Main

Humans rely on vision more than on any other of our senses. Vision is used to form our first social bonds in the immediate months after birth (1). At this age, visually guided behavior is based on processing occurring in subcortical structures such as the thalamus and the superior colliculus. Visual cortex does not make a functional contribution until the second month after birth (2,3). This observation implies that visual perception is served by two systems, a subcortical and a cortical system. The former is genetically determined whereas the latter is guided by experience. Visual pursuit or fixating appear to be fully developed at birth (4), whereas visual acuity and binocular vision improve within the first month of life independent of experience (5). This suggests that, even within the cortical, visual system, some mechanisms may be under genetic control whereas others are influenced by experience. This is reflected in the fact that the parvocellular system subserving color vision develops much more slowly than the magnocellular system used for perceiving movement (6,7).

At the time of its discovery, striate cortex was thought to be all that was necessary for visual perception (8). The discovery of the extrastriate visual areas in the primate visual system increased our appreciation of its complexity. These extrastriate cortical visual areas currently number 30 in the monkey, a number that is expected to be exceeded in humans (9,10). The complex interaction of the numerous brain regions involved in visual processing have led to the suggestion that the development of the visual system occurs in discrete steps with different parts of the system maturing at different times (2,3).

Functional MRI provides a noninvasive means of visualizing neuronal activation underlying perceptual and cognitive processes and so can be used repeatedly and safely on children. Functional MRI has already been used successfully to investigate linguistic development (11–13), motor development (14,15), and visual development in children (16–21). Still, special care has to be taken when it comes to interpreting the results of fMRI measurements on infants and children. The most important issues dealing with the special needs associated with fMRI measurements involving young children are summarized in the review article of Gaillard and colleagues (22).

These authors identified two alternative hypotheses of how the BOLD response of children could differ from that of adults. The immature nervous system could generate a more diffuse activation pattern or inefficient neuronal processing would be associated with an elevated metabolic cost. The first hypothesis suggests a larger extent of activation of the BOLD response in young children, whereas the alternative hypothesis suggests a higher rise in the HbR concentration and a smaller BOLD signal amplitude.

By studying the human visual system, we expect to gain insight into the changes in neuronal processing resulting from cortical maturation. We chose to perform our study on the visual cortex because much is known about its anatomical and functional organization.

The BOLD signal is linked to the vascular response arising from a neuronal challenge. To make any meaningful age-related comparison of the BOLD signal it is important to be certain that the vascular response at different ages is comparable. A number of studies have reported that the hemodynamic response to a neuronal challenge in children younger than 5 y of age and children older than 5 y of age is identical (15,23,24). We tested these alternative hypotheses using a modified version of the flashed and reversing stimulus paradigm used by Fox and Raichle (25). We successfully used this stimulus paradigm in adults to investigate the influence of doubling the OEF on the BOLD signal (26). Increasing the OEF increases the local HbR concentration, which in turn reduces the BOLD signal amplitude (27–31).

We hypothesized that a more diffuse neuronal activation to our stimuli would lead to a larger extent of activation in the younger children. If the neuronal processing in young children were less energy efficient, we would expect the measured BOLD signal amplitude to be lower in young children compared with older children. By comparing the BOLD response of visual cortex to the flashed and the reversing checkerboard stimulus in young and older children, we were in a position to gain an insight into the way maturational changes influence neuronal processing.

MATERIALS AND METHODS

Subjects.

The subjects were 21 children aged between 3 mo and 12 y. We performed functional MR measurements on children in two age groups: children up to 5 y of age (n = 11; males: 3 females: 8) and those older than 5 y of age (n = 10; males: 5, females: 5). This division corresponded to the ratio between CNS to body mass reaching a peak at 5 y of age, and the peak rate of cognitive development (32). All but half the children of the oldest age group were measured under anesthesia. The children measured under anesthesia were measured as part of a diagnostic MR imaging session that had to be performed under anesthesia because clinical indications necessitated it. Because these children usually suffered from or were suspected of a medical condition warranting a diagnostic MR, it was important to set strict inclusion and exclusion criteria. We excluded children with metabolic anomalies, micro-, macro-, or hydrencephaly, children having suffered from hypoxic asphyxia, children with abnormalities of the visual system or children with a tumour located in the cortex lying posterior to the central sulcus. Table 1 shows the different clinical indicators of the children measured under anesthesia. The study was approved by the hospital's ethics committee and informed consent from the child and at least one parent or guardian was obtained.

