Comparative utility of MRI and EEG for early detection of cortical dysmaturation after postnatal systemic inflammation in the neonatal rat

Background: Exposure to postnatal systemic inflammation is associated with increased risk of brain injury in preterm infants, leading to impaired maturation of the cerebral cortex and adverse neurodevelopmental outcomes. However, the optimal method for identifying cortical dysmaturation is unclear. Herein, we compared the utility of electroencephalography (EEG), diffusion tensor imaging (DTI), and neurite orientation dispersion and density imaging (NODDI) at different recovery times after systemic inflammation in newborn rats. Methods: Sprague Dawley rat pups of both sexes received single-daily lipopolysaccharide (LPS; 0.3 mg/kg i.p.; n = 51) or saline ( n = 55) injections on postnatal days (P)1, 2, and 3. A subset of these animals were implanted with EEG electrodes. Cortical EEG was recorded for 30 min from unanesthetized, unrestrained pups at P7, P14, and P21, and in separate groups, brain tissues were collected at these ages for ex-vivo MRI analysis (9.4 T) and Golgi – Cox staining (to assess neuronal morphology) in the motor cortex. Results: Postnatal inflammation was associated with reduced cortical pyramidal neuron arborization from P7, P14, and P21. These changes were associated with dysmature EEG features (e.g., persistence of delta waveforms, higher EEG amplitude, reduced spectral edge frequency) at P7 and P14, and higher EEG power in the theta and alpha ranges at P21. By contrast, there were no changes in cortical DTI or NODDI in LPS rats at P7 or P14, while there was an increase in cortical fractional anisotropy (FA) and decrease in orientation dispersion index (ODI) at P21. Conclusions: EEG may be useful for identifying the early evolution of impaired cortical development after early life postnatal systemic inflammation, while DTI and NODDI seem to be more suited to assessing established cortical changes.

One of the major patterns of brain injury observed by structural neuroimaging in infants born preterm involves diffuse changes in the white matter regions at term-equivalent age (Kline et al., 2022;Agut et al., 2020).Pathologically, diffuse white matter injury is characterized by the degeneration, regeneration, and maturational arrest of preoligodendrocytes, with associated chronic deficits in axonal myelination and gliosis (Laureano et al., 2024;Buser et al., 2012;Inder et al., 2023).Critically, modern imaging studies also show persisting impairments in the growth (Ma et al., 2022;Adrian et al., 2023;Kelly et al., 2023), microstructural development (Ball et al., 2013;Dimitrova et al., 2021), and connectivity (Toulmin et al., 2021;Jang et al., 2023) of grey matter regions (including the cerebral cortex) in children who were born preterm, including after postnatal systemic infection/inflammation (Jain et al., 2022;Kuban et al., 2019).Compelling experimental evidence suggests that such cortical deficits involve maturational disturbances in the growth, complexity, and connectivity of neuronal dendrites (i.e., neuronal dysmaturation), but without neuronal loss (Prasad et al., 2021;Stolp et al., 2019;Kelly et al., 2023;Dean et al., 2013;Ardalan et al., 2019).At present, the onset and speed of the evolution of neuronal dysmaturation remain unclear, partly because there are limited data on neuromonitoring outcomes that have been validated by neurohistological assessment of cellular pathology.
MRI can directly assess brain development and injury, and has well established utility for predicting neurodevelopmental outcomes, in preterm infants (Jansen et al., 2021;Cayam-Rand et al., 2019;Christensen et al., 2023).For example, even in the absence of focal cerebral lesions, diffusion tensor imaging (DTI) studies at preterm and later ages have shown developmental changes in both white matter and cortical microstructure and connectivity, indicative of impaired brain maturation, in preterm born infants (Ball et al., 2013;Liu et al., 2023;Vinall et al., 2013), the extent of which are related to poorer neurodevelopmental outcomes (Martini et al., 2023).However, DTI cannot identify the specific cellular compartments or cellular changes that contribute to changes in DTI-derived parameters.By contrast, the advanced diffusion imaging technique NODDI (neurite orientation dispersion and density imaging) is proposed to provide more specificity for assessing cellular microstructure by using separate parameters for cellular process density (i.e., the intra-cellular volume fraction [FICV]; also termed neurite density index [NDI]) and cellular process dispersion (i.e., the orientation dispersion index [ODI]) (Zhang et al., 2012;Eaton-Rosen et al., 2015;van de Looij et al., 2015).Maturational changes in NODDI parameters in the white matter and cerebral cortex, and their associations with functional outcomes, have been reported in children and adolescents following very preterm birth (Kline et al., 2022;Siffredi et al., 2023;Wang et al., 2022;Young et al., 2019).Experimentally, we have reported that NODDI (and DTI) parameters can detect long-term deficits in oligodendrocyte maturation and axonal myelination in the white matter, and reduced dendritic complexity of neurons in the cerebral cortex, after postnatal inflammation in newborn rats (Prasad et al., 2021).Nevertheless, neural abnormalities identified with MRI do not completely explain the heterogeneity in long-term outcomes associated with preterm birth (Arulkumaran et al., 2020;Banihani et al., 2021).
Electroencephalography (EEG) is increasingly used to assess functional brain maturation and injury in preterm infants (Stevenson et al., 2020;Pavlidis et al., 2020).During normal brain development to term age, the EEG progressively transitions from a discontinuous to a more continuous pattern of background activity, with appearance of sleepstate cycling, increased hemispheric synchrony, and loss of immature oscillations (with their replacement by adult frequency bands) (Bourel-Ponchel et al., 2021).EEG features associated with brain injury include acute background EEG depression with alterations in EEG amplitude, frequency, and continuity, as well as the development of disorganized (e.g., modified or abnormal waveforms) or dysmature (i.e., presence of immature EEG features that correspond to a normal child ≥2 weeks younger) activity patterns (Stevenson et al., 2020;Scher et al., 2011;Wikström et al., 2012;Tich et al., 2007), which are associated with motor, neurobehavioral, and neurocognitive disturbances (van 't Westende et al., 2022;El Ters et al., 2018;Deshpande et al., 2022).Interestingly, there is some evidence that disorganized EEG patterns may reflect more severe brain injury (e.g., cystic white matter lesions) and can predict motor outcomes (Watanabe et al., 1999;Hayakawa et al., 1997;Pittet-Metrailler et al., 2020), while dysmature patterns may reflect impaired brain maturation (e.g., of the cerebral cortex) and can predict cognitive outcomes (Kong et al., 2018;Okumura et al., 2002;Selton et al., 2013;Hayakawa et al., 1997;Scher, 1997).Nevertheless, the specific EEG features related to impaired cortical neuron development, and their relative timing, after postnatal inflammation, remain unclear.
