Maternal selenium dietary supplementation alters sociability and reinforcement learning deficits induced by in utero exposure to maternal immune activation in mice

,


Introduction
While autism spectrum disorder (ASD) and schizophrenia are traditionally considered separate conditions, they share multiple risk factors (Hamlyn et al., 2013) and exhibit similar changes in sociability and cognition (Eack, 2013;Stone and Iguchi, 2011;Solomon et al., 2011;Geana, 2021).One risk factor of particular relevance to both conditions, given the recent COVID-19 global pandemic, is viral infection during pregnancy (Atladóttir, 2010), which is thought to indirectly disrupt foetal brain development via the maternal proinflammatory response (Smith et al., 2007).Animal models of maternal immune activation (MIA) recapitulate a spectrum of behavioural and neurobiological phenotypes associated with ASD and schizophrenia, making them highly valuable tools for assessing the mechanism by which MIA causes neurodevelopmental dysfunction, and for testing the efficacy of novel treatments.Disruptions in foetal brain development following MIA may plausibly occur through various mechanisms, such as the priming of microglia (Otero and Antonson, 2022), altered expression of neurotrophic factors (Vasistha, 2020), and increased oxidative stress (Guma, 2022;Oskvig et al., 2012).MIA-exposed offspring also exhibit increased expression of IL-1β that extends beyond early age (Hameete, 2021), indicating that the immune system may become permanently shifted towards a proinflammatory state.These findings mirror multiple studies investigating ASD and schizophrenia, which have reported an increase in reactive microglia and astrocytes (Matta et al., 2019;Bloomfield, 2016;Catts et al., 2014), elevated blood and brain concentrations of multiple proinflammatory cytokines (Nour-Eldine et al., 2022;Theoharides et al., 2016;Na et al., 2014) and increased oxidative stress markers in post-mortem brain tissue (Rose, 2012;Zhang et al., 2010).
Dietary supplementation with selenium (Se) has received recent interest due to its role in redox regulation (Steinbrenner and Sies, 2009) and anti-inflammatory effects (Wu, 2022).Selenium mediates its activity primarily through selenoproteins, which contain selenocysteine (Sec) residues co-translationally incorporated into their amino acid sequence (Roman et al., 2014).Several selenoproteins have been previously identified in the placenta, such as the main Se transporter selenoprotein P (SELENOP); the iodothyronine deiodinases (DIOs) 2 and 3, which play a key role in the supply of thyroid hormones to the developing foetus; the antioxidant selenoproteins glutathione peroxidases (GPXs) 2 and 4; and thioredoxin reductases (TRXRs) 1, 2 and 3, which are associated with a protection of the embryo due to their redox regulation activity (Ojeda et al., 2021).Early findings have shown potentially positive outcomes for Se supplementation, with previous studies demonstrating benefits in reducing the stress response (Torres et al., 2021) and for Alzheimer's disease (Cardoso, 2019).Further, Se has also been shown to reduce serum concentrations of IL-6, IL-1β and TNF-α in the Black and Tan Brachyury (BTBR) mouse model of autism (Wu, 2022).However, to our knowledge, no studies have investigated the effect of Se on offspring exposed to MIA.Therefore, the primary aim of this study was to use Se as a prophylactic pregnancy supplement to protect against MIA-induced behavioural deficits.We hypothesised that Se treatment would dampen the MIA-induced increase in inflammatory cytokines, and consequently prevent MIA-induced deficits in learning, cognitive flexibility and sociability behaviours relevant to ASD and schizophrenia (Eack, 2013;Stone and Iguchi, 2011;Solomon et al., 2011;Geana, 2021).
While associations between maternal Se deficiencies and neurodevelopmental delays in the unborn child are well established (Ojeda et al., 2021;Hogan and Perkins, 2022), supratherapeutic concentrations of Se in maternal blood have also been associated with neurodevelopmental delays and a dose-dependent increase in the risk of ASD, ADHD and other developmental conditions (Lee, 2021;Amorós, 2018).While this narrow therapeutic window has received much attention in clinical research, recommended daily and upper limits for Se consumption are inconsistent (Vinceti et al., 2018).The safety profile of Se as sodium selenate in pregnant rodents appears similarly narrow, with previous studies reporting no teratogenic effects at doses 1.5 ppm, but the loss of offspring at doses of 2.5 or 3 ppm (Schroeder and Mitchener, 1971;Rosenfeld and Beath, 1954).Thus, our study also sought to confirm the safety of sodium selenate at a dose of 1.5 ppm in pregnancy, given the narrow therapeutic window identified in previous studies (Vinceti et al., 2018).
Another risk factor for ASD and schizophrenia is maternal iron (Fe) (Schmidt et al., 2014;Insel et al., 2008) deficiency, while maternal deficiencies in other essential elemental nutrients, such as copper (Cu) (Skogheim, 2021), calcium (Ca) (Li, et al., 2018) and zinc (Zn) (Skogheim, 2021) have also been reported for ASD.As many of these nutrients play a significant role in brain development (Lozoff and Georgieff, 2006;Georgieff, 2007), and are sequestered during a proinflammatory response (Tomkins, 2003), it is possible that MIA may also disrupt foetal brain development by restricting the maternal supply of these metal ions.Accordingly, this study also aimed to quantify changes in a range of elemental ions in foetal brains during the acute phase of MIA and investigate if changes in these ions were modified by Se supplementation.