Table 1 Clinical indicators of children measured under anesthesia
Full size table

Anesthetic management.

All anesthetic procedures were carried out by the institute's anesthetic team. The institutional guidelines relating to the performing of MR imaging under anesthesia were followed at all times. Each patient was administered midazolam (Dormicum; 0.5 mg/kg, maximum dose 10 mg) orally. This is a short-acting sedative and is used routinely to prepare patients for anesthesia. Tracheal intubation followed. The muscle relaxant pancuronium bromide (Pavulon, 0.1 mg/kg) was administered. Breathing and anesthesia was thereafter controlled by a mechanical ventilator. Throughout the entire MR session, all vital signs, including rectal body temperature, respiratory rate, heart rate, blood pressure, blood oxygen saturation, and expiratory CO2, were constantly monitored by the anesthetic team. Expiratory CO2 levels were kept between 4.8% and 6.0%.

The depth of anesthesia is measured using the MAC. One MAC is defined as the concentration of a specific anesthetic agent at which half of the patients still respond to a noxious stimulus (e.g. surgical incision) (33). We performed our fMRI measurements with an anesthetic level of 0.5 MAC using the anesthetic agent Sevoflurane in an oxygen-air mixture. This level of anesthesia is considered light (33–35). In those children measured under anesthesia, the eyelids were kept open using titanium eyelid retractors. Before applying the lid retractors, a short-term, local analgesic agent (Alcaine, Alcon Pharmaceuticals, Mumbai, India) was administered to the eye. This agent has no influence on the pupil or the lens. The conjunctiva and cornea were protected using a clear gel (Viscotears, Novartis Ophthalmics AG, Basel, Switzerland). We ensured that the gel sealed the eye and that it set forming an even surface so as to minimize optical distortion of our image.

We waited until the end-tidal, expiratory concentration of sevoflurane corresponded to 0.5 MAC before commencing our functional MRI measurements. In the time required for the anesthesia level to stabilize at 0.5 MAC, a single diagnostic MRI measurement was performed so as not to unduly delay the examination of the child. After the anesthetist signaled that the level of anesthesia was stable at 0.5 MAC, the functional measurements commenced. The functional MR measurements did not exceed 15 min. At the end of the functional MR measurements, the eyelid retractors were removed and the protective gel washed out. The eyes were then taped shut, as is customary for the institute when diagnostic MR imaging is being performed under anesthesia. The children that were tested awake were instructed to keep their eyes in the middle of the radial checkerboard and avoid large eye movements.

Apparatus.

Measurements were performed on a 2 Tesla Bruker Tomikon S200 whole-body scanner (Bruker, Medical, Faellanden, Switzerland) using a modified surface coil operated in cross-coil mode. This coil was modified by affixing a surface-coated mirror to it in such a way that an image projected onto a translucent screen at the back of the magnet could be reflected on to the subject's eyes. The coil was positioned so that it was most sensitive to signals from the occipital lobe. Inspecting the images in Figure 1 reveals inhomogeneities in the anterior part of the brain. These are the result of the signal loss from the surface coil. This was compensated for by high sensitivity in the region of the visual cortex. Functional imaging was performed using a gradient echo and an echo planar imaging sequence with the following parameters: TR = 2000 ms, TEeffective = 58 ms, spectral width = 100 kHz, and a tip angle = 90°. We used a fat suppression pulse with a 350 Hz bandwidth. A package of 14 slices with a thickness of 4.5 mm and a gap of 1 mm were placed horizontally through the calcarine sulcus, so that the lowest slice coincided with the base of the occipital lobe. The field of view measured 25 × 25 cm with a matrix size of 128 × 64 pixels, resulting in a spatial resolution of 2.0 × 4.1 mm. The echo planar images were then reconstructed to a matrix size of 128 × 128 elements, including zero filling.

