Knockdown of microglial iron import gene, Slc11a2, worsens cognitive function and alters microglial transcriptional landscape in a sex-specific manner in the APP/PS1 model of Alzheimer’s disease

All mouse breeding, maintenance, and procedures were approved in advance and conducted in compliance with the Institutional Animal Care and Use Committee at Vanderbilt University. For the primary cell experiments from young and aged mice, young 9-week-old control C57BL/6 J male mice were purchased from Jackson Laboratories (Bar Harbor, ME, USA) (#000664, JAX). For aged mice, young C57BL/6J male mice were originally purchased from Jackson Laboratories and were aged and maintained in the Vanderbilt mouse facility until they were 27–30 months old. To determine the effect of decreased microglial DMT1 on disease, we generated a novel transgenic mouse model with inducible knockdown of Slc11a2 in microglial cells in the APP/PS1 model of AD. Cx3cr1^Cre−ERT2 mice (B6.129P2(C)-Cx3cr1^{tm2.1(cre/ERT2)Jung}/J; #020940) purchased from Jackson Laboratories (JAX, Bar Harbor, ME, USA) express a tamoxifen-inducible Cre-recombinase driven by the promoter for the microglial/macrophage Cx3cr1 chemokine receptor gene, allowing for conditional knockdown of loxP-containing genes in Cx3cr1-expressing cells [46]. Slc11a2- ‘floxed’ mice (129S-Slc11a2^tm2Nca/J; #017789, JAX) [47] were bred with Cx3cr1^Cre−ERT2 homozygous mice to obtain Slc11a2^flfl;Cx3cr1^Cre++ homozygous animals. APP/PS1⁺ hemizygous animals were purchased from JAX and maintained in our facility (Tg(APPswe,PSEN1dE9)85Dbo; MMRRC_034832-JAX). These transgenic animals express a chimeric mouse/human amyloid precursor protein (Mo/HuAPP695swe) and a mutant presenilin-1 (PS1-dE9), and have been widely used in AD research, particularly in relation to amyloid-β associated pathology [48–50]. We chose this model of AD based on the well-characterized development of disease-associated symptoms (i.e., amyloid deposition, cognitive deficits) and the progressive nature of disease development over the course of several months. This slower onset compared to other models allows us to examine the early pathological changes that occur prior to the onset of symptoms later in the course of disease. Additionally, the APP/PS1 model has already been shown to exhibit significant microglial iron loading [21, 32], and an amyloid-driven model is relevant based on associations between iron and Aβ in the brain [15, 51]. APP/PS1⁺ hemizygous animals were bred separately with Slc11a2^flfl animals to yield Slc11a2^flfl;APP/PS1⁺ mice. Resulting progeny from these crosses were then bred with Slc11a2^flfl;Cx3cr1^Cre−ERT2++ animals to yield triple-transgenic Slc11a2^flfl;Cx3cr1^Cre−ERT2±;APP/PS1⁺ or APP/PS1⁻ (i.e., ‘WT’) mice (Additional File 1A). All mice used in experiments were Slc11a2^flfl;Cx3cr1^Cre−ERT2± and either APP/PS1⁺ hemizygotes or WT as littermate controls. Experimental mice were on a mixed 129S/BL6 background, with > 80% BL/6J genetic makeup as confirmed via Transnetyx strain analysis (Cordova, TN). All genotypes were confirmed with an ear snip via Transnetyx using real-time PCR. Mice were weaned at 3 weeks of age and had ad libitum access to food (LabDiets, standard rodent chow 5001, 240 ppm iron) and water. Both male and female mice were used in our experiments and were group-housed (2–5 per cage) by sex in transparent cages at 22–25 °C under a 12 h light/dark cycle in a specific pathogen-free facility. Control and experimental animals were randomly assigned across cages.

