The low-carbohydrate, high-fat ketogenic diet (KD) has been shown to be effective in managing a variety of diseases, including epilepsy, obesity, and disorders of the central nervous system [1–5]. It is also used for body-building purposes, and as a general lifestyle diet among healthy adults. This broad and growing use by adults, and particularly by women of child-bearing age, raises a concern about its use during gestation and its potential effects on offspring growth and brain development.

Studies on adult humans and rodents have shown that sparse carbohydrate and glucose availability initiates a process of ketogenesis that produces ketones, which all organs – including the brain – can use for energy [6, 7]. During gestation, since the embryo’s capability to carry out ketogenesis is very limited, it relies on ketone supply from maternal circulation. Ketone supply is modulated by the carriers Monocarboxylate Transporters (MCTs) [8, 9]. Studies using non-ketogenic gestational diets report that MCT expression in the placenta decreases at the end of gestation [10], implying restricted ketone availability to the fetus. This observation seems contrary to other studies that reported an increase in ketone utilization enzymes within the brain during the second half of gestation [6]. This increase in ketone utilization enzymes was speculated to facilitate lipid and myelin (white-matter) synthesis [11], implying greater availability would be favourable for the developing brain. A similar conclusion could be drawn from studies linking ketones and cerebral protein synthesis [11, 12]. Yet, while ketones may be important metabolic substrates, an excessive supply to fetal brain can also damage nucleotide bio-synthesis, impairing nucleic acid production and brain growth [13]. Hence, it is important to determine whether a gestational KD indeed alters early brain development.

Investigations of nutritional effects on growth and brain development are often conducted on the mouse due to the high genetic homology it shares with humans. However, unlike the human brain, the mouse brain continues in its rapid growth during the first few weeks post-natally, until weaning, at Post-natal day (P) 21.5 [14, 15]. Therefore, investigation of nutritional effects on early anatomical brain development using the mouse model must consider the first three weeks postnatally.

In our previous study we started investigating the physiological consequences of a gestational KD on the offspring [16]. That study focused on the embryonic period and revealed embryonic growth retardation as well as alterations in brain structure just prior to parturition. The current study focuses on the next developmental period - the neonatal period - and investigates the implication of a gestational KD for the neonatal mouse. The neonatal mouse brain previously exposed to a maternal gestational Standard Diet (SD) is compared to that exposed to a maternal gestational KD. This investigation is carried out using 3D magnetic resonance imaging (MRI). Conventionally, T1 and T2 weighted sequences are employed for such investigation. However, these sequences provide limited grey to white matter contrast in the neonatal brain, due to the sparse myelin at that period. As a result, image alignment and registration accuracy is also compromised, leading to poor structural assessment. Instead, and to improve the amount of anatomical information gathered from neonatal brain imaging, we utilized a Diffusion Tensor Imaging (DTI) sequence, which provides the regional grey/white matter contrast, as previously employed by Zhang et al. (2006) [17].

Six-week-old CD-1 male and female mice were weight-matched and randomly assigned to either a control group on a SD, or a study group on a KD. Mice were fed their respective diets for a period of 30 days, and then naturally mated by setting up a single male with a single female from their respective group. The morning newly-born pups were observed was considered day 0.5 of post-natal life (P0.5). Post parturition, pups were perfused and their skull and brain were extracted at one of two time-points: P11.5 and P21.5, corresponding to the end of the rapid brain growth and soon after the onset of the period of rapid myelination, respectively [14, 18].

At the beginning of the study we discovered that the KD dams were unable to lactate properly, as described in the results. The protocol was altered to ensure the viability of the KD dams and pups until weaning. Instead of remaining with their mothers, the KD pups were separated from their biological mothers at P2.5. Arbitrarily selected ones were then fostered by lactating SD foster dams whose pups were at the same age as the KDs. The control group was also fostered in a similar manner. The SD and KD pups who were not fostered, were euthanized.

Brain imaging

Fixed brains were imaged by MRI with a 3D Diffusion Tensor Imaging (DTI) sequence consisting of a fast spin echo train length of 6 and a TR of 325 ms. TE for the first echo was 30ms, and then the echo spacing was set to 6 ms for the remaining 5 echoes. The diffusion parameters were as follows: γ=2.675∗10⁸ r a d/s▪T, δ=4 m s, and Δ=15 m s. The imaging field-of-view was 14 x 14 x 25 mm³, the matrix size was 108 x 108 x 192, and the image was acquired using 2 averages. This yielded a final image with 0.130 mm isotropic voxels. Five images with minimal diffusion weighting (b = 0 s/mm²) and 30 high b-value images (b = 1918 s/mm², G=35 G/c m) were acquired according to the Jones30 scheme [23–25]. Total imaging time for this sequence was ∼14 hrs.

An example of a single MR image of a mouse brain acquired with DTI is depicted in Figure 1, along with labelling of several regions. Figure 1 shows the b0 image (Right) along with the average image of the 30 high-b diffusion weighted images (Left), which was used for image analysis. As can be seen, the image obtained from averaging the high-b diffusion weighted images yields a significantly improved anatomical contrast.