Early detection of cortical dysmaturation has the potential to provide the opportunity for neurodevelopmental interventions in high-risk preterm infants (Pineda et al., 2021;Siffredi et al., 2010).Therefore, in the present study, we compared the utility of MRI and EEG for detecting and monitoring the development of cortical dysmaturation after neonatal brain injury.Specifically, we tested the hypothesis that DTI, NODDI, and EEG parameters will detect the evolution of cortical neuronal dysmaturation (i.e., alterations in neuronal dendritic outgrowth) validated by neurohistopathology after prolonged postnatal systemic inflammation in newborn rats.

Animals and experimental design
All animal experiments were conducted with approval from the Animal Ethics Committee, University of Auckland (#R2020, #R2109), in accordance with the Guide for the Care and Use of Laboratory Animals (NIH) (Grundy, 2015).We also followed the National Institutes of Health guidelines on Rigor and Reproducibility in Preclinical Research, which included the use of randomization, blinding, both sexes, and statistical/power analyses.Pregnant Sprague-Dawley rats (specific pathogen free) were purchased from the small animal Vernon Jansen Unit (University of Auckland, New Zealand), and gave birth within the small animal facility (before 13:00 h to standardize for age).On the day of birth (defined as postnatal day [P]0), litter size were standardized to 10 pups of both sexes.Animals were housed with a 12 h light/dark cycle (light on 06:00-18:00 h) in a temperature-(~23 • C) and humidity-(~45 %) controlled environment, with food and water available ad libitum.
At P1, pups within each litter were randomly allocated to control (sterile saline; Thermo Fisher Scientific, Auckland, New Zealand) and lipopolysaccharide (LPS; ultrapure from E. coli [055:B5]; LOT #4231A1; Sapphire Bioscience Ltd., Auckland, New Zealand; 0.3 mg/kg) groups.Single daily intraperitoneal injections of saline and LPS (10 µL/g body weight) were given from P1-P3.To minimize the effects of circadian rhythm, experiments were performed at the same time each day (~12:00 h; at least 24 h after birth).Rat pups were euthanized at either P7 (by decapitation), or P14 or P21 (by CO 2 narcosis and cervical dislocation).Body weights were measured daily, while brain weights and sex were assessed at post-mortem (Supplemental Fig. 1).Brain tissues were collected at post-mortem for various analyses (detailed below).All surgery was performed under adequate anesthesia and analgesia, and all efforts were made to minimize suffering or stress, and to reduce the number of animals used.The welfare of the pups (e.g., health and feeding behaviors) and dams (e.g., nest building, pup retrieval/rejection, and feeding behaviors) was monitored regularly.
A timeline of the experimental protocol is shown in Fig. 1.A total of 116 animals from 24 litters were used in this study.Animals were excluded from final analyses if they met any of the following criteria: (i) were an LPS non-responder defined as an increase in body weight after the first LPS injection (i.e., from P1 to P2) that was >2 standard deviations of the mean weight gain in LPS animals from P1 to P2 (i.e., 0.39 + 0.95 g; n = 1 LPS); and (ii) if there was evidence of cortical damage caused by EEG electrode implantation (see details below) at postmortem and/or excessively noisy EEG record on both hemispheres that was unable to be filtered (n = 5 control; n = 2 LPS).There were also two deaths in the LPS group (mortality rate of 1.79 %; one death occurred after the first LPS injection, and one after the second LPS injection).Thus, a total of 106 animals were included in the final study (n = 55 control; n = 51 LPS).The percentage of males and females was 57 % and 43 %, respectively, in the control group, and 45 % and 55 %, respectively, in the LPS group.Note that a minimum of two independent litters were used for each specific outcome measure/timepoint.The numbers of animals used in each specific outcome measure are detailed in the respective figure legends.

EEG head-mount implantation surgery
EEG head-mount surgery was performed in a subset of pups randomly selected from control and LPS groups (n = 33; ~2-3 animals per litter from 18 litters; n = 17 control, n = 16 LPS).Animals used for EEG experiments were not used for other outcome measures.Animals in the EEG litters not receiving EEG surgery were used for other outcome measures.Prior to surgery, we performed scent familiarization to reduce the potential for maternal rejection of pups following surgery or for dams to chew on EEG head mounts.Cotton swabs coated with the chemicals used in EEG head-mount surgery were left to dry for 1 h, and then placed in each home cage for 5 min daily for the 4 days prior to surgery (i.e., from P2-P5 for P5 surgeries, or P7-P10 for P10 surgeries).
Note that two surgical ages were utilized because some animals receiving head-mount surgeries for P7 EEG recordings showed poor/ noisy EEG recordings at P21 (likely due to head growth over this time).
For EEG head mount surgery, pups at P5/P6 (for EEG recordings initiated at P7) or P10 (for EEG recordings initiated at P14) were anaesthetized with 3 % isoflurane (in 1.5 L/min O 2 ) and maintained at 1 % over the course of surgery.Animals were administered subcutaneous Temgesic (Indivior Pty Ltd, Macquarie Park, New South Wales, Australia; 30 µg/mL at 1 μL/g body weight) for post-operative analgesia.