Figure 1
figure 1

Luminance contrast version (A, B) and chromatic contrast version (C, D) of the flashed and reversing checkerboard stimulus pair. The two images used to generate the stimuli are shown in the top and bottom of each figure. The flashed checkerboard was alternated with an image of the background (A, C) and the reversing checkerboard was alternated with a checkerboard where the luminance contrast of all elements was inverted (B, D).

Full size image

Stimuli.

Our stimuli were a flashed and a reversing, radial checkerboard. Each checkerboard subtended a visual angle of 36° × 36° at a viewing distance of 1.13 m. The white and black elements of a checkerboard were equal in area. The reversing checkerboard was generated by reversing the contrast of the white elements only. The reversing checkerboard was generated by reversing the contrast of all elements (Fig. 1). We used a high luminance contrast (black and white) and a high chromatic contrast (red and blue) checkerboard stimuli. The luminance of the stimulus elements was measured using a luminance meter (Minolta LS 110; Konica Minolta, Tokyo, Japan). The luminance of the white elements was 144 cd/m2, that of the black background was 0.4 cd/m2. The luminance contrast ratio was 365:1. The luminance of the blue elements was 5.4 cd/m2 whereas that of the red background was 29.3 cd/m2. The luminance contrast ratio was 5.5:1.

We reversed the contrast twice per second. This is slower than the frequency used in other studies. In adults, a temporal frequency of 8 Hz leads to the largest CBF response (25,36–38), whereas the maximum increase in CMRO2 has been observed at 4 Hz (25,36–39). Our reason for choosing this frequency is that the lower temporal and spatial frequencies are perceived better by the immature visual system (5,40–44).

Procedure.

Each measurement was preceded by 10 dummy scans, which were immediately discarded and served only to bring the spin system into a steady state. The actual measurements started with 10 control scans (20 s), during which the subject viewed a static checkerboard. Ten stimulation scans (20 s) followed, during which either the flashed or the reversing checkerboard was presented. This cycle of control and stimulation period was repeated five times. The order in which the stimuli were presented was randomized between subjects.

Data analysis.

We used the commercial package MEDx 3.4 (Sensor Systems, Sterling, VA) to analyze our fMRI data. We followed the method outlined by Boynton and colleagues (45,46) and divided our analysis into two independent steps. First, we located the activated picture elements or voxels. A motion correction using the automated image registration (47) was applied. This was followed by an intensity normalization, which adjusted the mean of the data to a value of 1000. Subsequently, a multilinear correction for baseline drift was applied. Activated voxels with either a positive or negative BOLD response were determined using a cross-correlation analysis. The activated voxels detected were then subjected to a Bonferroni correction to correct for multiple testing. The correction threshold was obtained by convolving the uncorrected threshold (p = 0.05) with the number of voxels in a mask restricting the correction to the subject's head. The number of activated voxels remaining after this correction was taken as the extent of activation. We then used the x, y, and z coordinates of each activated voxels to extract the time series.

We determined the change in the CMRO2 by examining the change in the HbR concentration using ΔR2* as an indicator (30). The ΔR2* was calculated by comparing the transverse magnetization signal at each time point against that of the baseline, using the formula

Formula

We defined our baseline signal (Sb) as the average of the 10 control scans of the first cycle. Where TE is the echo time, St the MR signal at a specific time point, and Sb the baseline MR signal (28). A ΔR2* of 0 indicates no change in the HbR concentration compared with the baseline. A positive ΔR2* is associated with a negative BOLD response and indicates an increase in the concentration of HbR compared with the baseline. A negative ΔR2* is associated with a positive BOLD response and indicates a decrease in the concentration of HbR compared with the baseline. ΔR2* is proportional to the change in HbR concentration from the baseline.

RESULTS

We pooled the luminance and chromatic contrast data as the BOLD responses to the luminance and chromatic contrast stimuli which have been reported to be the same (48). Figure 2A shows the location of the activated voxels overlayed onto an anatomical image of a child from the younger age group. Figure 2B shows the activated voxels in a child from the older age group. In both cases, activated voxels were located in the striate cortex.