Tamoxifen (Sigma #T5648) was dissolved in corn oil (Sigma #C8267-2.5L, lot #MKCK6411, Saint Louis, MO) to generate a 20 mg/mL stock concentration by sonicating the mixture and stirring overnight in a glass vial at 37 °C. Slc11a2^flfl;Cx3cr1^Cre−ERT2±;APP/PS1^{+ or –} male and female mice at 5–6 months of age were administered a dose of 4 mg (maximum 200 μL volume) tamoxifen via oral gavage every day for five consecutive days [42, 52] (Additional File 1B). All mice that received tamoxifen are denoted as ‘Slc11a2^KD’, and are either APP/PS1 or WT. Littermate mice with the same genotypes (Slc11a2^flfl;Cx3cr1^Cre−ERT2±;APP/PS1^{+ or –}) were administered gavage with corn oil as a control for the presence of Cre based on work showing effects of Cx3cr1^Cre−ERT2 genotype alone on microglial function [53, 54]. Corn-oil-treated animals are denoted as ‘Control,’ and are either APP/PS1⁺or WT. The numbers of experimental animals used are shown in Supplemental Table 1 of Additional File 2. We chose to induce knockdown of Slc11a2 between 5–6 months of age in these mice, as it is a relatively early timepoint in this AD model when Aβ plaque deposition becomes visible and allowed us to assess the effect of early changes in microglial Slc11a2 on downstream disease development. Knockdown of Slc11a2 was confirmed in isolated microglia from all animals via RT-qPCR utilizing a primer targeting Slc11a2 exons 6–8 (Additional File 1C).

All behavioral assays were conducted in the Vanderbilt Murine Neurobehavioral Core after mice were acclimated to the facility for at least one to 2 weeks. All mice underwent testing between 12 and 15 months of age (mouse numbers and body weights shown in Additional File 2, Supplemental Table 1 and Supplemental Table 2). Control WT and APP/PS1 mice were randomly distributed across cages and litters, and the order of mice run through each assay was also randomly assigned. To avoid experimenter bias, mice of both sexes and all four experimental groups were evenly and randomly split between two different experimenters of opposite sex, who were blinded to the genotype and treatment of the mice before testing. The running order of assays was kept consistent for all animals in each study, and animals were run at the same time each day between 0630 and 1300 h with one task per day. For each task, mice were acclimated to the testing room for 30 min to 1 h prior to testing, and control and experimental groups were evenly and randomly distributed across cages, days, and time of each assay. Following completion of a trial, each apparatus was cleaned of feces, disinfected, and deodorized with an anti-bacterial spray (Peroxigard, Virox Technologies) in between animals. APP/PS1⁺ mice are known to be prone to spontaneous seizures [55] and any mouse that exhibited a seizure during an assay was excluded from that analysis (n = 3 male Slc11a2^KDAPP/PS1⁺). Additionally, two mice died spontaneously prior to completion of all tasks and euthanasia (one male Slc11a2^KDAPP/PS1 and one female Control APP/PS1). The data for these animals is recorded for the tasks completed prior to death.

As a measurement of general cognition and well-being, an overnight nest building assay was used. Nest building assessments were performed as the first behavioral task to minimize effects of stress on the mice from other behavioral assays. Mice were single-housed and given 5 g of cotton nestlet (Ancare, Bellmore, NY) in the afternoon the day prior. The next morning, amount shredded and quality of nests was scored by a blinded observer using a 0–5 scale adapted from previous work, in 0.5 increments [42, 56, 57]. Following nest building assessment, mice were re-housed in groups of 4–5 before all other behavioral tasks.

Several assays were used as control measures of anxiety and for locomotor activity assessment. An elevated zero maze (white maze, width 5 cm; diameter 50 cm; wall height 15 cm, Stoelting Co. IL) was used first, where mice underwent a single 5 min trial of free exploration. Mice were video-recorded using a ceiling-mounted camera and movement was automatically tracked and scored using AnyMaze (Stoelting Co., Wood Dale, IL). Analysis parameters were set to ensure 80% of the mouse needed to be present in either the ‘open’ or ‘closed’ zone for an entry into that zone to be recorded. Total time in the open and closed zones and total distance traveled were measured. Sound-attenuating transparent open field chambers (27.5 × 27.5 cm) were used for a second measurement of baseline locomotor activity. Mice were placed in the center of the chamber and allowed to explore freely for 45 min. Distance traveled was recorded automatically via the breaking of infrared beams (MedAssociates ENV-510 software, Fairfax, VT). Additionally, time spent in the center area (19.05 × 19.05 cm) versus time in the ‘surround’ was calculated as a control measure of anxiety-like behavior.