All surgical instruments were sterilized with 70 % ethanol.The head of the pup was cleaned with 70 % ethanol, and a rostro-caudal midline incision was made with a scalpel.The skin was then removed to form an ~1.2 cm 2 surgical field, the periosteum removed, and the skull cleaned using a cotton swab with acetone (Sigma-Aldrich Inc., St. Louis, MO, USA).A small animal drill (Stoelting Co., Wood Dale, IL, USA) with a 0.310″ gauge burr (SDR Scientific Chatswood, New South Wales, Australia) was used to drill four burr holes through the skull for bilateral electrode placement (approximately 4-5 mm lateral of the mid-sagittal suture, 2-3 mm posterior of Bregma, and 2-3 mm anterior of Lambda).Silver ball electrodes (1-mm diameter; fashioned from 0.5-mm thick silver wire; Sigma-Aldrich Inc.) were inserted into the burr holes to the level of the dura, and then sealed in place with a layer of cyanoacrylate glue.A layer of dental cement and accelerator (Rocket heavy red; Henry Schein Ltd., Auckland, New Zealand) was then applied to the skull to form a stable base for the EEG head mount.The electrode wires were soldered to a three-channel EEG head mount (8235-SM; 6-pin surface mount mouse head mount; Pinnacle Technology, Lawrence, KS, USA; 0.15 g, 3 × 5 mm) via intermediate wires soldered to the headmount connector prior to surgery, and the head mount and exposed wires were then secured with dental cement/accelerator.The in-built ground of the head mount was placed caudally to the electrodes.The duration of the surgery (including induction) was approximately 40 min per animal.Anesthesia was then terminated, and the pups were recovered for at least 1 h in a warmed chamber, and provided with oral saline for hydration and cleaned with sterile saline, before being returned to their dams.There were no animal losses related to these anesthesia, surgery, and recovery protocols.

Recording of EEG activity
All EEG recordings were acquired once daily for 30 min (between 10:00 to 12:00 h) at P7, P14, and P21.Individual pups were removed from their home cages, placed in a temperature-controlled recording chamber, and the head mount tethered to a preamplifier (PAL-8200 data acquisition system; Pinnacle Technology) attached to a lightweight and flexible 7-inch cable connected to a commutator located above the chamber.This setup allowed free movement of the pups in the recording chamber.The bilateral EEG signals were amplified (100×) and high-pass filtered (0.5 Hz) at the preamplifier.Data were then low pass filtered (50 Hz), and raw data digitized (with a sampling rate of 400 Hz) using Sirenia acquisition software (v2.2.3;Pinnacle Technology) (Ranasinghe et al., 2015).
The raw EEG traces (for both hemispheres) from control and LPS animals were manually reviewed for the presence of obvious artefacts or noise, which were marked and removed.These cleaned datasets were then used for spectral and time-frequency analyses.For spectral analyses, data were cut into 5-s epochs with 50 % overlap, and then a fast Fourier transform with a Hanning taper (0.2-40 Hz in 0.2 Hz increments) was performed.For each animal, spectral power data from both hemispheres (if available) were averaged, and the overall power (total sum of power from 0.2-40 Hz), spectral edge frequency 90 (SEF 90 ; closest frequency at which 90 % of total power was below), absolute power in the individual delta [0.2-4 Hz], theta [4-8 Hz], alpha [8-13 Hz], and beta [13-40 Hz] bands, and relative power (i.e., absolute power in each frequency band relative to absolute power in all frequency bands) were calculated.Spectral power was log-transformed using 'dB=10 × log (spectral power)' for visualization purposes.Time--frequency analysis was performed using a single Hanning taper convolution every 1 s with a variable window size equal to five cycles of the frequency of interest.The frequencies of interest ranged from 0.2-40 Hz in 0.2 Hz increments.Data from both hemispheres (if available) were averaged and used for qualitative assessment of mean power spectrograms (i.e., time-frequency plots) and power spectral density plots.For power spectrograms, spectral power was displayed with a fixed color bar from 0-100 µV 2 using the 'jet' color scheme (blue-greenred).As artefacts randomly occurred throughout the data causing gaps throughout the 10-min period, missing data were ignored and the mean time-frequency power spectrum calculated from available data.

Assessment of pyramidal neuronal morphology in the motor cortex
Golgi-Cox analysis of pyramidal neuron morphology was performed in a total of 37 brains (from four litters) collected at P7 (n = 9 control [8 male, 1 female]; n = 9 LPS [3 male, 6 female]) or P14 (n = 10 control [6 male, 4 female]; n = 9 LPS [7 male, 2 female]).Note that additional sham animals were collected at P1 (n = 2) and P2 (n = 2) to highlight neuronal complexity at the time of LPS exposure (see Fig. 2A, E).At postmortem, whole brains were rapidly extracted and processed for Golgi-Cox impregnation using the FD Rapid GolgiStain™ kit (FD Neurotechnologies, Inc., Columbia, MD, USA), as per the manufacturer's instructions.Each brain was then sectioned serially in the coronal plane (150-µm thick) using a cryostat (Leica CM3050S; Leica Biosystems, Wetzlar, Germany).The sections were mounted onto gelatin-coated slides, processed for Golgi visualization, dehydrated in a graded alcohol and xylene series, and then mounted with DPX non-aqueous mounting medium (Merck Millipore Co., Burlington MA, USA), as reported (Zaqout and Kaindl, 2016).
