Main

Processes of human brain development are integral to the understanding of the long-term impact of neurological disorders afflicting the immature brain. Animal models provide a means to investigate developmental processes in a controlled manner and to model the effects of disease and injury on the immature brain. Dobbing and Sands (1) proposed that the swine brain may represent an appropriate model for the immature human brain based on the timing of the brain growth spurt (rapid stages of brain growth), which is recognized as a period of increased vulnerability to injury. Additional benefits of the swine model for developmental neuroscience include similar histology, cortical folding, and gray-matter (GM)-to-white-matter (WM) ratio to humans (1). Moreover, recent advances in transgenic neural gene manipulations and the ability to model brain injury and neurologic disorders have also supported the increasing use of swine as a model for neuroscience and neurosurgical research (2,3).

Ontogeny of the brain typically focuses on structural changes during development; however, these changes may also be linked to cerebrovascular development based on the observed association between rapid stages of brain growth and increased cerebral perfusion that is believed to be required to match increased metabolic needs (4). Given that many pathologies of the immature brain involve compromised cerebral perfusion (e.g., neonatal encephalopathy and intraventricular hemorrhage), there is a need to characterize both cerebrovascular and structural changes occurring in animal models of early human brain development.

MRI is emerging as the prominent modality for noninvasive assessment of morphology and microstructure in the developing human brain and offers the potential to noninvasively assess cerebrovascular development. High-resolution three-dimensional MRI provides measures of brain volume and cortical folding, whereas diffusion tensor imaging (DTI) offers the ability to track microstructural organization and composition throughout the stages of development (5–10). Cerebral blood flow (CBF) levels at different stages of development have also been measured noninvasively by arterial spin labeling (ASL), an MR technique that exploits endogenous blood water as a contrast agent (11). Another potentially important parameter, which is not well-characterized during development, is cerebrovascular reactivity (CVR), which can be assessed using MRI measures of the CBF response to a vasoactive stimulus (e.g. CO2). Together, these MRI measures provide means to comprehensively assess structural and cerebrovascular development.

In this study, we aim to further characterize the developing swine brain as a model for translational neuroimaging research by quantifying microstructural and cerebrovascular changes during maturation via MRI. We present quantitative measures of brain volume changes using 3D high-resolution MRI, microstructural changes using DTI, baseline CBF, and finally, blood oxygenation level dependent (BOLD) signal changes in response to precise end-tidal (i.e. end exhaled) partial pressure of CO2 (PetCO2) to measure CVR in swine of various postnatal ages.

METHODS

Our institutional animal care committee approved this study and all animal procedures were conducted according to the Canadian Council on Animal Care. Thirteen 1-wk to 3-mo-old Yorkshire pigs were imaged on a 1.5-Tesla MRI scanner (General Electric Healthcare, Milwaukee, WI). An eight-channel receive-only head coil provided radiofrequency detection, and the body coil was used for radiofrequency transmission.

Animal preparation.

Anesthesia was induced with an intramuscular injection of 0.2 mL/kg Akmezine and switched to 3% isoflurane for intubation and insertion of a catheter into the ear vein for a constant delivery of i.v. anesthesia (22 mg/kg/h ketamine with 1 mg/kg/h midazolam). Paralysis was maintained with pancuronium (0.2 mg/kg bolus and 1 mg/kg/h constant infusion) to assist mechanical ventilation. Animals were under anesthetic for ∼2–3 h.

Anatomical imaging.

High-resolution 3D T1-weighted anatomical images were acquired using a fast spoiled gradient-recalled echo sequence: echo time (TE) = 4.2 ms, repetition time (TR) = 8.46 ms, flip angle = 20°, field-of-view (FOV) = 180 mm, slab thickness = 180 mm, and matrix size = 192 × 192 × 120. DTI was performed using 25 unique diffusion-encoding directions (b = 700 s/mm2) and one acquisition (b ≈ 0 s/mm2) without diffusion weighting. The DTI data were acquired using spin echo echo-planar imaging with parallel imaging (R = 2) and the following parameters, TE = 80 ms, TR = 5.4 s, FOV = 160 mm, matrix size = 128 × 128, slices = 16, and slice thickness = 4.5 mm. Pixel-wise mean diffusivity (MD) and fractional anisotropy (FA) images were calculated from the principal eigenvalues of the diffusion tensor.