Animals and maternal immune activation
C57BL/6J mice were obtained from the Monash Animal Research Platform and held within the Monash University Animal Facility.Breeders were housed under specific pathogen-free conditions as trios and checked daily for vaginal plugs.Females positive for a vaginal plug were single-housed until weaning.The presence of a vaginal plug was defined as gestational day (GD) 0. Selenium (as sodium selenate) was supplemented in drinking water (1.5 ppm, equivalent to a dose of ~0.15 mg/kg) from GD9 to birth, while control dams were provided with standard drinking water.Previous findings have shown that Se added to the drinking water at a concentration of 12 ppm did not affect water consumption (van Eersel, 2010), thus our much lower (almost 10-fold) concentration is unlikely to impact water consumption.At GD17, dams also received a single intraperitoneal (i.p.) injection of polyinosinic:polycytidylic acid (poly-I:C, Lot# 118M4035V, Sigma-Aldrich, 20 mg/kg) or an equivalent volume of 0.9 % saline (w/v).
We selected GD17 as previous studies by our group and others have shown that offspring exposed to MIA in late gestation exhibit altered performance in cognitive tasks (Nakamura, 2021;Schroeder, 2019;Zuckerman and Weiner, 2005), which was a major outcome of this study.Following injection, the temperature of dams was taken using a FPIR-V22 Famidoc infrared thermometer in triplicate.Dams were otherwise left undisturbed until weaning at postnatal day (PD) 21.Two litters per group (control, poly-I:C, Se + control, Se + poly-I:C), were tested for molecular changes in foetal brains and placenta, while 3 control, 4 poly-I:C, 3 Se, and 2 Se + poly-I:C litters were tested for all behavioural measures.For behavioural experiments, offspring were transported to a reverse light-cycle behavioural facility (7 pm lights on, 7am lights off) at approximately 6 weeks of age and allowed to acclimatize for 4 weeks before behavioural experiments were conducted.All dams and offspring were handled daily for at least 1 week prior to breeding or behavioural assessments, and were weighed weekly from 9 weeks of age.Fig. 1 shows an overview of the experimental design.While the range of the testing period varied, there were no group differences in the age of mice at the time of testing for all tasks, except social interaction whereby the age of the mouse did not correlate with behavioural outcomes (data not shown).Additional methodological information can be found in the maternal immune activation reporting guidelines checklist (Supplementary material 1) (Kentner, 2019).Mice had ad libitum access to standard mouse chow and water, except when undergoing the baited Y-maze task.All experimental procedures were approved by the Monash University animal ethics committee (Ethics numbers MCCB/2017/10 and MCCB/2020/03) and adhered to the ARRIVE guidelines of animal care.

Collecting and sexing foetal brains and placentas
Two dams per treatment group were euthanised by cervical dislocation 6 h following the injection of poly-I:C or saline, and foetal brains were isolated and snap-frozen on dry ice.In order to sex foetuses, a small sample of brain or limb tissue was collected during dissection.RNA was extracted using a Qiagen RNeasy kit (Cat# 74004), and residual DNA was digested on the column with Dnase I, according to the manufacturer's instructions.cDNA (200 ng) was synthesised from the RNA using a High-Capacity cDNA Reverse Transcription Kit (Applied Biosystems, Catalog #4368814) and quantified using a NanoDrop 1000 spectrophotometer.The sex of each foetus was determined using RT-qPCR (QuantStudio 6) or RT-digital PCR (Absolute Q) for the Y-chromosome gene Eif2s3y.

Sample preparation for ELISA, Luminex assays, and ICP-MS
Foetal brains and placentas were homogenised in ice-cold PBS (pH 7.4) and supplemented with 0.5 % (v/v) protease inhibitor cocktail set III (Calbiochem Cat #539134) and 2 % (v/v) phosphatase inhibitor cocktail set IV (Millipore, Cat #524628).Foetal brain homogenate (50 µL) was aliquoted for ICP-MS to measure concentrations of magnesium (Mg)-24, Ca-44, Fe-56, Zn-66, Cu-63 and Se-80.The remaining homogenate was left on ice for 10 min, solubilised on a rotator for 60 min at 4 • C, and then centrifuged at 14,000 × g for 15 min at 4 • C. The supernatant was stored at − 80 • C and used later in the ELISA and Luminex assays to assess inflammatory markers.

ELISA and Luminex assays
The total protein concentration for each foetal brain and placental sample was quantified using a BCA assay (Thermo Fisher Scientific, Cat #232255), according to the manufacturer's instructions.The concentration of the cytokines IL-6, IL-1β, IL-10, IL-17, TNF-α and MIP-2 were determined in foetal brains using 2 Milliplex Mouse Cytokine/Chemokine Magnetic Bead Panel kits (Millipore Cat# MCYTOMAG-70K, Lot#377336 (Plate 1) Lot#3833897 (Plate 2)), following the manufacturer's instructions.Briefly, ≤ 204 µg total protein was loaded in duplicate across two plates and incubated with magnetic beads coated with antibodies for all analytes.The concentration of each analyte was measured on a MAGPIX system with xponent software (Luminex Corp., Austin, TX, USA).For each plate, a 5-parameter logistic regression standard curve was generated for each analyte to calculate the concentration in each sample.
Singleplex ELISA plates were used to measure IL-6 (Elizakit.com,Cat #EK-0029), IL-1β (Invitrogen, Cat#BMS6002) and IL-17 (Invitrogen, Cat#BMS6001) in placental tissue.250 µg protein was loaded in duplicate for the IL-6 ELISA and 750 µg of total protein was loaded in duplicate for the IL-1β and IL-17 ELISA.All ELISA plates were run according to the manufacturer's instructions.The concentration of each sample was determined from a 5-parameter logistic regression standard curve for each analyte.The concentration of any sample less than the lowest standard was defined as half the concentration of the lowest standard (n = 2 for IL-1β, n = 9 for IL-17).

