Abstract

   Leukotrienes, a class of inflammatory bioactive lipids, are well
   studied in the periphery, but less is known of their importance in the
   brain. We identified that the enzyme leukotriene A4 hydrolase (LTA4H)
   is expressed in healthy mouse neurons, and inhibition of LTA4H in aged
   mice improves hippocampal dependent memory. Single-cell nuclear RNA
   sequencing of hippocampal neurons after inhibition reveals major
   changes to genes important for synaptic organization, structure, and
   activity. We propose that LTA4H inhibition may act to improve cognition
   by directly inhibiting the enzymatic activity in neurons, leading to
   improved synaptic function. In addition, LTA4H plasma levels are
   increased in both aging and Alzheimer’s disease and correlated with
   cognitive impairment. These results identify a role for LTA4H in the
   brain, and we propose that LTA4H inhibition may be a promising
   therapeutic strategy to treat cognitive decline in aging related
   diseases.
     __________________________________________________________________

   LTA4H is identified as a therapeutic target and plasma biomarker for
   cognitive impairment in aging and Alzheimer’s disease.

INTRODUCTION

   Leukotrienes (LTs) are critical mediators of inflammation but have been
   studied only minimally in the central nervous system (CNS). However,
   our recent understanding of the role inflammation plays in the aging
   brain provides new questions of how LTs may be implicated. Aging is
   associated with a progressive decline in numerous functions and an
   increased incidence of diseases ([48]1, [49]2). The continuous,
   low-level inflammation in the absence of active infection that
   contributes to age-associated disease has been coined inflammaging
   ([50]3). Inflammaging in the brain is characterized by increased
   neuroinflammation of astrocytes ([51]4, [52]5) and microglia ([53]6,
   [54]7), aberrant infiltration of immune cells into the parenchyma
   ([55]8), loss of synapses ([56]9, [57]10), and impaired cognition
   ([58]6, [59]11, [60]12). Inflammaging in the periphery ([61]13, [62]14)
   and within the brain ([63]15–[64]17) is detectable before the onset of
   disease symptoms, and failure to resolve this chronic inflammation has
   been hypothesized to be a major contributor to aging-related cognitive
   decline.

   LTs are a group of biologically active lipid mediators (LTA4, LTB4,
   LTC4, LTD4, and LTE4), which are secreted from multiple cell types and
   play a critical role in recruiting leukocytes and other immune cells to
   sites of inflammation in the periphery ([65]18). Cysteinyl LTs
   (Cys-LTs), LTC4, LTD4, and LTE4 have been shown to influence CNS
   diseases including traumatic brain injury, Alzheimer’s disease (AD),
   Parkinson’s disease (PD), multiple sclerosis, epilepsy, Huntington’s
   disease, depression, and gulf war illness ([66]19–[67]25). However, the
   non–Cys-LTs, LTA4, and LTB4, which bind to Leukotriene B4 Receptor 1
   (BLT1) and BLT2 receptors, respectively, LTB4R1 and LTB4R2 in mouse
   ([68]26), are understudied in the brain. In the periphery, these G
   protein–coupled receptors are broadly expressed in leukocytes and are
   responsible for the chemotaxis of immune cells to sites of inflammation
   ([69]27, [70]28). Recently, LTB4 has been identified in the cerebral
   spinal fluid (CSF) of patients with AD, and LTB4 levels have been shown
   to increase with worsening cognitive diagnosis ([71]29), suggesting
   that LTB4 may play a role in the brain similar to the Cys-LTs. The
   receptor for LTB4 ([72]30) and upstream processing enzymes ([73]31) are
   expressed in mouse and human neurons, and LTB4 has been detected in the
   brains of mice ([74]25), but the role of LTB4 signaling in these brain
   cell types is unknown. In addition, in humans, the enzyme
   5-lipoxygenase (5-LOX), which is upstream of LTA4 hydrolase (LTA4H), is
   increased in the brains of patients with AD ([75]32), and ablation of
   5-lipoxygenase–activating protein in an AD mouse model ameliorates
   cognitive impairment ([76]33), prompting further investigation into the
   connection of LTA4H and cognition.

