study. Furthermore, more sophisticated image reconstruction based on iterative algorithms that we are currently developing (15) has the potential to reduce dose while maintaining image quality.

Physiologically, there is an ongoing need for near real-time assessment of dynamic ventilation-perfusion relationships to evaluate gas exchange function in small animals in vivo, especially in mice, where the available molecular and pathological information now routinely exceeds the resolution of the available physiological tools. For instance, there are dozens of genetically modified mouse strains that show changes in lung development, aging-related lung pathology, or undergo unique changes in lung structure, including asthma, emphysema, and pulmonary fibrosis after various exposures or interventions. The physiological information available to interpret the impact of these structural changes in mice is limited mainly to arterial blood gas determinations and to global measurements of respiratory system compliance or resistance, with essentially no local intrapulmonary spatial resolution to provide information on heterogeneity. Dual-energy micro-CT is therefore a promising tool to provide this much needed regional functional information on ventilation-perfusion relationships in these settings.

Conclusions.

The advantage of our imaging method for preclinical lung research consists mainly of the ability to both provide volumetric distribution information and global metrics. In conclusion, our method has potential in pulmonary studies where various physiological changes can occur as a result of genetic changes, lung disease, or environmental or drug exposure.

GRANTS

Support for the study was provided by NIH NCRR/NIBIB grant p41-EB015897, with additional support from NCI grant U24-CA092656.

DISCLOSURES

No conflicts of interest, financial or otherwise are declared by the author(s).

AUTHOR CONTRIBUTIONS

Author contributions: C.T.B., S.M.J., and C.A.P. conception and design of research; C.T.B., X.G., and C.D.M. performed experiments; C.T.B., S.M.J., and C.A.P. interpreted results of experiments; C.T.B. and S.M.J. prepared figures; C.T.B. and C.A.P. drafted manuscript; C.T.B., D.P.C., S.M.J., and C.A.P. edited and revised manuscript; C.T.B., X.G., D.P.C., S.M.J., C.D.M., and C.A.P. approved final version of manuscript; D.P.C. analyzed data.