Figure 2
figure 2

(A) Location of the activated voxels to the flashed, chromatic contrast checkerboard in one of the children in the younger age group. (B) Location of the activated voxels to the same stimulus in one of the five children of the older age group measured under anesthesia. In both cases, the activated voxels area clearly located in visual cortex.

Full size image

All statistical analyses were performed using SPSS Version 11.5 (SPSS Inc., Chicago, IL). The data were analyzed using repeated measures of the general linear model with age as the between-subject factor and stimulus type as the repeated measures factor. As is standard practice we will only provide details of the statistical results that attained significance (p ≤ 0.05).

In a separate study examining the influence of light Sevoflurane anesthesia on the BOLD response, we found that the extent of activation was attenuated under anesthesia but that the CMRO2 and the CBF response were unaltered. We therefore excluded the five children measured awake from the analysis of the extent of activation because their inclusion would have introduced a confounding effect between the age group and the use of an anesthetic agent. These five children were included in the analysis of ΔR2*, as our previous study had shown ΔR2* to be unaffected by the anesthetic agent.

Extent of activation of the positive BOLD response.

Figure 3A shows the mean number of activated voxels with a positive BOLD response. There was no difference between the two age groups.

Figure 3
figure 3

(A) Mean number of activated voxels with a positive BOLD response to the flashed and reversing checkerboard stimuli in the two age groups (younger, n = 11; older, n = 5. (B) Data for the voxels with a negative BOLD response. The bars indicate the standard error of the mean.

Full size image

Extent of activation of the negative BOLD response.

Figure 3B shows the mean number of activated voxels with a negative BOLD response for the flashed and the reversing stimuli. There was no difference between the age groups.

CMRO2 and the positive BOLD response.

With our flashed and reversing checkerboards, we were in a position to use ΔR2* as an indicator of the change in the neuronal oxidative metabolism. Figure 4, A and B, shows ΔR2* at each time point in a “time-locked” graph. Figure 4A represents the data obtained from the younger children, and Figure 4B shows the data from the older children. Figure 6A shows the peak ΔR2* values for the two stimuli in the two age groups. The error bars in the figures represent the SEM. We found a statistically significant difference in the peak ΔR2* between the flashed and the reversing checkerboard (FWilks = 28.609; df = 1,18; p = 0.000; ETA2 = 0.614). We also found a significant interaction with the age (FWilks = 4.705; df = 1,18; p = 0.045; ETA2 = 0.207).

Figure 4
figure 4

Peak ΔR2* values obtained from the voxels with a positive BOLD signal. (A) Peak ΔR2* for each “time-locked” time point to the flashed and reversing checkerboard stimulus from the children in the younger age group (n = 11). (B) The same data for the children in the older age group (n = 5). The data from the flashed checkerboard is shown by the black curve, the data for the reversing checkerboard by the gray curve. The closer the ΔR2* values are to 0, the smaller the change in HbR concentration between that time point and the baseline. The bars indicate the standard error of the mean.

Full size image
Figure 6
figure 5

Peak ΔR2* for the flashed and the reversing checkerboard stimuli. (A) Peak ΔR2* from the pixels with a positive BOLD signal. The data from the younger children (n = 11) are shown in black, the data from the older children (n = 5) in gray. The closer the ΔR2* values are to 0, the smaller the change in HbR concentration compared with the baseline. (B) Data from the pixels with a negative BOLD signal. In A and B, the data from the younger children are shown in black, the data from the older children in gray. The closer the ΔR2* values are to 0, the smaller the change in HbR concentration compared with the baseline. The bars indicate the standard error of the mean.

Full size image

CMRO2 and the negative BOLD response.

Figure 5, A and B, shows “time-locked” graphs of the ΔR2* at each time point. Figure 5A represents the data obtained from children under 5 y of age, whereas Figure 5B shows the data from children older than 5 y. Figure 6B shows the peak ΔR2* values for the two stimuli in the two age groups. The error bars in the figures represent the SEM. We found no statistically significant difference in the peak ΔR2* from the voxels with a negative BOLD response.