Inductively coupled plasma-mass spectrometry (ICP-MS)
Elemental measurements of foetal brain tissue homogenates were performed on an Agilent 8900 triple quadrupole inductively coupled plasma-mass spectrometer (ICP-MS, Agilent Technologies).Samples were introduced directly from 1.5 mL polypropylene tubes via an integrated automation system autosampler (Agilent Technologies) using a peristaltic pump.Sample desolvation was performed using a MicroMist nebulizer (Glass Expansion, Australia).The instrument was calibrated for elements of interest using mixed 0, 2.5, 5, 10, 25, 50, 100, 125, 250 and 500 µg L − 1 standard calibration solutions in 1 % (v/v) nitric acid (HNO 3 , Suprapur, Merck, Australia) from commercially available certified reference standards (Multi-element Calibration Standard 2A, Agilent Technologies).A reference element solution containing 100 µg L − 1 yttrium (Y) (Agilent Technologies) was introduced via T-piece positioned after the peristaltic pump and was used to normalise all measurements.The tuning solution for the instrument contained 1 µg L − 1 of cerium, cobalt, lithium, thallium, and Y in 2 % (v/v) HNO 3 (ICP-MS Stock Tuning Solution, Agilent Technologies).Torch position, sample depth adjustment and lens optimisation were set according to manufacturer recommendation while all other instrument parameters were optimised during a batch-specific user tune prior to each experimental run (Supplementary Table S1).Tissue homogenates were digested in 25 μL of concentrated (65 % (v/v)) HNO 3 that included heating at 90 • C for 20 min.Samples were allowed to cool then briefly vortexed followed by addition of 25 µL of 30 % hydrogen peroxide (Merck, Australia).Samples were heated at 70 • C to complete digestion.Samples were diluted to a final volume of 1 mL in Milli-Q H 2 O and centrifuged for 25 min at 20,000 × g to remove any insoluble material prior to analysis.The elements Mg-24, Ca-44, Fe-56, Cu-63 and Cu-66 were analysed in MS/MS mode using helium (He) collision gas, and Se was monitored at m/z 80 using a mixture of nitrous oxide and hydrogen (N 2 O + H 2 ) reaction gas.

Baited y-maze
Before testing in the baited y-maze paradigm began, Offspring (PD136-229) were food-restricted one week before testing in the baited y-maze paradigm began and maintained at 90 % free-feeding body weight.Mice were tested in the baited Y-maze daily without breaks until the protocol was completed.In the habituation phase, mice were arbitrarily placed in one of the three arms (the 'home arm'), which did not change for the duration of the protocol.Mice were freely allowed to explore the other two arms, both of which contained a food reward (Nippy's strawberry milk).Once the rewards in both arms were consumed, the mouse was gently led back into the home arm and was restricted to the home arm for approximately 10 s.Food rewards were then replenished, and the mouse was allowed to explore freely again.Mice were required to consume both food rewards at least 7 times within a 15-min period for 2 days in a row before progressing to the acquisition phase the following day, to ensure that they were actively seeking the food rewards.In the acquisition phase, one of the two non-home arms was designated the correct arm and always contained a food reward, while the remaining arm was designated the incorrect arm and never contained a food reward.The mouse was placed in the home arm and allowed to freely enter one of the non-home arms.Once the mouse entered one of the non-home arms, access to the non-chosen arm was removed, and the mouse was allowed to freely explore the chosen arm and consume the food reward (if present).The mouse was then gently led back into the home arm, and was restricted to the home arm for 10 s.The trial was repeated a total of 40 times in one day.The following day, the acquisition phase was repeated for up to 8 trials.If the mouse chose the correct arm for 5 out of 6 of its most recent trials, it was immediately tested in the reversal phase.In the reversal phase, the correct and incorrect arms were switched, and the mouse was allowed to freely choose between the two arms for an additional 50 trials in one day.If the mouse did not meet criteria for reversal, it was exposed to an additional 40 trials without switching of the correct and incorrect arms, and was tested again the following day.The experimenter stood behind the home arm for the duration of all tests.Fig. 5a shows a diagram of the task.

Computational modelling (reinforcement-learning)
Computational modelling (using four competing variants of a Qlearning model (Watkins and Dayan, 1992) was used to analyse performance in the baited Y-maze.The four models that we compared included (A) a model with a single learning rate and a single softmax inverse-temperature parameter, (B) a model with two learning rates (one for reward outcomes, one for non-reward outcomes) and one inverse-temperature parameter, (C) a model with two learning rates (one each for the acquisition and the reversal phase) and one inversetemperature parameter, and (D) a model with two learning rates and two softmax inverse-temperature parameters (one each for the acquisition and reversal phases).We then compared the goodness of fit of these four models to our data.These four models were chosen based on theoretical considerations pertaining to reversal learning tasks.These considerations included determining whether the learning rates of mice differed between reward and non-reward outcomes, as in humans (Palminteri et al., 2017), or whether mice differed in their learning rates and exploration after beliefs related to the reward outcome had been established.Models were fit using the probabilistic programming language Stan (Team, 2023), using 4 parallel chains of 1250 post-warmup samples each.Models were compared using the WAIC statistic as implemented in the loo package for R (Vehtari et al., 2017).Models A-C fit the data equally well (as indicated by a ΔWAIC value within one standard error of the numerically best-fitting model), and the simplest model (A) was therefore selected in the interests of parsimony.From this model, for each mouse we estimated a learning rate α (reflecting the learning rate over both the acquisition and reversal phase) and a softmax inverse temperature β (reflecting exploration, or the degree to which the choice with the highest perceived expected value is prioritised for choice) as the median of the animal-level posterior distribution for each parameter.

Open field test
The open field test (OFT) consisted of a clear polypropylene box (39 × 57 cm), with the centre defined as 1.36 cm from the outer perimeter.Offspring (PD76-149) were placed in the centre of the open field and allowed to move freely over a 20-min period.The movements of each mouse were recorded by an overhead camera (Panasonic WV-CP304) and autonomously tracked using TopScan version 2.0.The location of the mouse was taken as the centre of its silhouette and the accuracy of all traces was manually verified by the experimenter.The maze was thoroughly wiped down with 80 % ethanol followed by water between tests.

Sociability test
Mice (PD78-184) were habituated to an open chamber (39 × 57 cm), containing two empty wire cups for 10 min.The mouse was then briefly confined to the centre of the chamber, while a conspecific was placed in one wire cup and a plastic mouse (Lego) was placed in the second wire cup.The mouse was then allowed to move freely for 10 min.The mouse was confined to the centre of the chamber again and the Lego mouse was replaced with a novel conspecific, before the mouse was allowed to move freely for another 10 min.Each mouse was considered to be interacting with the contents of the wire cup when its nose was within 3.2 cm of the cup.The sociability preference of each offspring was calculated using the formula [time (interaction with conspecific) − time (interaction with Lego mouse)]/[(time (interaction with conspecific) + time (interaction with Lego mouse)].The social novelty preference was calculated using the formula [time (interaction with novel conspecific) − time (interaction with familiar conspecific)]/[time (interaction with novel conspecific) + time (interaction with familiar conspecific)].The movements of each offspring were recorded by an overhead camera (Panasonic WV-CP304) and autonomously tracked using TopScan version 2.00.The accuracy of all traces was manually verified by the experimenter, and the location of the object and conspecifics was counterbalanced between and within treatment groups.Fig. 6a and 6b show a diagram of the sociability and social novelty preference components of the test, respectively.