A single-trial Y-maze was used as another measurement of baseline locomotor and exploratory behavior, as well as an assay to measure short-term working memory function. A clear plexiglass three-arm Y-maze (each arm 5 cm in width, 34.5 cm long) with differentiated arms (different colors of paper with or without patterns placed underneath the maze) was used. All mice were placed in the same point of the same arm and allowed to freely explore for 6 min. A ceiling-mounted camera recorded video of the mice and AnyMaze automatically measured total distance traveled and order of arm entries. Entry into another arm was predicated on having at least 80% of the mouse cross into at least 1 cm of the arm. Spontaneous alternation as a measure of intact working memory was calculated by hand using arm entry order data from AnyMaze. A ‘correct’ alternation is defined by three consecutive entries into three different arms (e.g., ABC, BCA, CAB). Percent alternation was calculated using: ((Number of spontaneous alternations) / (Number of total arm entries—2)) * 100.

Mice underwent testing in the Morris water maze (MWM) to assess the effect of Slc11a2 knockdown on learning and memory [58]. Briefly, a circular pool approximately 1 m in diameter filled approximately 30 cm deep with 22–27 °C water was used for this task. A white round platform (10 cm in diameter) was used to provide animals an escape from the water. Mice first underwent two visual training days, where the platform jutted above the water with a pole attached to allow mice to see the target platform. This platform was moved around to each of the four quadrants on each session during training days to allow the opportunity for each animal to swim and survey the room, which contained multiple visual spatial cues kept constant throughout. Each training day comprised four trials per mouse, and each mouse was given 60 s to find the platform. If a mouse did not reach the platform in 60 s, it was guided to and placed on the platform for at least 5 s. On subsequent days following the two visual training days, the water was made opaque with non-toxic tempura white paint, and the platform was submerged approximately 0.5 cm under the water. The platform was kept in the same location for each trial and day, and mice were randomly placed in different locations in the pool so that the use of spatial cues for navigation was necessitated. Mice underwent four trials per day for 5 days, with each trial lasting 60 s to assess learning and short-term memory. If mice did not find the platform within 60 s, they were guided to the platform and escape latency was recorded as 60 s. Following the final day of testing, the platform was removed and mice were allowed to swim freely for 60 s. Total time spent in the target quadrant where the platform used to be, time spent around the location of the platform, swim speed, total distance traveled, and time spent in perimeter were recorded as measurements of platform location memory.

Following completion of all other behavioral tasks, a fear conditioning assay was conducted to assess differences in fear-associated memory. Mice were placed in sound-attenuating chambers with a wire grid floor. On the 1st day (training trial), mice were placed in the chambers for 8 min and allowed to explore freely. Every 2 min, a 30 s tone was played, followed immediately by a small shock administered through the wire floor (1 s, 0.5 mA). This tone-shock pairing occurred 3 times during the training trial. To assess contextual fear conditioning, mice were placed back into the same chamber the next day and allowed to run around freely for 4 min with no tone or shock presented. Total time freezing—indicative of fear memory—was recorded automatically (VideoFreeze, MedAssociates). To assess cued fear conditioning (memory of the tone), mice underwent a second testing trial. This trial included a different experimenter handling the mice, significant alterations to the chamber with white walls, white floor inserts, and red light, and the scent of vanilla placed in an open tube outside the chamber. Mice freely explored the chamber for 2 min before the tone was administered for the final 2 min (without a shock pairing). Total time spent freezing during the no-tone and tone segments were recorded as a measurement of cued fear memory.

At the time of euthanasia, mice were deeply anesthetized with isoflurane and 500–700 μL of blood was collected via cardiac puncture. Immediately following blood collection, mice were trans-cardially perfused with 20 mL of cold 1 × Dulbecco’s phosphate-buffered saline (DPBS) to remove circulating blood and decapitated for rapid brain removal. Whole brains were either placed on ice for mincing and processing for cellular isolation, or bilateral hippocampus was isolated first before proceeding to cellular isolation.