Figure 5
figure 6

ΔR2* values obtained from the voxels with a negative BOLD signal. (A) Peak ΔR2* for each “time-locked” time point to the flashed and reversing checkerboard stimulus from the children in the younger age group. (B) The same data for the children in the older age group. The data from the flashed checkerboard is shown by the black curve, the data for the reversing checkerboard by the gray curve. The closer the ΔR2* values are to 0, the smaller the change in HbR concentration between that time point and the baseline. The bars indicate the SEM.

Full size image

DISCUSSION

Our study has shown that the BOLD signal response in young children is markedly different from that seen in older children. In older children, as in adults, the BOLD signal amplitude to the flashed checkerboard resulted in a larger BOLD signal amplitude than the reversing checkerboard. In the younger children, the BOLD signal amplitude to both stimuli corresponded to the amplitude observed with the reversing checkerboard stimulus in the older children. In this section we will examine the possible reasons for this difference in the BOLD response between the two age groups. This will include an examination of the origin of the BOLD signal, a discussion on the relationship between neuronal activity and CMRO2, and examination of the relationship between the BOLD signal and the OEF.

The origin of the BOLD signal.

The BOLD signal is used by virtually all functional MRI studies and is the result of a local change in the HbR concentration. An increase in neuronal activity leads to an increase in the local CMRO2. This in turn leads to more of the oxyhemoglobin (HbO) being turned into HbR. The diamagnetic property of HbO renders it invisible to MR imaging. The paramagnetic property of HbR on the other hand decreases the MR signal. A local increase in neuronal activity is generally believed to be followed by a local increase in blood flow (CBF). This vascular response dilutes the local HbR concentration to such an extent that the MR signal increases during neuronal activation compared with rest. This is what is commonly referred to as the BOLD response.

Defining neuronal activation.

Traditionally, neuronal activity has been synonymous with electrical discharge activity. Functional MRI using the BOLD response is based on detecting changes in the HbR concentration in a brain volume as small as 1 mm3. A volume of this size contains millions of cells. The CMRO2 and consequently the HbR concentration within this small volume will depend not only on the electrical discharge activity, but also, and probably more so, on the number of activated cells. In our investigation we manipulated the size of the activated neuronal population while keeping their electrical activity constant.

The effect of our stimuli on CBF.

We did not actually measure the CBF response using blood flow– or perfusion-sensitive MR sequences such as arterial spin labeling. A recent study has reported that the measurement of the CBF response using arterial spin labeling is not as reliable as that obtained by PET (49). Measuring the CBF response to our stimuli would have more than doubled the time the child had to be kept under anesthesia or the child would have to be subjected to a PET measurement. We relied instead on the published PET data of Fox and Raichle (25), demonstrating that the CBF response to a flashed and a reversing stimulus is identical. The finding of our study of a larger ΔR2* for the flashed when compared to the reversing checkerboard is consistent with the prediction by Buxton (50) for stimuli with identical CBF responses but differing CMRO2. These observations, taken together, confirm that our checkerboard stimuli did induce identical CBF responses.

Effect of our stimuli on neuronal oxidative metabolism (CMRO2).

For reasons discussed by Marcar and colleagues (26), the number of activated neurons and the mean electrical activity resulting from the presentation of the reversing checkerboard was double that resulting from the presentation of the flashed checkerboard. Doubling the size of the activated neuronal population will also double the CMRO2. Both checkerboard stimuli induced identical vascular responses so that doubling the CMRO2 implies that more oxygen has to be extracted from the blood, thus increasing the OEF. Increasing the OEF will result in an elevation of the local HbR concentration.

Influence of anesthesia on the BOLD response.

The effect of performing fMRI measurements on children during sleep, be it natural (16,17) or induced (18,51), is well documented. In studies where sleep was induced, a sedative agent was used. The administration of a sedative agent has been found to reduce glucose metabolism (CMRGlc) by 25–30% (52) and results in a reduction of the BOLD response (51). What effect measuring children under light anesthesia has on the BOLD signal is unknown. The influence of light anesthesia on CMRO2 and its implication for BOLD measurements are reported in our companion paper (53). In summary, we found that sevoflurane anaesthesia attenuated the extent of activation, whereas its effect on CBF and CMRO2 were found to be minimal (53).