   To better understand the role of LTB4 in the brain, we characterized
   the expression pattern of the LTB4 receptor LTB4R1 and processing
   enzyme LTA4H. We hypothesized that the reduction of LTB4 levels could
   have a direct impact on the function of neuronal cells that express
   LTB4R1 and LTA4H and tested two small-molecule LTA4H inhibitors:
   SC57461A ([77]34, [78]35) and AKST1220 in aged mice. We evaluated
   cognitive effects with behavioral experiments and performed RNA
   sequencing to investigate transcriptomic changes within neurons of the
   hippocampus. LTA4H inhibitors have been extensively studied in the
   clinic for peripheral inflammatory diseases ([79]36), yet they are
   relatively understudied in CNS diseases. To address this, we measured
   LTA4H levels in human plasma, across age from healthy donors and
   patients with AD, and identified an increase in LTAH across aging and
   disease. Together, our findings identify a role for LTA4H in the brain
   and suggest that LTA4H inhibition may provide a therapeutic approach to
   treat cognitive decline.

RESULTS

LTA4H and LTB4R1 are localized to neurons

   LTB4, a lipid signaling molecule, has been widely characterized
   throughout the periphery as a chemoattractant that recruits immune
   cells to sites of inflammation ([80]37). More recently, it has been
   identified in the CSF of AD human patients ([81]29), suggesting that
   this bioactive lipid may have functional consequences in the brain. To
   better understand whether LTB4 signaling could play a role in the CNS,
   brain expression of LTA4H, the enzyme responsible for its generation,
   was examined ([82]Fig. 1A). LTA4H protein was detectable in crude brain
   cortex homogenates by Western blot ([83]Fig. 1B). Via
   immunohistochemistry (IHC), LTA4H was found to be expressed in neurons,
   colocalizing with the nuclear maker NeuN throughout the brain ([84]Fig.
   1C). In addition, in the hippocampus, the Lta4h gene is highly
   expressed by approximately 10% of neurons (fig. S1 and table S1). The
   nuclear localization of LTA4H in other cell types, such as macrophages
   and basophils, was previously identified to be important for the
   coordination of LTB4 processing from arachidonic acid (AA) by first
   5-LOX and subsequently LTA4H ([85]Fig. 1A) ([86]38). The unexpected
   identification of LTA4H in mouse neurons suggests that LTB4 may be
   synthesized in and secreted from neurons directly.

Fig. 1. LTA4H and LTB4R1 are localized to neurons.

   [87]Fig. 1.
   [88]Open in a new tab

   (A) Schematic depicting LTB4 processing and signaling in neurons. (B)
   LTA4H is detected in crude phosphate-buffered saline (PBS) and
   radioimmunoprecipitation assay (RIPA) Western blot homogenates from
   mouse cortex and is unchanged with age. Kruskal-Wallis test, P =
   0.3536. All data shown as means ± SEM. (C) LTA4H is localized in the
   nucleus of neurons. Representative images of LTA4H (green) and NeuN
   (purple) in adult mouse brain. Scale bar, 20 μm. (D) LTB4R1 is
   localized along the dendrites of neurons. Representative images of
   LTBR41 (green) and MAP2 (purple) in adult mouse brain. Scale bar, 10
   μm. (E) Biochemical postsynaptic density (PSD) enrichment from mouse
   hippocampus identified LTB4R1 to be enriched in membrane and
   extrasynaptic fractions. LTA4H is located within crude and cytosolic
   fractions. Presynaptic marker SYNAPSIN1 and postsynaptic maker PSD95
   are used to show enrichment.

   To understand where LTB4 signaling is occurring in the CNS, mouse brain
   tissue was stained for the high-affinity LTB4 receptor, LTB4R1, and its
   expression was detected along the dendrites of neurons and colocalized
   with microtubule-associated protein (MAP2) in the cortex and the cornu
   Ammonis 1 (CA1) region of the hippocampus ([89]Fig. 1D). Furthermore,
   using biochemical fractionation of postsynaptic density (PSD)–enriched
   synaptic membranes, LTB4R1 was enriched in the membrane and
   extrasynaptic membrane fractions and is detectable in PSD-enriched
   fractions ([90]Fig. 1E). Together, this characterization suggests that
   the inflammatory lipid LTB4 is both produced by and can signal to
   neurons within the mouse brain.