Brains were rapidly removed and briefly placed in 3 mL cold, sterile 1 × Hank’s buffered saline solution (HBSS, Gibco, #14175095) containing 1% fetal bovine serum (FBS, heat-inactivated; Gibco, #10082147) to remove any residual blood. Microglia isolation was performed following published protocols, with slight modifications [59–61]. Briefly, whole brains were transferred and finely minced with scissors in cold, sterile “IMG media” [Dulbecco’s modified Eagle’s medium (DMEM) with high glucose (4.5 g/L) and l-glutamine media (Gibco, #11965092) containing 10% FBS and 1% penicillin–streptomycin (Gibco, #15140122). Minced tissue was transferred into 50 mL conical tubes and 5 mL of digestion media (IMG media + 100 units Papain, #LK003176↗; 500 Kunitz units DNase, #LK003170↗, Worthington Biochemicals, Lakewood, NJ) was added to each tube. Whole-brain samples were enzymatically-dissociated by placing in an orbital shaker for 1 h at 37 °C, diluted with 10 mL IMG media, and strained through 70 μm sterile filters (Corning, #431751). Bilateral hippocampus samples in the RNA-sequencing studies were treated in the same manner described above, with slight modification. Bilateral hippocampus samples were isolated and immediately placed in 15 mL conical tubes on ice and 2.5 mL digestion media was added. Samples were incubated for 30 min in an orbital shaker at 37 °C. Every 10 min, samples were triturated up and down with a serological pipette of decreasing size before straining samples through filters and proceeding with subsequent steps. All samples were further processed at 4 °C unless otherwise indicated.

Cells were centrifuged for 5 min at 500 × g, re-suspended in a solution of 30% isotonic Percoll and IMG media (Cytiva, #17-0891-01), and slowly layered onto a 70% Percoll gradient with HBSS + 1% FBS. HBSS + 1% FBS (without Percoll) was layered on top, and samples were centrifuged for 15 min at room temperature at 600 × g with the brake set to the lowest setting to allow for density separation. The supernatant containing myelin and neuronal debris was removed, and cells at the interface between the 30–70% gradients were carefully collected and placed on ice into 8 mL HBSS + 1% FBS in a fresh tube to wash residual Percoll. Cells were centrifuged at 500 × g for 5 min at 4 °C, and pelleted cells were re-suspended in appropriate media for downstream assays.

For experiments conducted in isolated primary cells from young and aged mice, all steps above were performed under sterile conditions in a cell culture hood with autoclaved tools and sterile-filtered reagents. Mixed glial cells isolated and pelleted from the Percoll gradient were re-suspended in 1 mL IMG media for counting and plating. Cells were counted using the Nexcelom Cellometer Auto T4 Cell Counter (Nexcelom Biosciences) and plated at a density of 100,000 cells per well in poly-L-lysine-coated 48-well plates in pre-warmed sterile IMG media containing 5 ng/mL GM-CSF (R&D Systems, #415-ML-010). Media was changed the next day, and then every other day for 5 days before stimulation with Aβ as described below.

For experiments analyzing gene expression (RNA-sequencing and RT-qPCR) in microglia from the triple-transgenic animals, Percoll-isolated glial samples were further processed for enrichment of CD11b⁺ microglia. Following Percoll gradient separation, centrifugation, and pelleting, cells were re-suspended in 400 μL cold “MACS” buffer (1 × PBS containing 0.5% FBS and 2 mM EDTA) and transferred to 5 mL tubes. Cells were centrifuged at 4 °C for 5 min at 500 × g, pelleted, and re-suspended in 90 μL MACS buffer for magnetic labeling and separation according to manufacturer’s instructions (Miltenyi Biotec, Bergisch Gladbach, Germany). Briefly, samples were incubated with magnetic anti-CD11b MicroBeads (Miltenyi Biotec, #130-093-634; 10 μL per 90 μL buffer/brain) for 15 min at 4 °C. Magnetic separation was performed utilizing MS columns, and CD11b⁺ cells and the effluent non-magnetic fractions (CD11b⁻ cells) were obtained. Following a final centrifugation for 5 min at 500 × g, cells were immediately re-suspended in RLT lysis buffer (Qiagen, #74004) supplemented with 1% beta-mercaptoethanol, briefly vortexed, and flash-frozen in liquid nitrogen. Samples were stored at –80 °C until RNA isolation.