Statistical analyses
Maternal temperature and offspring weight (assessed with a A&D EJ-610 precision compact lab balance) were analysed by fitting a mixed model with Se, poly-I:C and sex as between-subject variables, and time as a within-subject variable.Sphericity was not assumed, and a Greenhouse-Geisser correction was applied for all analysis incorporating repeated measures.The number of surviving litters at weaning, litter size at weaning and maternal weight at GD17 were analysed using 2way ANOVA with Se and poly-I:C as between-subject variables.All cytokines and nutrients, as well as the results from the three-chamber sociability test and the OFT, were analysed using 3-way ANOVA, with sex, Se supplementation and poly-I:C as between-subject variables.When no significant effect or interaction with sex was found, sexes were collapsed and a 2-way ANOVA was performed.With respect to the acquisition and reversal phase of the baited y-maze, trials were grouped into non-intersecting bins of 10 trails each, and the percentage that the correct arm was chosen for each bin was calculated.The acquisition and reversal phases were then analysed using a 4-way ANOVA and by fitting a 4-way mixed model with sex, Se supplementation and poly-I:C as between-subject variables, and trial bin as a within-subject variable.Win-stay and lose-shift behaviour was analysed using 4-way repeated measures ANOVA, with sex, poly-I:C and Se as between-subject factors, and testing stage (acquisition or reversal) as a within-subject factor.A win-stay was defined as the percentage of times a mouse chose the correct arm after choosing correctly in the previous trial, while a loseshift was defined as the percentage a mouse changed its choice after choosing incorrectly in the previous trial.Between-group differences in win-stay, lose-shift, α learning parameters and β exploration parameters were analysed using a 3-way ANOVA, with sex, poly-I:C and Se as between-subject factors.Data was excluded from the analysis if it was identified as a statistical outlier [according to non-iterative Grubbs method for cytokine data (with small N), non-iterative ROUT method for ICP-MS and behavioural data (with a larger N)], if the mouse became ill and required humane euthanasia before the test was performed, if data was lost due to a technical fault, if logistical difficulties prevented the mouse from being tested or, in the case of the reversal phase of the baited y maze task, if the mouse failed to develop a preference for the baited arm during the acquisition phase.Post-hoc analyses were performed using Šídák's multiple comparisons test.The statistical analysis of maternal temperature and offspring weight was performed using SPSS Statistics 27, and other statistical analyses were performed using PRISM v9.1.0.ƒ.

Poly-I:C causes a maternal inflammatory response
Dams exposed to poly-I:C exhibited hypothermia (Fig. 2a, time × poly-I:C interaction, F (3, 74) = 3.977, p = 0.011).As there was no effect of Se, the Se and control groups were collapsed for the purposes of posthoc analysis, which showed that the temperature of Poly-I:C exposed mice differed significantly from controls at 3-and 6-hours post-injection (p = 0.04 and p = 0.029, respectively).Supplementation with Se did not affect maternal temperature, maternal weight at GD17 (Fig. 2b), or the average number of surviving offspring in litters that reached weaning (Fig. 2c).However, a trend towards a reduced number of offspring in surviving litters was observed for poly-I:C exposed mice, (Fig. 2c, F (1, 9) = 4.918, p = 0.054).Offspring weight increased with age (F (5, 230) = 132.076,p < 0.001, Fig. 2d-e), and was significantly higher in males compared to females (F (1, 46) = 126.415,p < 0.001), as expected.

Poly-I:C increases inflammatory cytokines in the placenta, which is recovered by Se
Poly-I:C and Se had no effect on placental IL-6 (Fig. 3a).However, 17 was observed in poly-I:C exposed offspring (main effect of poly-I:C (F (1, 37) = 2.954, p = 0.094, Fig. 3f) regardless of Se treatment or sex, with > 36 % mice exposed to poly-I:C displaying elevated IL-17.TNF-α and MIP-2 were both below the limits of detection in foetal brain samples for all treatment groups.In summary, we found that poly-I:C led to increased IL-1β in placenta regardless of sex and increased IL-17 in placentas associated with female foetuses, both of which were normalised by prophylactic Se treatment.A trend towards increased IL-17 in poly-I:C exposed foetal brains was also observed, which was not affected by Se treatment or sex.

MIA and maternal Se supplementation alter metal ion micronutrient levels in foetal brains
As there was no effect of sex or interactions of sex with poly-I:C or Se for any elements quantified by ICP-MS, sexes were combined for all analyses.Selenium supplementation significantly increased concentrations of 80 Se, 24 Mg, and 63 Cu in total foetal brain, however this increase was mitigated by poly-I:C (Fig. 4a-c; see Table 1 for detailed statistics). 56Fe was also elevated in Se exposed offspring, however, exposure to both Se and poly I:C reduced concentrations compared to either intervention alone (Fig. 4d). 44Ca was elevated in poly-I:C exposed offspring, and Se supplementation alleviated this (Fig. 4e).Selenium supplementation reduced foetal brain concentrations of 66 Zn independently of poly-I:C exposure (Fig. 4f).These findings indicate that Se supplementation altered uptake and/or storage of a broad range of metal ion micronutrients within total foetal brain tissue, which is generally inhibited by MIA.