Region-matched brain sections at the level of bregma/mid-striatum were used to assess the dendritic morphology of cortical pyramidal neurons in the motor cortex (Neurolucida software; MBF Bioscience Inc., Williston VT, USA) (Prasad et al., 2021;Dean et al., 2013).For each brain section, the motor cortex region of interest (ROI; encompassing both the primary motor and secondary motor cortices; defined using age-specific atlases ( (Paxinos and Watson, 2013;Ramachandra and Subramanian, 2016); e.g., see Fig. 3A and (Prasad et al., 2021)) were outlined on both hemispheres under a 5 × objective (NA = 0.1) using a Zeiss Axio Imager M2 upright microscope (Carl Zeiss, Oberkochen, Germany) and a ZEISS Axiocam 506 color digital camera (Carl Zeiss).Using the meander scan function under a 20× objective (NA=0.5),all pyramidal neurons (except for Betz cells [giant pyramidal neurons]) within the motor cortex ROIs were digitally selected when they met standard criteria (Prasad et al., 2021;Juarez et al., 2008)-i.e., (i) triangular-shaped soma and apical dendrite perpendicular to the pial surface, (ii) complete Golgi-impregnation of the neuron allowing for visualization of the entire dendritic arbor, (iii) neuronal soma and processes not obscured by other neurons, glia, or vasculature, and (iv) neurons exhibiting largely complete basilar dendritic tree with few broken or truncated processes.These criteria were then confirmed on the preselected neurons under a 40× objective (immersion oil, NA=1.3;Carl Zeiss) to provide a final population of pyramidal neurons for morphological analysis.No distinctions were made between subtypes of pyramidal neurons.This sampling procedure provides a randomized and unbiased selection of pyramidal neurons throughout the cerebral cortex, which is independent of cortical region/layer.
Using reflected light microscopy (Jespersen et al., 2012), image stacks (0.5-μm step size; throughout entire z-plane) containing the complete basilar arbor of 4-5 preidentified neurons evenly distributed throughout each motor cortex ROI (i.e., 8-10 neurons/image stacks per brain section) were captured and saved for offline analysis (number of neurons at P7: n = 80 control, n = 80 LPS; number of neurons at P14: n = 80 control, n = 72 LPS).Outlines of the neuronal soma and the entire basilar dendritic tree structure were traced from the image stacks in the x-, y-, and z-coordinates using the user-guided tracing function of Neurolucida 360 software (MBF Bioscience).Apical dendrites were excluded from analysis because of their high rates of truncation after tissue sectioning.Morphometric analysis of total dendritic length (summed lengths of all basal dendritic branches per neuron) and dendritic complexity, which includes the number of primary dendrites off the neuronal soma, total number of dendritic branches, total number of dendritic endings, and total dendritic length for basal dendrites, of all reconstructed neurons was performed with Neuroexplorer software (MBF Bioscience).(caption on next page) P. White et al. Brain Behavior and Immunity 121 (2024) 104-118 2.6.MRI assessment of cortical microstructure MRI was performed in a total of 32 brains (from four litters) collected at P7 (n = 8 control [6 male, 2 female]; n = 8 LPS [4 male, 4 female]) or P14 (n = 8 control [6 male, 2 female]; n = 8 LPS [3 male, 5 female]).At postmortem, whole brains were rapidly extracted, immersion fixed in 4 % paraformaldehyde (Sigma-Aldrich Inc.) at 4 • C for 1.5-2 days, and stored in 0.1 M phosphate buffer at 4 • C for a minimum of 6 weeks until imaging.Prior to imaging, rat brain tissues were immersed in Fomblin©, a fluorinated lubricant that does not transmit any MR signal.Ex vivo MRI image acquisition was performed using an actively shielded 9.4 T/ 31 cm magnet (Varian Inc., Steinhausen, Switzerland) equipped with 12 cm gradient coils (400 mT/m, 120 µs) and a transceiver birdcage radiofrequency coil (2.5-cm diameter).A spin-echo sequence was used to acquire a multi-shell diffusion-weighted imaging protocol.A field of view of 20 × 15 mm 2 (P7) or 23 × 17 mm 2 (P14) was sampled on a 128 × 92 Cartesian grid.Brain slices (slice thickness = 0.6 mm; P7: ~7 slices; P14: ~8 slices) were acquired in the axial plane, centered on the level of bregma/mid-striatum (Fig. 3A) and largely enclosing the genu to the splenium of the corpus callosum (e.g., P7: plates c7-c14; P14: plates c8-c22 (Ramachandra and Subramanian, 2016).
Scans were averaged three times, with an echo time of 45 ms and a repetition time of 2000 ms.For each brain, 96 diffusion-weighted images were acquired.Of these, 15 were b 0 reference images, while the remaining 81 non-collinear images were uniformly distributed in three shells using the following distributions, which are represented as the number of directions/b-value (s/mm 2 ): 21/1750 s/mm 2 , 30/3400 s/ mm 2 , and 30/5100 s/mm 2 .The DTI images were spatially normalized to a study-specific DTI template using DTI-TK (https://dti-tk.sourceforge.net/pmwiki/pmwiki.php),from which mean diffusivity (MD) and fractional anisotropy (FA) were derived.The acquired data were also fitted using the NODDI toolbox (v1.05; https://www.nitrc.org/projects/noddi_toolbox) (Zhang et al., 2012) to estimate the diffusion in each voxel as three independent compartments-the isotropic volume fraction, which represents water diffusing isotropically (i.e., with the diffusivity of free water); the intra-cellular volume fraction (FICV; also termed the neurite density index [NDI]), which represents the fraction or density of space occupied by cellular processes within a voxel; and the extra-cellular volume fraction, which represents diffusion hindered perpendicular to processes.The dispersion of water diffusion along a cellular process was also used to calculate the orientation dispersion index (ODI), which represents the angular variation of cellular process orientations (reflecting the bending and fanning of processes and areas of crossing fibers)-low ODI values represent tightly-aligned diffusion (i.e., aligned cellular processes/fibers), while high ODI values represent more dispersed diffusion (i.e., more dispersed cellular processes/fibers).A supplementary parameter (the intra-restricted volume fraction), which corresponds to water trapped in the tissue following fixation and without any diffusion (Dhital et al., 2018), was used to improve the analysis of our ex-vivo imaging data.
To avoid partial volume effects, direction-encoded color maps were used to manually draw an ROI in the motor cortex (defined using age-specific atlases (Paxinos and Watson, 2013;Ramachandra and Subramanian, 2016); e.g., see Fig. 3A and (Prasad et al., 2021)) of both brain hemispheres on each coronal brain slice (i.e., over the brain levels of the genu to the splenium of the corpus callosum) for each animal.The median DTI-(MD and FA) and NODDI-(NDI, ODI) derived parameter values within the motor cortex ROI were calculated for each MRI slice, and the median value across the different image planes calculated to obtain one dataset per animal (i.e., over all MRI brain slices).For secondary analysis, MRI data from the motor cortex were also examined at the individual brain slices.