The immortalized microglial cell line, “IMG” [62], was used for in vitro experiments to assess the direct effect of pharmacological inhibition of DMT1 on Aβ-induced inflammation. IMG cells were purchased from Millipore (Cat. #SCC134, RRID:CVCL_HC49), and cultured as described using Accutase for dissociation and passaging [45]. Briefly, cells were cultured up to a maximum of 10 passages in sterile Dulbecco’s modified Eagle’s medium (DMEM) with high glucose (4.5 g/L) + 2.5 mM glutamine (Gibco, #11965092) supplemented with 10% fetal bovine serum (FBS, heat-inactivated, Gibco, #16140071) and 1% penicillin/streptomycin (“IMG media”).

The drug ebselen [2-phenyl-1,2-benzisoselenazol-3(2H)-one] was chosen as a robust inhibitor of DMT1 [63]. Ebselen was purchased from Focus Biomolecules (#10-2288) and re-suspended in sterile dimethyl sulfoxide (DMSO; Sigma, #276855). IMG cells were plated and grown overnight in six-well-plates (150,000–200,000 cells/well) in IMG media. The next day, cells were treated for 24 h with either 25 μM ebselen or control DMSO. This concentration of ebselen was chosen as the treatment dose following preliminary experiments indicating this dose decreased cellular iron content and following similar reported doses from previous work [64]. Following 24 h of ebselen/DMSO treatment, cells were further treated as described below.

In both IMG cells and primary isolated microglia in the young and aged mice experiments, oligomeric Aβ_1-42 was used as an acute AD-associated inflammatory stimulus. Aβ (HFIP-treated, rPeptide #A-1163-2) and scrambled Aβ (rPeptide #A-1004-2) were purchased from rPeptide and 5 mM stock solutions were prepared with sterile, anhydrous DMSO (Sigma #276855) and sonicated for 15 min before storing aliquots at – 20 °C. The day before cell stimulation, oligomeric Aβ_1-42 was prepared as previously described [39] using cold, sterile phenol-free Ham’s F-12 media (R&D Systems, #M25350↗) and allowed to rest at 4 °C for 24 h. The next day, cells were treated with 1 μM Aβ_1-42 or scrambled Aβ for 24 h before lysis and collection for RNA isolation. For in vitro experiments in IMG cells, ferric ammonium citrate (FAC, Sigma, #F5879) was used as a non-transferrin-bound form of iron. FAC was re-suspended fresh in sterile RNase-free water immediately before each experiment, and cells were treated for 24 h with 50 μM FAC based on literature recommendations [39, 65] or water (control), with or without Aβ prior to lysis and collection for RNA isolation or ICP-MS, as described below.

Following 24 h of treatment with scrambled Aβ or 1 μm Aβ_1-42 ± FAC, IMG cells were collected for ICP-MS analysis of intracellular iron content. After washing twice with ice-cold 1 × PBS, cells were collected into metal-free tubes using Accutase, and total cell counts were measured for data normalization. After centrifugation at 600xg for 5 min and removal of supernatant, cells were acid-digested in 150 μL trace-metal grade nitric acid (70%, OPTIMA Grade HNO_3, Fisher-Sci, #A467-250), and 30% ultra trace-grade hydrogen peroxide (Thermofisher) was added at a 1:4 dilution (37.5 μL H₂O₂). Samples were vortexed, incubated at 65 °C overnight, and diluted the next day with Ultrapure Milli-Q water (Ω18.2) at 10 times the volume of nitric acid (1.5 mL water). ICP-MS was performed at the Vanderbilt Mass Spectrometry Research Center using an Agilent 7700 ICP-MS (Agilent) attached to a Teledyne autosampler (CETAC Technologies, Omaha, NE). The following settings were used: cell entrance = − 40 V, cell exit = − 60 V, plate bias = − 60 V, OctP bias = − 18 V, and collision and cell helium flow = 4.5 mL/min. Samples were introduced by peristaltic pump and taken up at 0.5 rps for 30 s, followed by 30 s at 0.1 rps for signal stabilization. A calibration curve for each isotope was made at 0, 1, 10, 100, 1000, 5000, and 10,000 ppb, and blanks were run following standard calibration to wash out signal from the 10,000 ppb standard. Data were acquired and analyzed using the Agilent Mass Hunter Workstation Software version A.01.02.