Offspring exposed to MIA exhibit significant impairments in learning and cognitive flexibility, which is driven by reduced win-stay behaviour
As no significant effects of sex or interactions of sex with poly-I:C or Se were observed, sexes were combined for all analyses.In the acquisition phase, a main effect of trial-bin (F (2.8412, 168.7) = 16.24,p < 0.001) was found, indicating that the mice were successfully learning the location of the food reward (Fig. 5b).As no significant interactions between trial bin and Se or poly-I:C were found, all trial bins were collapsed and the percentage of correct responses across the entire task were analysed.A significant interaction between Se and poly-I:C (F (1, 60) = 4.668, p = 0.035) was identified.Post-hoc findings showed a significant difference between control and Se-exposed mice (p = 0.012), with offspring exposed to Se alone choosing the arm containing the food reward less than controls (Fig. 5c).Poly-I:C exposed offspring also showed a similar reduction in choosing the arm containing the food reward, however this did not reach significance (p = 0.054).A significant main effect of trial bin (F (3.61, 218.2) = 88.16) and a significant trial bin × Se × poly-I:C interaction was found during the reversal phase (F (4, 242) = 5.393, p = 0.004, Fig. 5d).This interaction was explored   Se + poly-I:C-exposed offspring showed no significant differences in performance when compared to controls across all trials (Fig. 5e-g).Impaired learning and reversal learning coincided with reduced win-stay (poly-I:C vs Se interaction, F (1, 61) = 5.101, p = 0.028; control vs poly-I:C p = 0.019; control vs Se p = 0.015; Fig. 5h), but not lose-shift behaviour (Fig. 5i).Increased win-stay, and reduced lose-shift behaviour was also observed in the reversal phase compared to the acquisition phase irrespective of treatment (main effect of phase, F(1, 62) = 9.073, p = 0.004).
To further probe the behavioural dynamics underlying impaired performance in both phases of the task, a Q-learning model was fit to the dataset.The computational modelling indicated that poly-I:C, regardless of the Se treatment, reduced learning rate across both the acquisition and reversal trials (F (1, 60) = 3.994, p = 0.05, Fig. 5j), while Se had no effect on the learning rate.In brief, reductions in the learning rate indicate that the poly-I:C-exposed mice place less emphasis on recent experiences when selecting future actions.This can delay flexible adjustments in behaviour when environments change but can be advantageous when the environment contains spurious feedback that should be ignored.A significant poly-I:C × Se interaction was found for the β exploration parameter (F (1, 54) = 14.84, p < 0.001), with posthoc tests indicating that mice exposed to either poly-I:C or Se exhibited reduced β-exploration parameters compared to controls (p = 0.002, p = 0.001 respectively, indicating increased levels of exploration in the treated groups, Fig. 5k).However, mice exposed to both poly-I:C and Se showed increased β exploration compared to poly-I:C alone (p < 0.01, Fig. 5k).The β exploration parameter describes the tendency of the mouse to select the behaviour it currently perceives to be most advantageous, as opposed to exploring alternative options that may ultimately be more beneficial.These findings suggest that prophylactic treatment with Se could be partially beneficial in normalising reinforcement learning in offspring exposed to MIA, but may alter reinforcement learning in offspring not exposed to MIA.

MIA causes sex-specific sociability deficits, which are partially recovered by Se
As a significant main effect of sex was observed in the sociability test (F (1, 68) = 5.283, p = 0.025), the data was split by sex.Female mice showed a significant poly-I:C × Se interaction (F (1, 37) = 6.443, p = 0.016), with poly-I:C exposed offspring showing reduced preference for the conspecific relative to the object compared to controls (p = 0.06), and poly-I:C + Se-exposed offspring (p = 0.02).No significant group differences were observed in male offspring (Fig. 6c).Data were also split by sex following a trend towards a three-way interaction (F (1, 68) = 3.734, p = 0.06) during the social novelty phase of the test (Fig. 6d).
Here, no significant differences in preference for the novel conspecific compared to the familiar conspecific were found in female offspring.However, in contrast to the sociability test, male offspring showed a significant main effect of poly-I:C (F (1, 31) = 5.579, p = 0.025), with poly-I:C-exposed offspring showing reduced preference for the novel mouse relative to the familiar mouse.There was no significant effect of Se and no poly-I:C × Se interaction, indicating that Se treatment did not recover this effect.
For locomotor activity, while no significant effect of sex was observed, a significant main effect of Se (F (1, 67) = 6.510, p = 0.013) and a poly-I:C × Se interaction (F (1, 67) = 4.674, p = 0.034) was found with respect to the average velocity of offspring during the OFT (Fig. 6e).Post-hoc analysis indicated that offspring exposed to both poly-I:C and Se moved slower than offspring exposed to poly-I:C (p = 0.004) or Se (p = 0.036) alone.Female offspring spent significantly less time in the centre of the open field compared to male offspring (F (1, 68) = 5.847, p = 0.018, Fig. 6f).However, there was no effect of poly-I:C or Se, indicating no group differences in anxiety levels.In summary, MIA caused female specific reductions in sociability, which were not present in Se exposed offspring, and a male specific deficit in social memory, which was not recovered by Se supplement.Further, while no differences in anxiety-like behaviour were observed in the OFT, both sexes exhibited reduced locomotion when prenatally exposed to both MIA and Se.

Discussion
Infections during pregnancy are a risk factor for neurodevelopmental conditions such as ASD and schizophrenia, which is thought to be driven by MIA (Patterson, 2011;Sørensen et al., 2009).Here, we investigated Se as a prophylactic pregnancy supplement to protect against MIAinduced behavioural deficits, investigated the safety of sodium selenate supplementation in pregnancy, and assessed the levels of elemental micronutrients within foetal brains exposed to poly-I:C and/or Se supplementation.