Computer-controlled gas delivery system.

Cerebrovascular imaging was performed with precise control of the end-tidal partial pressures of CO2 and O2 using a model-driven prospective end-tidal targeting (MPET) system (RespirAct; Thornhill Research Inc., Toronto, Canada) (12) integrated with mechanical ventilation by use of a purpose-built secondary breathing circuit. Expired gases were sampled from the endotracheal tube and the continuous partial pressure of CO2 and O2 were digitally recorded (LabView; National Instruments Corporation, Austin, TX). Breath-by-breath PetCO2 and end-tidal partial pressure of O2 were identified from the CO2 and O2 partial pressure waveforms in real time using the same customizable software and confirmed manually post hoc.

Arterial spin labeling.

For baseline CBF measurements, we targeted and maintained PetCO2 at 40 mm Hg via the MPET system. ASL data were collected using single-shot gradient-echo spiral imaging using the flow-sensitive alternating inversion recovery technique (13) with a label inversion time = 1700 ms and the following imaging parameters: FOV = 160 mm, matrix size = 64 × 64, slices = 6, slice thickness = 5 mm, slice separation = 1 mm, TE = 4 ms, and TR = 3.75 s. Tissue water signal was suppressed to mitigate effects of physiological noise and motion (14), and an arterial saturation pulse was applied (850 ms postlabel) to define the label duration. Two separate acquisitions of 64 alternating “tag” and “control” image pairs were collected. Perfusion-weighted images (ΔM) were generated by successive pair-wise subtraction of tag and control images. Four proton-density (Mo) images were also acquired separately.

To account for brain tissue R1 relaxation rate (R1tissue) differences, a Look-Locker approach was used to collect a train of 11 low flip angle (20°) T1-weighted images at inversion times separated by 300 ms. The inversion-recovery signal was fit to extract pixel-wise R1tissue values. Custom Matlab and IDL (Interactive Data Language, Research Systems, Boulder, CO) scripts were used for ASL postprocessing. The CBF was calculated from the mean of the ΔM images using the following equation:

where α (0.85) accounts for signal decrease caused by imperfection in the background-suppression inversion pulses, λ (0.9 mL/g) is the water partition coefficient, R1a (0.8 s−1) is the longitudinal relaxation rate of arterial water (15), ΔR1=R1a−R1tissue, TI is the inversion time (adjusted for each slice), τ (1130 ms) is the tissue transit time for newborn pigs (16), and d is the delay between the tag and arterial saturation pulses. Voxels with CBF >200 mL/100 g/min were considered to be dominated by vascular effects and removed.

BOLD CVR.

BOLD imaging was performed in two separate acquisitions using single-shot gradient-echo echo-planar imaging during MPET controlled iso-oxic PetCO2 cycling between 40 and 55 mm Hg. PetCO2 changes were delivered in a repeated square-wave pattern with normocapnia followed by hypercapnia in 60 s steps (total = 9 minutes). Imaging parameters included: FOV = 160 mm, matrix size = 64 × 64, number of slices = 14 − 16, slice thickness = 4.5 mm, slice separation = 0.5 mm, TE = 35 ms, and TR = 2 s. Image preprocessing included spatially smoothing (Gaussian kernel full-width half-maximum = 5 mm), motion correction, and high-pass filtering using FEAT (FMRIB Expert Analysis Tool, v 5.98, FMRIB, http://www.fmrib.ox.ac.uk/fsl, Oxford University, United Kingdom). We temporally aligned the PetCO2 and BOLD signals by visually estimating the delay between the PetCO2 and whole-brain BOLD signal changes. Pixel-wise CVR was determined using the slope for the linear regression between the BOLD and PetCO2 signals averaged between the two acquisitions. A temporal derivative was included to improve the match of the BOLD and PetCO2 data on a per-voxel basis by allowing small temporal shifts to be included in the regression model.

Image analysis.