Lysed cell samples from all experiments (i.e., CD11b⁺ microglia and primary isolated glia) were processed for total mRNA using an RNeasy Micro Kit with DNase treatment according to manufacturer’s instructions (Qiagen, Hilden, Germany, #74004), with the exception of the IMG experiments, which used the RNeasy Mini Kit (Qiagen, # 74104). Following on-column RNA purification and elution, cellular RNA was reverse transcribed into cDNA at equal concentrations across samples using iScript Reverse Transcriptase (BioRad, Hercules, CA). RT-qPCR was conducted to assess the expression of several genes and confirm Slc11a2 knockdown using FAM-conjugated TaqMan Gene Expression Assay primers (Thermofisher, shown in the table in Additional File 3) and iQ Supermix (BioRad). PCR reactions were performed in duplicate under thermal conditions: 95 °C for 10 min, followed by 40 cycles of 95 °C for 15 s, and 60 °C for 45 s. The expression of each gene was normalized to a housekeeping gene (either 18S or ActinB where indicated), and relative expression values were analyzed utilizing the comparative cycle threshold 2^−ΔΔCT method [66].

Following on-column purification and DNase treatment with the Qiagen RNeasy Micro Kit, total mRNA extracted from hippocampal CD11b⁺ samples was submitted to the Vanderbilt Technologies for Advanced Genomics (VANTAGE) Core facility for sample quality control assessment and RNA-sequencing (RNA-seq). Only hippocampal CD11b⁺ microglia isolated from female animals were used for RNA-seq, following earlier findings of significant behavioral differences primarily in Slc11a2^KD female APP/PS1 animals. The concentration of RNA samples was determined by NanoDrop (ThermoScientific). Sample Quality Control analysis was assessed using fluorometry Qubit and integrity by BioAnalyzer, and a RIN value of > 7 was confirmed for all samples before proceeding to library preparation and sequencing. Paired-end sequencing libraries were constructed using a standard mRNA NEBNext Poly(A) selection Library Prep Kit (Illumina). Library Quality Control analysis was performed by using Qubit and BioAnalyzer to determine the concentration and size bp. Samples were then sequenced at multiplex Paired-End 150 bp using the Illumina NovaSeq 6000 sequencing platform. To confirm sequencing quality, Illumina Quality Scores were calculated utilizing the following equation: Q = − 10log₁₀I. All samples sequenced reached sequencing quality of at least Q30.

Gene alignment, read mapping, gene counts quantification, and differential gene expression analyses were conducted at the Creative Data Solutions (CDS) Core at Vanderbilt. RNA-seq reads were adapter-trimmed and quality-filtered using Trimgalore v0.6.7 [67] and Cutadapt 1.18 [68] to remove adapter sequences and pairs that were either shorter than 20 bp or that had Phred scores less than 20. An alignment reference was generated from the mm39 mouse genome and GENCODE comprehensive gene annotations (M31), to which trimmed reads were aligned and counted using Spliced Transcripts Alignment to a Reference (STAR) v2.7.9a [69] with the –quantMode GeneCounts parameter. About 30–50 million uniquely mapped reads were acquired per sample. DESeq2 package v1.36.0 [70] was used to perform sample-level quality control, low count filtering, normalization and downstream differential gene expression analysis. Genomic features counted fewer than five times across at least three samples were removed. The default significance cutoff (0.1) for optimizing the independent filtering in DESeq2 was also used.