MIA increased pro-inflammatory cytokines in the placenta, which was alleviated by Se
Reduced maternal temperature and litter size are strong indicators of a MIA response (Mueller, 2019), and a trend towards reduced litter size was observed in this study.Despite the increase in pro-inflammatory cytokine IL-1β in the placenta, the lack of elevated IL-1β, IL-6, IL-10 and TNF-α in MIA-exposed foetal brains is consistent with the findings of a recent systematic review and meta-analysis, which indicated that MIA is associated with elevated IL-6 and TNF-α in the blood and brain of offspring exposed at mid, but not late, gestation (e.g.GD17), while MIAinduced elevations in IL-10 and IL-1β in foetal brains did not generally occur (Hameete, 2021).One possible mechanism underlying this difference is the reduced placental permeability to IL-6 from mid to late gestation (Dahlgren et al., 2006), which may also occur for TNF-α.Interestingly, we identified a female-specific increase in placental IL-17 and a trend towards increased IL-17 within foetal brains, which has been implicated in mediating MIA-induced neurobiological and behavioural differences reminiscent of ASD (Choi, 2016).To the best of our knowledge, three studies have reported no differences in IL-17 (Mueller, 2019;Arrode-Brusés and Brusés, 2012;Arsenault et al., 2014), while another reported substantial increases that did not reach statistical significance (Ehninger, 2014).Prophylactic treatment with Se partially mitigated the inflammatory effects of MIA by reducing placental IL-1β and IL-17 but had no impact on IL-17 within the foetal brain.These observations corroborate previous findings on the effects of Se treatment during pregnancy, which have reported reduced inflammatory markers, such as IL-1β and IL-6, in sows (Mou, 2020).Further, in vitro and ex vivo studies have indicated that Se downregulates NF-κB signalling (Gandhi, 2011), possibly via various prostaglandins that are mediated by the action of selenoproteins (Gandhi, 2011).

MIA and Se during pregnancy altered metal ion micronutrients within the foetal brain
Redox-active metal ions, such as Fe and Cu, and as well as other elements, such as Se, Cu and Ca, are essential for a numerous cellular processes and pathways involved in the regulation of antioxidant systems and DNA repair within post-mitotic neurons (Wandt, 2021).Prior research has shown that elemental micronutrient dyshomeostasis is strongly associated with impaired neuronal development and function (Wandt, 2021;Chin-Chan, 2022;Cardoso et al., 2015), however, this has not been investigated in the context of MIA or Se supplementation.Our study demonstrates that maternal supplementation with Se leads to elevated Se in whole foetal brain homogenate, consistent with previous studies that reported increases in multiple foetal tissues (Sakamoto, 2013;Abdelrahman and Kincaid, 1995;Hawkes, 1994).As Se has been shown to readily cross the placenta (Chen, 2014), this increase is unsurprising.Selenium supplementation also increased Mg, Cu, and Fe in foetal brain tissue, which was mitigated by MIA.However, an important caveat is that this work was restricted to analyses of whole foetal brain homogenate.As such, differences in the cellular and spatial distribution of elements within the foetal brain in response to Se supplementation and/or MIA exposure could not be interrogated.Nevertheless, our approach opens the way to develop novel hypotheses for how MIA and/ or Se may influence elemental homeostasis in the brain.
The increase in Fe following Se treatment was consistent with prior studies of BALB/c mice fed on either Se-deficient (Sharma et al., 2019) or Se-supplemented diets (Sharma et al., 2019).Selenium and Fe are both involved in the regulation of ferroptosis, a programmed nonapoptotic cell-death pathway that is dependent on Fe and triggered by the accumulation of lipid peroxides.The selenoprotein glutathione peroxidase 4 is a key regulator of ferroptosis (Cardoso et al., 2017), and previous research has demonstrated the positive effects of Se treatment in inhibiting the ferroptotic cell death pathway (Ingold, 2018).Although the molecular mechanisms underpinning this Se mediated increase in Fe warrants further investigation, this likely has direct implications for brain development.Selenium supplementation also decreased Zn levels independently of MIA, aligning with previous research showing that cosupplementation of Se and Zn lowered Zn bioavailability in rats (Daragó et al., 2016).It has been proposed that Zn works in balance with Cu and Mg levels in neurons.As such, the observed Se-induced increases in foetal brain Cu and Mg may reflect an indirect response to the reduction in Zn and MIA-mediated dyshomeostasis.However, further studies are required to understand the precise molecular basis for how this occurs.Irrespective, Fe and Cu are known to play essential regulatory roles in neuronal development by regulating the redox system, acting as essential metal cofactors in the DNA base excision repair pathway, and contributing to neurotransmitter synthesis (Wandt, 2021).Further, the cellular pathways of Mg and Ca homeostasis are also known to interact and regulate cell growth, differentiation, and death (Staneviciene, 2022).While these elements serve essential roles in neuronal development and function (Wandt, 2021;Staneviciene, 2022), imbalances in their concentrations may lead to deleterious neuronal impacts that manifest as behavioural and neurobiological alterations in MIA exposed offspring.Therefore, appropriate regulation of elemental homeostasis within the foetal brain is fundamental for neuronal development.
An increase in Ca has been reported in the tissues and serum of mice exposed to viral infections (Ilbäck et al., 2003) and in humans exposed to bacterial infections (Aslan, 2023).Notably, Ca mediates a proinflammatory response via the Ca-sensing receptor (Sundararaman and van der Vorst, 2021).Pro-inflammatory cytokines promote the release of Ca into the bloodstream, and high levels of extracellular Ca are found at sites of inflammation in response to infections.This in turn stimulates the Ca-sensing receptor to induce activation of the NLRP3 inflammasome and NF-κB pathway, promoting further release of proinflammatory cytokines, with implications in various inflammatory diseases including Alzheimer's disease and COVID-19 (Iamartino and Brandi, 2022).As such, MIA may drive the release of Ca into the serum of dams, which then ultimately accumulates in the foetal brains of offspring.This may have been alleviated in Se treated offspring via selenoprotein-mediated downregulation of pro-inflammatory pathways.However, it should be noted that our study incorporated a limited number of litters per treatment group, with only two litters contributing to foetal brains and placentas exposed to both MIA and Se.Confirming these findings through additional studies that incorporate a larger number of litters may therefore be warranted.