Six brain regions were investigated, including cortical GM (CGM), cortical WM (CWM), deep GM (DGM), deep WM (DWM), mesencephalon, and cerebellum. The six brain regions were each segmented from the rest of the brain using a combination of automatic (FMRIB Automated Segmentation Tool, FAST) (17) and manual segmentation of the anatomical images (Fig. 1). The DTI, ASL, and BOLD-CVR parameter maps were transformed to the anatomical images via low resolution T1-weighted images collected at corresponding slice locations (FMRIB Linear Image Registration Tool, FLIRT) (18) and mean parameter values were extracted for each region. Limited ASL spatial coverage meant mesencephalon and cerebellum CBF estimates were not available.

Figure 1
figure 1

Saggital T1-weighted anatomical MRI with tissue regions-of-interest overlaid, including: 1, CGM; 2, CWM; 3, DGM; 4, DWM; 5, mesencephalon; and 6, cerebellum. The image was cropped to isolate the brain.

Full size image

Statistical analysis.

Because of the uncertainty in the animals' gestational and postnatal ages, we opted to use body weight to assess developmental changes. For the analyses, body weight was the dependent variable and the following measures were independent variables: brain volume, MD, FA, CBF, and CVR. Nonparametric Spearman rho was computed for each parameter to determine body-weight dependencies for each independent variable without assuming a specific model for the relationship. Next, regression analysis determined the relationship between body weight at scan and brain volume, MD, FA, CBF, and BOLD-CVR. We compared linear, logarithmic, and exponential model functions for each parameter and determined, based on the greatest Pearson correlation coefficient (r), that the logarithmic function provided the optimal fit for the majority of brain regions for all MRI parameters investigated. In total, the logarithmic fit yielded greater r values compared with the linear model for 20 of the 28 regressions (mean r increase = 0.04 ± 0.07). Therefore, data were fit to the logarithmic function:

All statistical analyses were performed using SPSS v 11.0 (SPSS Inc., Chicago, IL). Results were considered statistically significant for p < 0.05. As this is a pilot study providing descriptive characteristics of developmental changes, the critical p-value was not corrected for multiple comparisons.

RESULTS

The median [range] body weight at scan for the different postnatal age groups was as follows: 1 wk old = 3.35 kg [2.3–4.4 kg] (n = 2); 1 mo old = 6.85 kg [5.2–9.4 kg] (n = 4); 2 mo old = 12.6 kg [10–14.8 kg] (n = 3); and 3 mo old = 24.6 kg [18.4–43.8 kg] (n = 4).

Nonparametric analysis demonstrated that total tissue volumes (whole brain, GM, and WM) exhibited body-weight dependencies. Follow-up parametric regression demonstrated a significant logarithmic increase in brain tissue volumes versus body weight (Fig. 2). Significant logarithmic relationships were also observed between brain volume and body weight for all individual brain regions investigated, except the mesencephalon (Fig. 3). Table 1 provides regression coefficients and measures of fit quality (r) for total brain volumes and individual regions.

Figure 2
figure 2

Log-linear regression of whole brain volume (•), total GM (), and total WM (○) with respect to body weight. The solid lines represent the log-linear regression.

Full size image
Figure 3
figure 3

Log-linear regression of body weight and brain volume for each tissue region investigated: (A) CGM (•) and CWM (○), (B) DGM (•) and DWM (○), and (C) mesencephalon (▪) and cerebellum (□).The solid lines represent the log-linear regression. Note that the scales for panels A, B, and C are different.

Full size image
Table 1 Logarithmic regression coefficients and fit quality measures for brain tissue volume vs body weight (N = 13)
Full size table

Representative MD and FA images for 1-wk, 1-mo, 2-mo, and 3-mo-old pigs are provided in Figure 4. The DTI measurement of MD did not reveal a consistent pattern of developmental changes across the brain regions investigated (Fig. 5). Nonparametric correlation analysis revealed that MD was related to body weight in all regions except the CGM and DGM. Based on the parametric regression, a negative logarithmic relationship between MD and body weight existed in the CWM; whereas, MD in the DWM, mesencephalon and cerebellum exhibited a positive logarithmic relationship (Table 2). The FA exhibited a consistent increase with body weight across all regions after one outlier subject was removed before the analysis. The decision to remove this subject was based on poor FA map SNR, as FA values are sensitive to DTI noise levels. Parametric regressions revealed a consistent positive logarithmic relationship between FA and body weight across all six regions (Fig. 6, Table 2).