Image statistical analysis

At each of the two time-points, an average of the 30 high b-value diffusion weighted images of the SD and KD pups were geometrically aligned with one another, and saved within a common coordinate system. The overlapping images were then used to construct a single consensus average image by first undergoing linear (6 parameter, followed by a 12 parameter) alignment through a series of rotations, translations, scales, and shears. Then, an iterative non-linear alignment procedure created local deformations in each image, thereby bringing all images into alignment. All registrations were performed using a combination of ANTS algorithm [26] and mni_autoreg tool [27]. The collection of all linear and non-linear deformations across each image formed a deformation field, which contained the transformation from each embryo image to the consensus average image. The Jacobian matrix was calculated for every point in the deformation field, and the determinant of the Jacobian was then calculated for each image voxel (i.e. 3D pixel). The log-transform of these determinants were computed, assigning positive values to the determinants that were >1 and negative values to those <1. Smoothing was applied to the resulting values using a 0.2mm FWHM Gaussian kernel, to improve normal distribution and facilitate the following t-statistics. Analysis of these results using t-statistics allowed to identify whether the KD group significantly differed from the SD group of brains in any particular 3D region. The results from this deformation-based analysis are reported as a structural brain image overlaid with the voxel’s t-statistics, as shown in Figure 2 by the red and blue scale-bars in regions that were significantly different. This analysis accounted for multiple comparisons through a False Discovery Rate (FDR) technique [28, 29], as reported along with each image. To quantify whole-structure volume changes, brain regions were segmented from the final average image using a pre-existing atlas [30]. The inverse deformation field from the registration was then applied to back propagate the atlas onto the original unregistered images, allowing the regional volume differences to be calculated. Only regions that were significantly different at an FDR of <15% in this region-based analysis are reported in the results. FDR was calculated based on all 62 regions in the atlas.

Our study first reveals that a gestational ketogenic diet decreases maternal fertility and litter size. This may be attributed to the lower dietary protein contents in the KD versus the SD. A study by Goettsch (1960) [31] indicated a direct relationship between protein consumption and maternal fertility, as measured by estrus cycle length. The same study also found that reduced dietary protein could lead to smaller litter size. A similar observation was made in a human study using a low carbohydrate, high protein diet, which revealed increased fertility with increasing protein intake [32]. However, other researchers have not observed a reduction in litter size as a result of a protein deficiency, per se [33]. These researchers speculated that during lactation the “females could compensate for a dietary protein deficit by reducing their reproductive effort to match the lower availability of protein” [32]. In our study, the smaller litter size was noted only at P2.5. Hence, it is possible that the litter size upon parturition was not statistically different between the study groups, and that the difference noted at P2.5 was due to maternal cannibalization of the KD pups. Such maternal behaviour could be caused by the high metabolic stress the KD dams were experiencing, and would indirectly reduce their reproductive efforts [34].

Our study also shows that KD dams are at a high risk for developing ketoacidosis during lactation. The elevated maternal serum ketone along with the statistically lower serum glucose concentration, as measured at P2.5, were the early signs of ketoacidosis and an indication of physiological strain due to lactation. Since lactation significantly raises the metabolic rate, to more than twice the normal level [35], this high metabolic demand may be the reason for the development of ketoacidosis in the KD dams. Cessation of milk production and mammary involution, as expected to be induced by separation from the litter at P2.5 [36, 37], facilitated stabilization of the KD dams’ metabolism within 4 days. Their elevated ketone concentration at P2.5 dropped to below ∼2mmol/L, and their below-normal P2.5 glucose concentration increased to ∼6mmol/L, both of which were within the normal stable-ketosis values [38].

Change in maternal body weight over the first three weeks post parturition exhibited a different trend depending on whether the dams were breastfeeding. The weight of the breastfeeding foster SD dams increased gradually between P2.5 and P10.5, but then decreased until P20.5, when the offspring were weaned. On the other hand, the weight of the dams who were separated from their litters at P2.5 only decreased thereafter, and stabilized by around P14.5. In this case, there was no intermediate increase in body weight. In the case of both the SD-control and KD dams, their separation from the litter soon after parturition lead to milk stasis, cessation of milk production, along with mammary involution [36, 39]. In fact, their pattern and timing of weight loss is consistent with those of mammary involution and mammary cell apoptosis, as reported by Richert et al. [40]. The difference in the rate of weight loss of the SD-control and KD dams may be attributed to differences in mammary tissue volume. The fact that the KD dams encountered metabolic difficulties lactating, also supports our speculation that their mammary glands were under-developed. Under-developed mammary glands have smaller milk storage capacity, leading to decreased production of milk [40, 41]. The insufficient milk production is, in turn, supported by our observation of the retarded growth of the KD pups who originally remained with their biological KD mothers during lactation.