Statistical analysis
All analyses were performed with statistical software (GraphPad Prism v9.0; GraphPad Software Inc., La Jolla, CA, USA).All data sets were confirmed to be normally distributed using the Shapiro-Wilks test.
For neuronal morphology and MRI data, the effects of treatment (control vs LPS) × age (i.e., P7, P14, and P21) or treatment × MRI slice were assessed by two-way ANOVA.For EEG data, the effects of treatment × age were assessed using a mixed effects model to account for repeated EEG measurements.When there was a significant interaction of treatment × age or treatment × MRI slice (p < 0.05), post hoc analysis was performed using the Tukey's multiple comparisons test.For all analyses, if there was both a significant main effect of treatment and an interaction of treatment × age/MRI slice, the main effect was presented.Specific details of the statistical tests performed in this study are provided in the figure legends.All quantitative data (except for MRI slice analyses) are presented as box plots of medians with 10 %-90 % confidence intervals.MRI slice data are presented as mean ± standard error.All analyses were performed by investigators blinded to the study groups.This study was not designed to test the effect of sex.

Postnatal systemic inflammation causes early disruptions in neuronal arborization in the developing cortex
We previously reported that repeated systemic LPS exposure in newborn rats caused long-term impairments in the arborization of basal dendrites of pyramidal neurons in the motor cortex (e.g., see P21 dataset in Fig. 2) (Prasad et al., 2021).In terms of cortical development, the P21 rat approximates a 2-3-year-old human (Semple et al., 2013).However, the timing of evolution of these neuronal dendritic deficits at earlier developmental ages is unknown.Herein, we performed threedimensional morphological reconstruction of basal dendritic development of pyramidal neurons in the motor cortex of P7 and P14 rats (equivalent to late preterm/early term to term human brain development, respectively (Semple et al., 2013) exposed to LPS from P1-P3.
Overall, these data suggest that postnatal systemic inflammation during a period of rapid cortical neuronal development causes an early but evolving dysmaturation of neuron dendritogenesis.

DTI and NODDI have limited capacity to identify early impairments in cortical neuron development after postnatal systemic inflammation
We previously reported that DTI and NODDI parameters could identify the long-term impairments in dendritic development in cortical neurons in the motor cortex of P21 rats (e.g., see P21 datasets in Fig. 2 and Fig. 3) exposed to repeated LPS at birth (Prasad et al., 2021).However, the utility of DTI and NODDI for detecting the evolution of these neuronal dendritic deficits at early developmental ages (e.g., see P7 and P14 datasets in Fig. 2) is unknown.Herein, we utilized ex vivo high-field advanced diffusion imaging of the motor cortex of P7 and P14 rats exposed to LPS from P1-P3 to examine the capacity of DTI and NODDI to detect early alterations in cortical neuronal development.
Representative images of diffusion-derived parameters from control and LPS rats are shown in Fig. 3A, D, G, J.There were no differences in the median FA (Fig. 3B), MD (Fig. 3E), ODI (Fig. 3H), or NDI (Fig. 3K) values calculated over all MRI brain slices in the motor cortex between control and LPS rats at P7 or P14, while there was a significant increase in FA (Fig. 3B) and decrease in ODI (Fig. 3H) in LPS rats at P21. Next, we examined changes in cortical MRI parameters over individual caudal to rostral MRI brain slices (Fig. 3C, F, I, L).LPS rats showed no changes in FA at P7 (Fig. 3C), but significantly higher FA at brain slices 4, 6, and 7 (i.e., more frontal brain slices) at P14 (Tukey's post hoc, p < 0.05 for all), compared with controls.By contrast, LPS rats showed no differences in ODI (Fig. 3F), MD (Fig. 3I), or NDI (Fig. 3L) at P7 or P14 between the two groups.
Overall, these data suggest that DTI and NODDI have limited capacity to identify the early evolution of cortical neuronal dysmaturation after postnatal systemic inflammation.

Postnatal systemic inflammation causes early and persisting impairments in the maturation of cortical activity assessed by EEG
EEG provides a direct measure of neural activity, which is related to the structure and function of neurons.Thus, we utilized serial EEG recordings in awake rats at P7, P14, and P21 rats to examine its capacity to detect deficits in cortical neuronal development after postnatal systemic inflammation induced by repeated systemic LPS exposure at P1-3.
As previously reported in rodents (de Camp et al., 2017;Akman et al., 2018), qualitative examination of raw EEG traces in control rats showed evidence of discontinuity at P7 (Fig. 4A), with maturation to predominantly continuous activity by P14 (Fig. 4D) and P21 (Fig. 4G).By contrast, rats exposed to LPS generally showed more evidence of discontinuity at P7 (Fig. 4B), which evolved into a more continuous activity by P14 (Fig. 4E) and P21 (Fig. 4H).Power spectrum analysis showed a progressive developmental increase in the EEG spectral power in both control and LPS rats (power spectrograms: control group, Fig. 4A, D, G; LPS group, Fig. 4B, E, H; power spectral density plots: Fig. 4C, F, I).In control rats, the peak power density occurred in the delta range (δ; 0.2-4 Hz) at P7 (Fig. 4C; i.e., peak at 2.0 Hz), with a developmental shift to theta oscillations (θ; 4.0-8 Hz) by P14 (Fig. 4F; i.e., peak at 4.6 Hz), and progressive appearance of alpha (α; 8-13 Hz) and beta (β; 13-40 Hz) rhythms (Fig. 4C, F, I).By contrast, LPS rats showed a higher overall EEG power at all ages compared with controls (power spectrograms: Fig. 4B, E, H; power spectral density plots: Fig. 4C, F, I).Further, LPS rats showed a delay (by ~7 days) in the developmental shift in peak power density from delta to theta oscillations (i.e., the peak power density appeared in theta band by P14 in controls vs P21 in LPS rats; see Fig. 4L).