Figure 4
figure 4

Representative MD maps for each age group: 1-wk-old (A), 1-mo-old (B), 2-mo-old (C), and 3-mo-old (D) pigs. Corresponding FA maps (E–H) are also provided.

Full size image
Figure 5
figure 5

Relationship between body weight and MD for each tissue region investigated: (A) CGM (•) and CWM (○), (B) DGM (•) and DWM (○), and (C) mesencephalon (▪) and cerebellum (□). The regression lines are displayed as solid lines for all regions except the cerebellum, which is displayed as a dashed line.

Full size image
Table 2 Logarithmic regression coefficients and fit quality measures for MD (N = 13) and FA (N = 12) vs body weight
Full size table
Figure 6
figure 6

Log-linear regression of body weight and FA for each tissue region investigated: (A) CGM (•) and CWM (○), (B) DGM (•) and DWM (○), and (C) mesencephalon (▪) and cerebellum (□). The regression lines are displayed as solid lines for all regions except the cerebellum, which is displayed as a dashed line. Triangle symbols represent data excluded from the analysis because of poor FA image quality.

Full size image

Representative baseline CBF and BOLD CVR image for four different ages are provided in Figure 7. Only one ASL data set was acquired for the 5.2 and 6.2 kg pigs (due to imaging-session time constraints), only one BOLD CVR data set was acquired for the 43.8 kg pig (due to secondary breathing circuit technical issues), and no BOLD CVR data were available for the 18.4 kg pig (due to imaging-session time constraints). A detectable BOLD response to CO2 was not present for the smallest pig (2.3 kg) and was excluded from the regression analysis. The mean CGM CBF estimate was 52.6 ± 8.5 mL/100 g/min. No correlations existed between baseline CBF and age for either parametric or nonparametric analyses (Fig. 8, Table 3). The R1tissue measurements exhibited a negative linear correlation with body weight in all regions; however, because the maximum R1tissue difference was small (13 ± 4%), the impact on CBF estimates was minimal (<4%). For BOLD CVR measurements, significant body-weight dependencies were observed for all regions except DWM, and significant logarithmic relationships with body weight were found for all six brain regions (Fig. 9, Table 3).

Figure 7
figure 7

Representative CBF images for 1-wk-old (A), 1-mo-old (B), 2-mo-old (C), and 3-mo-old (D) pigs. Corresponding BOLD CVR images (E–H) are also provided.

Full size image
Figure 8
figure 8

Relationship between body weight and CBF for each tissue region investigated: (A) CGM (•) and CWM (○) and (B) DGM (•) and DWM (○). The solid lines represent the log-linear regression.

Full size image
Table 3 Logarithmic regression coefficients and fit quality measures for baseline CBF (N = 13) and BOLD CVR (N = 11) vs body weight
Full size table
Figure 9
figure 9

Relationship between body weight and the BOLD CVR for each tissue region investigated: (A) CGM (•) and CWM (○), (B) DGM (•) and DWM (○), and (C) mesencephalon (▪) and cerebellum (□). The solid lines represent the log-linear regression for GM regions, and the dashed lines represent the log-linear regression for WM regions.

Full size image

DISCUSSION

We investigated the swine brain as a model for early human brain development using noninvasive MRI methods performed on a clinical MRI system with reasonable scan times. Although previous studies have used invasive methods or ex vivo MRI techniques to characterize structural and microstructural animal brain development with greater detail (19,20), an advantage of our study is the ability to easily translate the methods and observations to clinical human studies. Unique data generated from this study include DTI parameters and ASL measures of CBF and BOLD CVR during swine brain development.

For this study, we used body weight as the dependent variable to assess developmental changes instead of performing between-age-group comparisons. By doing so, we reduced the number of animals needed to characterize developmental changes and avoided the poor statistical power and need to use nonparametric statistical tests for a group-wise analysis. We consider body weight to be an appropriate surrogate for age, as we and others (21) found a smooth and continuous relationship between total brain volume and body weight. Swine brain volume changes in the current study of 1-wk to 3-mo-old pigs qualitatively match the relationship between brain volume and body weight observed in human infancy and prepubertal adolescence (22).