The measure of standard deviation (sd) and quantiles on principal component 1 (PC1) among samples was used to assess whether any samples were a statistical outlier. One sample in the Control APP/PS1 group was removed from analyses after exhibiting a deviation of > 2 standard deviations and an interquartile range of > 1.5 in PC1 compared to its respective group (sample shown in Additional File 4B and C). Five to six biological replicates per condition were included for the differential expression analysis. Differentially expressed genes were identified using a false discovery rate (FDR) adjusted p-value threshold of 0.05, calculated using the Benjamini-Hochberg (BH) procedure for multiple hypothesis testing correction, and a log2 fold change threshold of greater than 1. Gene set enrichment analysis (GSEA) [71] was performed using the R package Clusterprofiler [72] with gene sets from the Mouse MSigDB database [73]. Coverage of reads across annotated exons in the Slc11a2 gene analysis was done using the R package ggcoverage 1.3.0 [74]. All data processing was performed at the Advanced Computing Center for Research and Education (ACCRE) at Vanderbilt University.

Data are presented as mean ± S.E.M. All experiments were analyzed using analysis of variance (ANOVA) for multiple comparisons followed by appropriate post-hoc analyses unless otherwise noted. Male and female data were first compared using ANOVA (2(Sex) × 2(Genotype) × 2(Treatment), followed by Sidak’s corrections for multiple comparisons and analysis of interaction effects. Based on our previous work showing sex differences in Slc11a2 expression between males and females, most primary analyses were conducted within each sex separately to assess the effect of Slc11a2 knockdown in each sex. To do this, a 2(Genotype) × 2(Treatment) ANOVA followed by Sidak’s corrections was used. In analyzing MWM data, repeated measures ANOVA (2(Knockdown) × 2(APP/PS1 Genotype) × 5(Day)) was used to analyze latency data from multiple training days and Tukey’s post-hoc analysis was used following significant F values to establish differences among all groups. Data from primary cell and IMG cell experiments were analyzed using either 2(Treatment) × 2(Age) ANOVA or 3(Treatment) × 2(ebselen/DMSO) ANOVA, respectively. Sidak’s post-hoc analysis was used for interaction effects and corrections for multiple comparisons. Statistical outliers within each group for all studies were identified using either the ROUT or Grubb’s method for outliers and excluded from statistical analyses. GraphPad Prism 9 (GraphPad Software, San Diego, CA, USA) was used for statistical analyses outside of RNA-seq analyses conducted in R. Differences among groups were considered significant at values of p < 0.05.

The data presented are from experiments in primary microglia isolated from young and aged mice treated with Aβ in vitro, immortalized IMG cells treated with a DMT1 inhibitor followed by Aβ and iron, and data from a triple-transgenic AD mouse model with microglial Slc11a2 knockdown in both sexes. These experiments were designed to examine relationships between Aβ and microglial DMT1 at the cellular level, to directly target DMT1 in vitro to examine changes at the molecular level, and to determine the effects of targeting microglial Slc11a2/DMT1 at the behavioral level.

To determine the effects of knocking down Slc11a2 in vivo, we generated a transgenic mouse line allowing for inducible knockdown of Slc11a2 in microglia between 5 and 6 months of age. Between 7 and 9 months after tamoxifen treatment, when mice were 12–15 months of age, male and female control WT, control APP/PS1, Slc11a2^KD WT, and Slc11a2^KDAPP/PS1 mice were run through a series of behavioral assays to assess the effect of microglial Slc11a2 knockdown on aspects of behavior and cognition.