Quantitative EEG assessment also showed a progressive developmental increase in the overall EEG power in both control and LPS rats (Fig. 4J), and a significantly higher overall EEG power in LPS rats compared with controls (at P7 and P14: Tukey's post hoc, p < 0.05 for both; Fig. 4J).Further, in support of the delay in the developmental shift of peak power density from delta to theta oscillations (see Fig. 4C, F, I), LPS rats showed a significantly lower SEF 90 compared with controls (at P7 and P14: Tukey's post hoc, p < 0.01 for both; Fig. 4K).
Given the potential frequency-dependent effects of LPS on EEG power shown by power spectrum analysis, we next examined the specific changes in the absolute and relative EEG power in the individual delta, theta, alpha, and beta frequency domains.As for overall EEG power (see Fig. 4J), both control and LPS rats showed a progressive developmental increase in the absolute EEG power in all individual frequency bands (Fig. 5A, C, E, G).However, LPS rats showed a significantly higher absolute EEG power in the delta band at P7 and P14 (Tukey's post hoc, p < 0.01 for both; Fig. 5A), theta band at P7 and P21 (Tukey's post hoc, p < 0.05 for both; Fig. 5C), and alpha band at P21 (Tukey's post hoc, p < 0.05; Fig. 5F) compared with controls.There were no differences in absolute power in the beta band between the two groups (Fig. 5G).Relative spectral power analysis (i.e., absolute power in each frequency band relative to absolute power in all frequency bands) showed a predominance of delta oscillations at P7 in both control and LPS rats, and a developmental decrease in the proportion of delta band activity (Fig. 5B) and increase in the proportion of theta band activity (Fig. 5D).However, LPS rats showed a significantly higher relative delta power at both P7 and P14 (Tukey's post hoc, p < 0.01 for both; Fig. 5B), as well as a significantly lower relative beta power at both P7 and P14 (Tukey's post hoc, p < 0.005 for both; Fig. 5B), compared with controls.There were no differences in relative theta (Fig. 5D) or relative alpha (Fig. 5F) power between the two groups.
Overall, these data provide evidence of persisting immature EEG features that overlap with the timing of cortical neuron dysmaturation after postnatal systemic inflammation.

Discussion
The present study examined the structure-function relationships between developmental changes in cortical neuron maturation, MRIderived measures of cortical microstructure, and EEG-derived measures of cortical maturation after postnatal systemic inflammation in newborn rats.Our main findings were that early postnatal inflammation, during a period of rapid cortical neuronal development, led to early and evolving deficits in cortical neuronal dendritogenesis, with associated disturbances in the normal maturation of cortical EEG activity.By contrast, DTI and NODDI had a limited capacity to identify the early evolution of cortical neuronal dysmaturation.Thus, EEG may be particularly useful for identifying the evolution of impaired cortical development after early life postnatal systemic inflammation, while DTI and NODDI may be more suited to assessment of more established injury/dysmaturation.
Persisting impairments in the growth, microstructural development, and connectivity of the cerebral cortex are major pathological features associated with neurodevelopmental disability in modern cohorts of children born preterm (Ma et al., 2022;Kelly et al., 2023;Dimitrova et al., 2021;Toulmin et al., 2021).Experimentally, we and others have  A, B, D, E, G, H) Representative raw cortical EEG traces (60 s) and mean power spectrograms (i.e., time-frequency plots; 10 min [600 s]) are shown from awake control (A, D, G) and lipopolysaccharide (LPS; B, E, H) rats at postnatal day (P)7 (A, B), P14 (D, E), and P21 (G, H).For power spectrograms, spectral power was displayed with a fixed color bar from 0-100 µV 2 using the 'jet' color scheme (blue-green-red).(C, F, I) Power spectral density plots are presented as mean values of control (blue lines) and LPS (red lines) rats at P7 (C), P14 (F), and P21 (I).The delta (δ; 0.2-4 Hz), theta (θ; 4-8 Hz), alpha (α; 8-13 Hz), and beta (β; 13-40 Hz) frequency bands are defined.(J, K) Quantitation of the overall power of cortical EEG activity (J) and the 90 % spectral edge frequency (SEF 90 , i.e., the frequency below which 90 % of the power resides; K).Overall EEG power and SEF 90 are presented as box plots (median with 10 %-90 % confidence intervals; mean values are indicated by + ).(L) Frequency of the peak in power spectral density data from (C, F, L).For all graphs: P7 (control: n = 9; LPS: shown that such cortical deficits involve chronic impairment of maturation of cortical neurons, including reduced growth, complexity, and connectivity of neuronal dendrites (Prasad et al., 2021;Stolp et al., 2019;Kelly et al., 2023;Dean et al., 2013;Ardalan et al., 2019).For example, in the same newborn rat model of LPS exposure as the present study, we reported impaired arborization and spine formation of basal dendrites of cortical pyramidal neurons at ~3-weeks recovery (i.e., at P21) (Prasad et al., 2021)-this age approximates a cortical development of 2-3-year-old humans (Semple et al., 2013).Similarly, hypoxia-ischemia in preterm fetal sheep was associated with altered cortical neuron arborization at 4 weeks of recovery (i.e., near-term) (Dean et al., 2013).Two small case studies also reported evidence of reduced dendritic length of cortical neurons in very preterm infants with either diffuse (Stolp et al., 2019) or severe necrotic (Takashima et al., 1982) white matter injury.However, to our knowledge, the timing of evolution of these neuronal dendritic deficits after injury have not been reported.Using Golgi-Cox analysis of pyramidal neuron morphology, we found that LPS animals developed mild disruptions in measures of basal dendritic complexity (i.e., total numbers of dendritic branches and dendritic endings) by P7, which persisted at P14 and P21 (note the largest differences were seen at P21).Importantly, however, these dendritic complexity changes did not translate into a reduction in total dendritic length until P21.These data suggest an early but evolving dysmaturation of neuronal dendritogenesis after early postnatal systemic inflammation.Given the progressive expansion of neuronal dendritic complexity and total dendritic length that occurs after birth in both control and LPS animals (e.g., see Fig. 2), this evolving injury likely relates to a slower trajectory of dendritic outgrowth that becomes progressively distinct from control animals over time.Nevertheless, despite the very simple neuronal dendritic morphology observed in the P1-3 rat, we cannot rule out potential for acute overt neuronal damage (i.e., dendritic degeneration or retraction) during the inflammatory insult itself.Further studies are required to assess dendritic complexity during and immediately after LPS exposure in this model, as well as to determine whether this neuronal dysmaturation relates to a primary neuronal insult, or evolves secondary to other changes in the brain such as white matter injury and impaired axonal myelination (Volpe, 2021).