Consistent changes have been observed in the course of the human brain development with respect to increased FA and decreased MD with age (5–10). In the current study, we observed an increase in FA with body weight that followed a log-linear relationship. The log-linear model is often used to describe FA changes with age in the developing human brain (9,10), although models with greater complexity have been used (e.g. biexponential model) (8). Correspondence between swine and human FA changes during maturation is supported by previous invasive studies that suggest a similar pattern of myelination and cellular development exists in both species (21,23). That FA changes are likely attributed to myelination is based on a previous study showing that myelin sheath area was most strongly correlated with FA during development (24). However, other processes of neural development may also affect properties of water diffusion. Despite the similar pattern of FA development between our study and the developing human brain, WM FA values observed in our study (FA ≈ 0.2–0.4) were much lower than those reported in humans (<2 y old, FA ≈ 0.4–0.6) (5). This may be related to between-species differences in WM microstructure, as well as experimental factors such as increased partial volume effects in the smaller pig brain.

The small and inconsistent MD changes observed during swine brain development were unexpected as studies of human brain development observed considerable MD decreases with age (6). This discrepancy between MD trends observed in pigs versus that in humans is most likely due to between-species differences in the timing of brain water content changes. First, MD values in the current study were lower than values reported in the immature human brain (8) and MD in newborn pigs (25) is similar to the adult human brain. Second, water content of the human brain increases rapidly after birth (22), whereas this period of rapid water content increase occurs primarily during the prenatal period of pig brain development (26). Together, these observations suggest MD changes during swine brain development occur prenatally.

A key component of brain development is the concomitant increase in CBF during periods of rapid brain development. In humans, CBF peaks at approximately the age of 2 y, with a second smaller peak around puberty (11,27). The early phase of rapid CBF increase is mirrored by changes in glucose utilization (28) and increases in the total cell count (22). In contrast, our study failed to observe any significant changes in ASL-based measures of CBF in the 1-wk to 3-mo-old pig brain. This observation is consistent with a previous microsphere study that compared CBF in young (1–2 wk old) and older (6–8 mo old) pigs anesthetized with pentobarbitol (29). However, a more recent study measuring CBF via microsphere infusion demonstrated increased CBF in juvenile pigs (4–10 wk old) pigs compared with younger pigs (1–2 wk old) anesthetized with fentanyl (30). It is possible that different anesthetic combinations or CBF measurement methods could account for these inconsistent results. In the current study, use of ketamine, which tends to decrease CBF (31), and the duration of anesthesia may have obscured any CBF changes related to brain development.

One aspect of brain development that has received limited attention is the CBF reserve. Despite not observing any increases in baseline CBF, BOLD CVR increased significantly during maturation, which mirrored changes observed in brain volume and FA. These developmental changes in BOLD CVR are in contrast to a previous microsphere study that failed to observe differences in the CBF response to CO2 between young (1–2 wk) and adult (6–10 mo) pigs (32). Our results also differ from another microsphere study that suggested that CO2 reactivity is depressed for the first week of life, but quickly reaches adult levels between 1 wk old and 1 mo old (33). A major distinction between the current study and the cited studies is the CBF measurement approach. The BOLD signal response to CO2 used in the current study is an indirect measure of CBF based on deoxyhemoglobin concentration changes and may also be influenced by metabolism, cerebral blood volume, and hematocrit levels. Therefore, BOLD CVR changes we observed may be more representative of overall hemodynamic and metabolic development, including factors such as the cerebral metabolic rate of O2. Evidence for the complex interplay between the hemodynamic and metabolic changes during human brain development is provided by the absence or negative BOLD response to visual stimulation observed in neonates (34). Results of our current study have implications for the interpretation functional MRI studies different stages of development (35).

In summary, we have further characterized the complex development processes of swine brain development using noninvasive MRI techniques that can be easily translated to the developing human brain. We found that increases in brain tissue volume and FA mirror changes observed during human brain development; however, unlike the human brain, we failed to observe substantial MD decreases. Despite the lack of maturation-related baseline CBF changes in the current study, we observed significant changes in the BOLD response to CO2 that point to rapid changes in the metabolic and hemodynamic status during early swine brain development. The lack of MD and CBF changes in the swine brain may limit its suitability as a model for some aspects of neurodevelopment. However, for certain applications, such as morphological and DTI measures of anisotropy, future investigations of brain injury and associated neurodevelopmental sequelae in the swine brain may provide inferences into neurological disorders affecting the immature human brain.