Effect of cortical maturation on CBF.

As mentioned above, our study did not include a measurement of the CBF response. Instead, we drew on PET measurements in adults to assess the CBF response to our flashed and reversing checkerboards. The hemodynamic response to a neuronal challenge has also been found to be the same in children younger than 5 y of age compared with children older than 5 y of age (15,23,24). These studies clearly indicate that the CBF response does not vary between younger and older children. Maturational changes in CBF show a similar time course as the maturational changes in CMRGlc (54,55). A tight coupling between the baseline CMRGlc and CBF exists throughout cerebral development (56) and adulthood (57). These studies indicate that the observed difference in the BOLD response between the two age groups is not the result of a change in the vascular response with age. Having excluded a maturational effect on the vascular response as a cause of our observed result, the most likely reason for the observed difference in the BOLD signal amplitude is a change in the oxidative metabolism (CMRO2) with age.

Effect of cortical maturation on CMRO2.

A change in the oxidative metabolism is likely to result from a change in the number of cells being activated or in the number of synapses. Both neuronal density and synaptic density are subject to a significant change with age. The ratio of gray to white matter is larger in young children than adults (58). This implies that the brains of young children will have a higher metabolic requirement simply because their brains contain more neurons. Anatomical studies indicate that the neuronal density reaches its peak at the time of birth. It is unlikely that a maturational change in CMRO2 is the result of a change in neuronal density. The more likely cause is a change in synaptic density with age. An assessment of the energy expenditure of neurons revealed that the majority of the energy is expended at the synapses (59–61). Synaptic density has been found to peak between 4 and 7 y of age (62,63). A PET study has shown that CMRGluc increases up to 5 y and then declines (56). These are indications that the neuronal metabolic activity that is higher in young children is perhaps due to a higher synaptic density.

Effect of CMRO2 on the BOLD signal.

We have already presented our arguments relating to the influence of our stimuli on the BOLD signal. In this section, we examine the influence of a change in neuronal density compared with a change in synaptic density on the BOLD signal. A higher neuronal density elevates the baseline metabolic activity and thereby reduces the dynamic range of the BOLD signal. A higher neuronal density, however, would not alter the relative difference in CMRO2 between our flashed and the reversing checkerboard stimuli. The BOLD signal in all age groups should therefore reflect this relative difference in CMRO2. The maturational change in synaptic density is not the result of a change in the number of neurons but the result of a change in the number of synaptic connections between them. Changing synaptic density has little effect on the baseline CMRO2 but has a strong effect on the CMRO2 associated with neuronal activity. An immature visual system containing many superfluous neuronal connections will be less energy efficient than a mature visual system. This reduced energy efficiency will manifest itself more strongly during neuronal activation. A change in synaptic density will not alter the baseline CMRO2 and so has little influence on the dynamic range of the BOLD signal. The presence of superfluous neuronal connections raises CMRO2 during neuronal activity because more neurons than actually required will be activated by a given stimulus, making the system less energy efficient. A high CMRO2 level during neuronal activity implies a high OEF, a high HbR concentration, and an attenuation of the BOLD signal amplitude.

CONCLUSIONS

Our study set out to investigate the hypotheses put forward by Gaillard and colleagues (64), by examining whether the immature nervous system leads to a more diffuse BOLD response or to a lower BOLD amplitude and less activation. The results of our study indicate that the BOLD response in young children is most strongly influenced by CMRO2. As a result of the reduced energy efficiency of the immature nervous system, the local OEF and the HbR concentration are higher. The BOLD signal amplitude in response to neuronal activation will tend to be smaller in young children compared with older children or adults. A lower BOLD signal amplitude makes it more difficult to detect neuronal activation using statistical methods. Only when the superfluous connections and members of a network of neurons are pruned, does the energy efficiency improve and the metabolic cost of neuronal processes decline. A decline in CMRO2 reduces the OEF and the local HbR concentration. This in turn increases the BOLD signal amplitude which makes it easier to detect neuronal activation.