The present finding of progressively evolving cortical injury after early LPS exposure is supported by longitudinal studies in infants born preterm showing progressive impairments in cortical growth up to termequivalent age, which were associated with reduced neurocognitive abilities at 2 years (Rathbone et al., 2011;Ball et al., 2013) and 6 years of age (Rathbone et al., 2011).In turn, numerous cross-sectional studies have shown that these cortical volume deficits in infants born preterm persist into adolescence and adulthood (Nosarti et al., 2008;Pascoe et al., 2019).A similar progressive impairment in cortical growth was reported in a preterm fetal sheep model of hypoxia-ischemia (Dean et al., 2013), with comparable cortical growth to controls at 1-week recovery, but increasing deficits in cortical growth deficits at 2-weeks and 4-weeks recovery, as well as reduced dendritic arborization at 4 weeks.
The present study also assessed the utility of diffusion MRI for monitoring the evolution of cortical neuronal dysmaturation.The human cortex shows a developmental decrease in FA (Ball et al., 2013;Batalle et al., 2019) and increase in ODI (Eaton-Rosen et al., 2015;Batalle et al., 2019) up to term age, which is thought to reflect the progressive increase in neuronal dendritic outgrowth and complexity that occurs during this period (Mrzljak et al., 1992).Human studies also show evidence of impaired development of cortical microstructure (e.g., higher cortical FA and lower cortical NDI/ODI) in some preterm born infants when measured at term-equivalent age or later (Dimitrova et al., 2020;Dimitrova et al., 2021;Wang et al., 2022;Bouyssi-Kobar et al., 2018), which may relate, at least in part, to reduced neuronal complexity (i.e., neuronal dysmaturation (Volpe, 2021)).In contrast to our hypothesis, we found no changes in any DTI or NODDI parameters in the motor cortex (calculated over all MRI brain slices) at P7 or P14 in LPS rats, despite evidence of mild deficits in dendritic complexity at this time.However, as we reported (Prasad et al., 2021), by P21 there was a higher FA and lower ODI (both indicative of reduced dendritic complexity) in the motor cortex of LPS rats, at a time when more marked deficits in dendritic outgrowth were evident.These data suggest that DTI and NODDI have limited capacity to identify milder deficits in neuronal development, such as those during the early evolution of cortical neuron dysmaturation after postnatal systemic inflammation, but may be appropriate for detecting more marked neuronal deficits, such as those during later evolution of injury or with more severe or protracted insults.This is supported by studies showing neurodevelopment delay but without MRI evidence of brain injury in some infants (Banihani et al., 2021;Wu et al., 2023), and may also partly explain why brain MRI at term-equivalent age has a greater prognostic power (including for neurocognitive function) than earlier assessment (Banihani et al., 2021;Van't Hooft et al., 2015).Note that when we examined changes in cortical MRI parameters over individual MRI brain slices, there was a significant increase in cortical FA at more rostral (i.e., frontal) brain levels in LPS animals at P14, which fits with the global increase in cortical FA observed in LPS animals at P21.Given that cortical maturation can progress in a 'back to front' manner (Gogtay et al., 2004), there may be differential timing of cortical dysmaturation depending on the specific brain level and the timing of MRI measurement.Thus, assessment of MRI data using brain level-based approaches may also be useful in the assessment of preterm brain dysmaturation.
During preterm human brain maturation, cortical EEG activity transitions from a discontinuous pattern with a predominance of highamplitude low-frequency delta waves, to more continuous higher-frequency/lower-amplitude delta, theta, and beta oscillations by termequivalent age and into infancy (Bourel-Ponchel et al., 2021;Shany et al., 2014).In terms of EEG spectral power, these EEG features manifest as a gradual decrease in relative power in the delta range, and a progressive increase in relative power in the higher frequency ranges (i.e., a shift towards the higher frequency ranges), leading to increased SEF (Bell et al., 1991;Niemarkt et al., 2011).We found a comparable shift from a predominance of delta oscillations in the P7 rat to higher frequency EEG activity by P14 and P21.Critically, however, this development shift to higher frequency EEG activity was delayed in LPS rats-this included a significant impairment of the developmental decline in relative delta power (i.e., a higher relative delta power), a lower relative beta power, and a lower SEF 90 compared with controls.This increase in the proportion of lower frequency (but higher power) EEG activity was the likely driver of the increases in overall cortical power, and in total power in the lower frequency ranges, observed in LPS rats.Importantly, the timing of this developmental delay in EEG maturation in LPS rats largely overlapped with the timing of impaired cortical neuronal development.Thus, evidence of persisting immature EEG features, particularly those related to lower frequency delta activity, may be useful for identifying the early evolution of impaired cortical neuronal development after postnatal systemic inflammation.Of note, these immature EEG features in LPS rats tended to resolve by P21 (except for absolute power in the theta and alpha frequency ranges), despite ongoing impairment of neuronal dendritogenesis at this time.The reasons for this are unclear, although it is feasible that maturation of other brain circuits may predominate the EEG signature at this time, and thus obscure these immature EEG features.Nevertheless, this apparent EEG recovery does not preclude future neurodevelopmental problems.Alternatively, there may be some repair of brain function post injury in our model.Further studies examining longer-term EEG outcomes are required to determine whether this catch up persists at older ages, and whether other EEG features may be more suitable for assessing brain dysmaturation as this time.