MIA and Se during pregnancy cause deficits to reinforcement learning in offspring
Our findings in the baited-y maze task indicate that MIA led to reduced performance during the reversal phase, and a trend towards reduced performance during the learning phase.This is partially consistent with previous literature, which has generally reported deficits in the reversal but not learning phase of positive reinforcement learning tasks (Schroeder, 2019;Nakamura et al., 2022).These defects were underpinned by reduced win-stay behaviour, reminiscent of individuals with ASD (Solomon et al., 2011;Solomon, 2015) and schizophrenia (Reddy et al., 2016;Culbreth et al., 2016;Waltz, 2013;Waltz et al., 2007) in probabilistic learning tasks.As a form of positive reinforcement, win-stay behaviour is associated with phasic increases in dopaminergic signalling within the midbrain, which influences striatal and prefrontocortical circuitry to guide learning and goal-directed behaviour.People with ASD and schizophrenia exhibit disruptions to this reward circuit (Pavǎl and Miclutia, 2021;Deserno et al., 2016), which has been suggested as a potential explanation for reduced win-stay behaviour in reinforcement learning tasks (Solomon et al., 2011;Saperia, 2019).It is possible that MIA exposed offspring exhibited similar changes in this reward circuit, however this will need to be confirmed by future studies.As such, our findings mirror this behavioural phenotype observed in both conditions.While previous studies investigating reinforcement learning in the MIA model have typically restricted their analysis to overall task performance, we also used computational modelling to formally investigate the behavioural dynamics that may underpin differences in performance.We report that, while MIAexposed offspring exhibited a trend towards a subtle reduction in learning rate, reduced performance was primarily driven by reductions in the inverse temperature parameter β.This factor reflects the extent to which mice prioritise spreading their choices among alternatives regardless of their beliefs of the underlying value of each choice, which allows them to gather additional information about their environment to guide future decisions (i.e., exploration) (Wilson et al., 2021).Reductions in the inverse temperature parameter reflect an increase in exploratory behaviour and may reflect neurobiological differences in MIA-exposed offspring, such as increased tonic dopaminergic signalling within the midbrain (Beeler et al., 2010;Aguilar-Valles et al., 2020).Alternative explanations include the impaired computation of value representations in the prefrontal cortex (Kolling et al., 2016), which may arise due to altered GABAergic signalling that has been widely reported in the MIA model (Canetta, 2016), or altered noradrenergic signalling in the locus coeruleus, which has received little study in the MIA model (Csatlosova, 2021).These possibilities are not mutually exclusive and represent promising avenues for future research.Our findings of increased exploratory behaviour are also reflected in adults with ASD and schizophrenia in reinforcement learning tasks (Yechiam et al., 2010;Speers and Bilkey, 2023), providing further support for the translational validity of the MIA model.Surprisingly, we also observed reduced task performance, reduced win-stay behaviour and increased stochastic exploration in mice exposed to Se alone.These findings are crucial given the contemporary interest in Se as a supplement for multiple conditions (Torres et al., 2021;Cardoso, 2019;Duffield-Lillico, 2002).The negative effects of Se have been attributed in part to the hermetic relationship between Se and redox status.Increased Se intake is associated with the upregulation of multiple selenoproteins and reduced markers of oxidative stress (Steinbrenner and Sies, 2009).However, excessive Se is metabolised into a series of highly redox-reactive compounds once selenoprotein synthesis is saturated (Misra et al., 2015).While we did not observe a reduction in offspring weight or weight gain during pregnancy that is characteristic of severe Se toxicity, it is possible that the dose used in our study induced moderate levels of oxidative stress.Dopaminergic and GABAergic interneurons are particularly susceptible to increased oxidative stress (Scheuer, 2021;Gatzke-Kopp, 2011), and have been found to play critical roles in reinforcement learning (Beeler et al., 2010;Kolling et al., 2016).Differing rodent strains between laboratories may possess varying capacities for selenoprotein induction and thus may exhibit Se-mediated increases in reactive oxygen species at differing doses.Regardless of cause, our study indicates that additional caution for the use of Se during pregnancy and in women of childbearing age may be warranted.This is especially true given that translational studies reporting the benefits of Se in other conditions used doses higher than that used in our study (Corcoran et al., 2004).
Given the increased exploration observed in offspring exposed to MIA or Se alone, it was also surprising to observe that offspring exposed to both exhibited comparable exploratory behaviour to controls.One explanation is that exposure to both led to comparable dysfunctions in striatal dopamine release, but also led to the loss of dopaminergic neurons by further elevating oxidative stress.As such, normative concentrations of tonic striatal dopamine may have been maintained through a smaller population of hyperactive dopaminergic neurons in offspring exposed to Se and MIA.This loss of dopaminergic interneurons would also be expected to impair phasic dopamine release and lead to learning impairments (Popescu et al., 2016).Indeed, a reduced learning rate was observed in MIA-exposed offspring, particularly in offspring exposed to both MIA and Se.This theory is also consistent with our findings of reduced locomotion in the OFT (Ryczko and Dubuc, 2017).While it must be noted that Se prevented MIA-induced increases in IL-1β and IL-17, MIA reduced concentrations of Mg, which promotes the antioxidant activity of superoxide dismutase (Shafeeq and Mahboob, 2020).Thus, exposure to both MIA and Se may lead to increased oxidative stress and/ or reduced antioxidant capacity within foetal brains.