To determine the effect of Slc11a2 knockdown on hippocampal microglia, we first compared microglial gene expression between Slc11a2^KD and Control WT females. As a result of knockdown alone, we only found 10 DEGs (Additional File 7A). Top genes altered included Ccr6 and Cd5 (Additional File 7B, C). We then aimed to determine how Slc11a2 knockdown affects microglial gene expression in the APP/PS1 female animals. There were 449 genes significantly upregulated and 130 downregulated in microglia isolated from Slc11a2^KDAPP/PS1 animals compared to microglia from control APP/PS1 mice. Of these DEGs, Enpp2 and Ttr were robustly upregulated in knockdown cells compared to controls (Fig. 7B). Of the top 50 identified DEGs between Slc11a2^KDAPP/PS1 and control APP/PS1 females, Apoe (encoding apolipoprotein E), Cybb (gene for NOX2), and lipid-droplet-accumulating microglia (LDAM) marker, Ly9, were also significantly downregulated in the knockdown cells compared to the control APP/PS1 cells (Fig. 7B, C). GSEA in the Slc11a2^KD and control APP/PS1 microglia revealed significant increases in genes associated with cellular metabolism—in particular, oxidative phosphorylation and fatty acid metabolism—and reactive oxygen species (ROS) pathways (Fig. 7D). Slc11a2 knockdown cells also exhibited significant decreases in genes associated with TNF and NFκB inflammatory signaling and Wnt signaling, as indicated by gene-set enrichment pathway analysis (Fig. 7D). When comparing relative expression of specific genes in the sequencing dataset, we observed significant alterations in several genes involved in inflammatory and oxidative stress-associated pathways in Slc11a2^KD versus control cells from APP/PS1 females. Specifically, we observed a significant decrease in Ctsb and Csf1 (markers associated with DAMs) [84] in the knockdown cells, as well as a significant increase in Tgfbr1 (p < 0.05) and increase in Trem2 compared to control cells (although not statistically significant, p = 0.068) (Fig. 7E). In examining genes related to iron handling and redox status, we observed a significant increase in iron exporter gene Slc40a1 and antioxidant gene Gpx4 in Slc11a2^KDAPP/PS1 cells compared to control APP/PS1 microglia (Fig. 7F). Additionally, Slc11a2 knockdown cells exhibited decreases in pro-oxidant genes, such as Hif1α and Cybb, and a robust decrease in the iron-related gene encoding ceruloplasmin (Cp) (Fig. 7F). To assess changes in specific DAM-like markers further, we conducted targeted analysis of gene sets related to different subsets of DAMs reported in the literature [36, 85–88], including a ‘pro-inflammatory’, ‘anti-inflammatory’, ‘lipid-associated DAM (LDAM)’, and ‘ferroptosis’ gene set (Fig. 7G). These data further demonstrate that Slc11a2 knockdown resulted in changes to some, but not all, DAM markers. Although Slc11a2^KD cells isolated from APP/PS1 mice exhibited significant differences in the expression of several markers compared to control APP/PS1 microglia, Slc11a2^KDAPP/PS1 microglia displayed a transcriptional profile still distinct from control, non-APP/PS1 WT cells (black dotted line, Fig. 7E, F). In comparison to control WT cells, Slc11a2^KDAPP/PS1 microglia upregulated DAM and aging-related markers Csf1, Hif1α, Cybb, and Ctsb—albeit, to a lesser degree than control APP/PS1 microglia.

Initial assessment of overall gene expression via PCA and DEGs in these samples revealed significant variance in gene expression in one sample in the Slc11a2^KDAPP/PS1 group compared to the rest of the Slc11a2^KDAPP/PS1 biological replicates (sample labeled as -0004 in PCA plot shown in Additional File 4B and in heat map Fig. 7B). Although this sample was not considered to be a statistical outlier, further RNA-seq analyses conducted following the removal of this sample are shown in Additional File 8. In this analysis, there were 2230 genes significantly upregulated and 2210 significantly downregulated in the Slc11a2^KD versus control APP/PS1 females (Additional File 8B). The top DEGs revealed upregulations in genes including phagocytic-associated Igkc, along with Ttr and Enpp2, and prostaglandin-signaling molecule, Ptgds (Additional File 8C and D). Slc11a2^KD cells also exhibited significant downregulations in several DAM markers—particularly from the ‘pro-inflammatory’ gene subset—including Cybb, Stat1, and Ctsb, as well as Ly9, Hif1α, homeostatic marker Bin2, and Apoe (Additional File 8G). Overall, these data suggest that microglial Slc11a2 knockdown in females decreases expression of some markers related to subsets of DAMs and aged cells in the APP/PS1 model. Several DEGs found in female Slc11a2 knockdown microglia were probed via RT-qPCR in male hippocampal microglia and are shown in Additional File 9. There were no significant changes in Hif1α, Cybb, or Il1β in male microglia due to Slc11a2 knockdown, mirroring the lack of behavioral differences in the male Slc11a2^KD mice.