Offspring growth, as measured by the average weight gain of the pups, was also different in the adopted KD compared with the adopted SD groups. The KD pups exhibited a slower weight gain, which became particularly noticeable after P12.5, and resulted in a statistically significant difference in weight at P20.5 (p <0.05). This difference in weight gain could be attributed to the dietary mismatch the KD pups experienced between pre- and post-adoption periods. Prior to adoption, the KD pups received a high-fat diet through maternal KD milk; whereas post-adoption, they received a low-fat diet through SD milk. This change from abundance to sparsity in fat renders the postnatal, post-adoption phase a state of undernutrition. Since postnatal undernutrition has been shown to result in slower postnatal growth [42], this would explain the significantly lower body weight of the KD pups at P20.5. Another potential explanation is that the dietary mismatch rendered the KD pups weaker compared with the biological SD ones, limiting their access to maternal milk, and resulting in hindered growth. Overall, the early dietary mismatch is a confound that may play a role in the slowed growth of the KD pups betweem P12.5 and P21.5.

Offspring brain structure was also altered as a result of prenatal and early-postnatal exposure to a maternal KD. At P11.5, when compared with the SD brain, the KD brain had a slight bilateral enlargement of cortical volume, and unilateral reduction in mid-brain volume, when computed as a percentage of the total brain volume. At the later time-point, P21.5, the relative bilateral enlargement of the cortex became more prominent. In addition, the KD hypothalamus was also relatively enlarged; whereas the hippocampus, corpus callosum and olfactory bulb were smaller in relative volume. These changes could be due to the augmented prenatal nutritional environment of the KD mouse offspring. Since the mouse brain undergoes significant organizational processes in prenatal life, such as neurogenesis, neuronal migration, and differentiation, destabilization of this prenatal period is detrimental for proper brain development [43]. A metabolic insult that alters any of these processes can impact neuronal circuit formation and cerebral architectural organization, which would become apparent only during the brain’s postnatal growth spurt in the mouse [43]. We demonstrate that the prenatal exposure to a maternal KD is the cause of the observed structural alterations, confounded by the KD-to-SD switch at P2.5.

The exact effects of a maternal, gestational KD have not been thoroughly investigated. Thus far, it is known that ketones are natural by-products of fat metabolism [44], and can serve as metabolic fuels in lieu of glucose [12, 45, 46]. Yet, their yield during ATP hydrolysis is greater than that of glucose [4, 5, 47], suggesting ketosis provides a greater amount of energy compared with glycolysis. While the greater energy availability may promote cellular growth, it may be counteracted by ketones’ inhibition of pyrimidine synthesis [45], alteration of cerebral nucleotide bio-synthesis and of cerebral amino acid metabolism [13, 48]. The overall effect of ketones on the developing brain, would then depend on the balance between these various factors. Specifically, a positive net effect could explain the increase in relative volume observed in the hypothalamus and medulla; whereas, a negative net effect could explain the relative decrease in volume observed in the hippocampus and corpus callosum.

Studies on maternal ketosis in the pig revealed an increase in weight, protein content and cell-size in the fetal pig brain [49]. Studies using the mouse showed that its brain preferentially utilizes ketones for energy, and as precursors of amino-acid and lipid synthesis [50, 51]. Similar studies on the rat found elevated levels of the ketone β−hydroxybutyrate, particularly in the striatum and cortex, in late gestation and early postnatal period [52]. Together, these studies indicate that ketones do play a dominant role in prenatal brain growth. They also imply that an increase in ketone availability, as may occur during maternal KD could, in fact, alter the volume of various brain regions in the offspring, supporting our observations. The altered brain structure may be indicative of functional and behavioural changes in the KD offspring. The exact outcome remains to be elucidated by post-natal behavioural studies of the offspring. The genetic similarity between the mouse and the human suggests that our observations may resemble the potential impact on the human fetal brain, if exposed to a similar gestational ketogenic diet at the corresponding developmental time period.

To reveal the potential impact of a gestational Ketogenic diet on brain structure in the offspring, we imaged neonatal brains from mice whose mothers were either fed a Standard Diet (SD) or a Ketogenic Diet (KD) during gestation. Our initial observations of the KD dams indicated reduction in maternal fertility and litter size, as well as a substantially increased risk of developing fatal ketoacidosis during lactation. To ensure viability of both the dams and their pups, we carried out fostering at P2.5. Following separation from their litters at P2.5, the KD dams exhibited a more gradual weight decrease, compared with the control SD dams. The average fostered KD pup exhibited retarded weight gain during the first 3 weeks postnatally, compared with the average fostered SD pup. Finally, assessment of brain structure revealed several major structural differences at both P11.5 and P21.5. The KD brain had a bilateral relative decrease in the cortex, hippocampus, corpus callosum, fimbria, and lateral ventricles, but a relative enlargement in volume in the hypothalamus and medulla. These anatomical differences may be the consequence of the altered nutritional environment during prenatal life, when significant organizational processes occur in the mouse brain. These changes could also be triggered by certain brain regions’ preference for ketones during development, as well as by the increased efficiency of energy production from ketones. In conclusion, these cerebral anatomical alterations could be indicative of underlying functional and behavioural changes which could become apparent in later postnatal life. Overall, our study points out the significance of prenatal nutrition, and indicates that it drives subsequent postnatal brain development, even when matching postnatal nutrition exists.