Delays in the developmental maturation of EEG activity have also been reported in preterm infants, even in the absence of overt brain injury.For example, using serial EEG measurements initiated with 72 h of birth and continued to term-equivalent age, Okumura et al. reported dysmature EEG patterns (defined as presence of some EEG features that are observed in normal infants at ≥2 weeks younger, including persisting high-amplitude low-frequency delta waveforms) in infants born <33 weeks gestation, which included those without severe white matter lesions or intraventricular hemorrhage (Okumura et al., 2002).Similarly, preterm infants (born <27-29 weeks gestation) without major structural brain lesions had an increased incidence of immature EEG waveforms (e.g., delta brushes and theta sawtooth patterns) at ages ≥34 weeks compared with infants born at later gestational ages (Conde et al., 2005).Further, preterm infants (born <27 weeks) without overt brain damage showed higher relative alpha-band power, and lower relative theta-, alpha-, and beta-band power, on EEG measured at discharge (~35 weeks old) compared with healthy late-preterm born infants (~35 weeks gestation) (Suppiej et al., 2017).Sher et al. also reported reduced relative power in the theta-through to beta-frequency ranges in preterm infants born <32 weeks when assessed at term-equivalent compared with term-born infants (Scher et al., 1994).Interestingly, a number of studies have shown that persisting dysmature EEG patterns in preterm infants were correlated with later cognitive impairment and were largely independent of white matter imaging findings (Kong et al., 2018;Selton et al., 2013;Hayakawa et al., 1997), suggesting that dysmature patterns may be more related to cortical neuronal disturbances than white matter lesions (Hayakawa et al., 1997).
There have been few studies examining maturational changes in EEG associated with perinatal infection/inflammation.In line with our findings, in a small study very preterm infants with severe bronchopulmonary dysplasia and requirement for mechanical ventilation (these are associated with systemic inflammation (Bose et al., 2013;Cakir et al., 2023)), ~64 % showed dysmature EEG patterns at near-term equivalent (Hahn and Tharp, 1990).Further, extremely preterm infants treated for sepsis during their NICU stay exhibited lower functional brain maturation assessed by amplitude-integrated EEG at near-term age compared with those who had not received treatment for sepsis (Lee et al., 2021).By contrast, in a smaller study of extremely preterm infants, there were no differences in amplitude-integrated EEG maturation slopes (measured at 28, 32, and 36 weeks) between those with or without clinical sepsis (Helderman et al., 2010).Experimentally, male mice exposed to a 'two-hit' model of embryonic polyinosinic:polycytidylic acid (12.5 d gestation) and postnatal LPS (P9) also showed a pattern of increased relative delta and theta activity, and decreased relative alpha and beta activity, at 7 weeks of age (Missig et al., 2018).Similarly, repeated low dose LPS exposure in preterm fetal sheep was associated with delayed cortical EEG maturation characterized by increased relative delta power, decreased relative alpha and beta powers, and decreased SEF at 7-10 days after LPS initiation (Keogh et al., 2012).Further, in near-term fetal sheep, progressive systemic inflammation induced by LPS exposure was associated with increased relative delta power, reduced relatively beta power, and reduced dendritic complexity and spine density of cortical neurons at 4 days after LPS initiation, with a positive correlation between mean relative beta band power during the final 12 h of the experiment and dendritic length (Kelly et al., 2023).
Despite the temporal relationship between dysmaturation of cortical EEG and pyramidal neuron development observed in the present study, the specific contributions of these neuronal changes to the observed EEG changes are unclear.For example, there may be more widespread changes in neuronal maturation and circuit development in and between other cortical and subcortical regions including the subplate and thalamus-these structures contribute to the development of mature EEG (Wallois et al., 2021) and are known to be altered in preterm brain injury (recently reviewed in (Volpe, 2021)).Deficits in interneuron development and inhibitory GABAergic circuitry have also been reported in preterm brain injury (Robinson et al., 2006;McClendon et al., 2014), including after postnatal inflammation (Stolp et al., 2019;Ardalan et al., 2019), which may contribute to the persistence of immature cortical EEG features after LPS exposure in the present study.Indeed, the maturation of inhibitory GABAergic circuitry is thought to regulate the developmental shift from discontinuous to continuous EEG (Vanhatalo et al., 2005), and the predominance of higher frequency activity including beta oscillations (Porjesz et al., 2002;Vanhatalo and Kaila, 2006), both of which were altered in the present study.Finally, white matter damage (including deficits in axonal myelination) is a major component of preterm brain injury (Guillot and Miller, 2021), including in the present rat model (Prasad et al., 2021).As axonal myelination is an important component of brain circuit development and maturation of the EEG signal (Nunez et al., 2015), white matter injury may also contribute to the observed changes in cortical EEG in the present study.Note that a limitation of our study was that EEG, MRI, and Golgi-Cox analyses were performed in different animals, which limited our capacity to perform correlations between changes in these parameters.Further studies are also required to assess functional outcomes and how these correlate with EEG findings.
Overall, the present study suggests that EEG and MRI are likely to be complementary tools for assessing changes in cortical development in preterm infants.EEG was most effective for identifying the early onset and evolution of cortical neuronal dysmaturation, while DTI and NODDI seem to be more suited to identifying more established neuronal changes at term-equivalent age, when EEG may be less sensitive.Early identification of infants at heightened risk of adverse neurodevelopmental outcomes may enable further research into targeted habilitative or treatment strategies to reduce the adverse long-term consequences of preterm birth.
the work reported in this paper.

Fig. 1 .
Fig. 1.Experimental design.Neonatal rats received single daily intraperitoneal (i.p.) injections of lipopolysaccharide (LPS) or saline on postnatal days (P)1-P3, and were recovered to a range of timepoints for various outcome measures.Surgery for electroencephalography (EEG) electrode and head mount placement was performed at P5/P6 for animals receiving EEG recordings initiated at P7, or at P10 for animals receiving EEG recordings initiated at P14.Note that separate animals were used for the tissue collection, magnetic resonance imaging (MRI), and EEG studies.Figure created with BioRender.com.