MIA and Se during pregnancy cause sex-specific alterations in sociability and social memory in adult offspring
Our study also identified sex-specific effects of MIA on social behaviour, with females exhibiting a strong trend towards reduced sociability, and males exhibiting a reduced preference for a novel conspecific.Given that elevated IL-17 is necessary for MIA-mediated deficits in sociability (Choi, 2016), this aligns with our observation of a female-specific increase in placental IL-17 and more pronounced elevations of IL-17 within female foetal brains.While the importance of investigating both sexes in animal research is increasingly recognised, sex differences in the MIA model remain understudied (Coiro and Pollak, 2019).To our knowledge, only four studies have investigated sex differences in sociability following MIA in late gestation.Those studies have reported no effects of MIA (Aavani et al., 2015;Batinić, 2016), deficits in both sexes (Bitanihirwe et al., 2010), or female specific deficits (Fernández de Cossío et al., 2017).Similarly, while increased asocial behaviour has been generally accepted in ASD (Chevallier et al., 2012) and schizophrenia (Fulford et al., 2018), studies investigating sex differences in asocial tendencies remain scarce.Thus, our data supports the further investigation of sex differences in sociability for the MIA model and in both conditions.
Selenium successfully recovered poly-I:C-mediated increases in both placental IL-17 and IL-1β, and the emergence of sociability deficits.As mice treated with Se alone also exhibited no differences in sociability, deficits in this domain may have occurred due to pathogenic mechanisms others than oxidative stress, such as IL-17-mediated microglial priming (Otero and Antonson, 2022).With respect to deficits in social memory, studies investigating the impact of MIA in late gestation are inconsistent, with some studies reporting deficits and others reporting no differences (Okamoto et al., 2018;Meyer, 2006;Fernández de Cossío et al., 2017).Again, these inconsistencies may be due to the variable elevation of IL-17 in MIA exposed offspring.While deficits in social memory have been reported in mixed sex cohorts of patients with ASD (Williams et al., 2005;Griffin et al., 2021) or schizophrenia (Philippe-Olivier and Martin, 2014;Pelletier et al., 2005), sex differences in social memory have also not been comprehensively explored in either condition, and we would encourage considering potential sex effects in future research.One limitation of our study is that we did not assess potential changes in olfaction, which plays an important role in mediating sociability in mice (Ryan et al., 2008).However, while previous studies have indicated that MIA induces neurobiological changes in the olfactory bulb (Liu et al., 2013), performance in olfactory tests appears generally unchanged in MIA exposed offspring (Braun, et al., 2019;Gzieło et al., 2023;Malkova et al., 2012).Furthermore, while the three -chamber test remains one of the most common paradigms to assess social behaviour in rodents, it possesses several limitations (Jabarin et al., 2022), and the translatability of social memory deficits in rodents to human populations is unclear.MIA-exposed offspring did not exhibit increased anxiety-like behaviour, as measured through time spent in the centre of the OFT.This is consistent with previously reported findings, which have generally implicated MIA in early, but not late gestation with increased anxiety-like behaviour (Depino, 2015).

Conclusions
In mice, MIA induced autism and schizophrenia-like behaviours in a sex-specific manner.MIA led to increased pro-inflammatory cytokines in placentas, particularly in females, which was prevented by Se.Se also led to increased levels of Se, Mg, Fe and Cu in whole feotal brain homogenate, and mitigated MIA-induced elevations in Ca.MIA reduced sociability in female offspring, which was recovered by Se, and reduced social memory in male offspring, which was not recovered by Se.Exposure to Se or MIA alone led to impaired performance in reinforcement learning task.Computational modelling indicated that this was chiefly due to increased exploratory behaviour, rather than alterations in the rate of learning.In summary, while Se may be beneficial in ameliorating sociability deficits associated with MIA, it may have negative effects in other behavioural domains.Further research into the relationship between maternal Se supplementation and neurodevelopmental disorders is needed and caution in the use of Se supplementation during pregnancy and in women of childbearing age that are Se-replete may be warranted.
Funding Sources: This work was supported by the 2020 C. Andrew Ramsden Monash Health Translational Precinct Early Career Researcher Collaborative Award, awarded to Barbara R Cardoso and Anna Schroeder as well as a National Health and Medical Research Council (NHMRC) ideas grant (GNT2000893) awarded to Rachel Hill.Brendan Gillespie is supported by a Research Training Program Scholarship, Monash University.The funding sources were not involved in the study design, the collection, analysis and interpretation of data, in the writing of the report, or the decision to submit the article for publication.

Fig. 1 .
Fig. 1.Study design.GD = gestational day; PD = postnatal day.Selenium (Se) supplementation via 0.15 mg/100 mL sodium selenate in drinking water.Poly-I:C = polyinosinic:polycytidylic acid, delivered via a single intraperitoneal injection at a dose of 20 mg/kg, or an equivalent volume of 0.9 % saline (w/v) control, with body temperature of dams monitored following injection.Two litters per group were culled to investigate changes in foetal brains and placenta.Remaining litters (3 control, 4 poly-I:C, 3 Se, 2 Se + poly-I:C) were tested for all behavioural measures.ICP-MS = inductively coupled plasma mass spectrometry, used to assess changes in metallic micronutrients within foetal brains.Luminex & ELISA assays were used to measure cytokines in placentas and foetal brains.

B
.Gillespie et al.

Fig. 5 .
Fig. 5. Learning and cognitive flexibility differences in adult offspring prenatally exposed to MIA and/ or selenium (Se) supplementation.a. Experimental design of the baited y-maze task; b.Performance during the acquisition phase as a function of trial bins; c.Performance during the entire acquisition phase of the baited-y maze task.d.Performance during the reversal phase as a function of trial bins; e-g: First ten, penultimate ten and final ten reversal trials (n = 10-22/group); h-j: Percentage of win-stay and lose-shift behaviour during the baited y-maze task (12-22 /group); j.Learning rates during the baited y maze task (12-20/group); k.Exploration parameter during the baited y maze task (12-20/group).Data is expressed as mean ± SEM. # p < 0.1, * p < 0.05, ** p < 0.01.

Fig. 6 .
Fig. 6.The performance of offspring exposed to maternal inflammation and activation and selenium (Se) in the sociability and open field tasks.a-b: Experimental design of sociability (a) and social memory phases (b) of the three-chamber sociability test.Pink lines represent the location of a representative control mouse for the entire analysed testing period.In this example, the left chamber contains a Lego mouse during the sociability phase, and a novel conspecific during the social memory phase.c.Preference of offspring to spend time investigating a conspecific relative to an object during the first 2 min of the sociability test (n = 5-12/group); d.Preference of offspring to spend time investigating a novel conspecific relative to a familiar conspecific during the first 2 min of the sociability test (n = 5-12/group); e. Velocity in the OFT.(n = 13-23 /group); f.Percentage time spent in the centre in the OFT (5-12/group).AU = arbitrary units.Data is expressed as mean ± SEM. # p < 0.01, *p < 0.05.(For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Table 1
Statistical analysis of ICP-MS data.
a SC = saline control, Se = selenium, PC = poly-I:C control, SS = saline + Se, PS = poly-I:C + Se.Main effects and interactions in bold indicate statistical significance.