LTA4H inhibition modulates LT signaling in the brain

   To determine the potential role of LTA4H in neurons, two specific and
   potent small-molecule inhibitors of LTA4H enzymatic function AKST1220
   and SC57461A (figs. S2 to S4) were tested in aged mice. As plasma LTB4
   levels were elevated in aged mice ([91]Fig. 2A), plasma LTB4 levels
   were used to confirm compound efficacy in reduction of LTB4 levels in a
   target engagement assay ([92]Fig. 2B and fig. S2), where both compounds
   show comparable levels of inhibition. LTA4H inhibition did not affect
   the LTB4 precursor lipid AA (fig. S3).

Fig. 2. AKST1220 and SC57461A are specific and potent small-molecule
inhibitors of LTA4H.

   [93]Fig. 2.
   [94]Open in a new tab

   (A) Plasma LTB4 measured from unstimulated blood is significantly
   increased from 3 to 22.5 months (M). Unpaired t test, **P = 0.0011, F =
   2.952 (n = 15 and 11). (B) Plasma LTB4 levels from
   calcimycin-stimulated blood are significantly reduced following 4 weeks
   of dosing with LTA4H inhibitors AKST1220 or SC57461A. Kruskal-Wallis
   test, P = 0.0006 followed by Dunn’s multiple comparisons: vehicle
   versus AKST1220, **P = 0.0034; vehicle versus SC57461A, *P = 0.0212;
   and AKST1220 versus SC57461A, P > 0.9999 (n = 5, 6, and 9). (C)
   Small-molecule structures of AKST1220 and SC57461A. (D) Molecular
   docking of AKST1220 into the structure of human LTA4H (Protein Data
   Bank ID: 6ENC). Right panel shows zoom in of hydrolase binding pocket
   (ball and stick cartoon) with individual interacting amino acids
   highlighted. (E) Molecular docking of SC57461A into the hydrolase
   binding pocket of human LTA4H. (F) Overlay of AKST1220 (red) and
   SC57461A (yellow) in the hydrolase binding pocket of human LTA4H shows
   similar binding modes of the two small molecules. All data are shown as
   means ± SEM.

   The expected binding modes of both small molecules to LTA4H are
   illustrated via molecular docking ([95]Fig. 2, C to F) ([96]39),
   demonstrating that AKST1220 and SC57461A bind in the hydrolase pocket
   of LTA4H and exhibit very similar amino acid interactions ([97]Fig.
   2F). The carboxylate substituents undergo polar contacts with the
   catalytic zinc responsible for hydrolysis of LTA4 to generate LTB4
   ([98]Fig. 1A), indicating that the mechanism of inhibition for both
   molecules is through competitive binding to the active site of the
   protein. Binding affinity calculations and ligand-enzyme energy scores
   also demonstrate that AKST1220 and SC57461A exhibit similar binding
   affinities toward human LTA4H (table S2).

Inhibition of LTA4H with AKST1220 in aged mice improves hippocampus-dependent
cognitive impairment

   To determine the impact of long-term LTA4H inhibition in aged mice, 20-
   to 22-month-old mice were dosed with AKST1220 or SC57461A. Mice dosed
   for 1 month using the SC57461A LTA4H inhibitor showed no statistically
   significant improvement in the y-maze spatial learning task ([99]Fig.
   3A) or in the radial arm water maze (RAWM) (fig. S5, A to D).
   Conversely, mice treated with AKST1220 for 1 month demonstrated
   improved cognitive performance in the y-maze ([100]Fig. 3A and fig.
   S5D). To understand whether AKST1220 treatment could affect cognitive
   performance with a shorter dosing window, we tested the performance of
   aged mice using the contextual fear conditioning task following 10 days
   of dosing ([101]Fig. 3B and fig. S5E) and identified enhanced cognitive
   performance with LTA4H inhibition. Furthermore, the detrimental effect
   of elevated LTA4H levels on cognition was further confirmed by direct
   administration of human recombinant protein to young mice (fig. S6).

Fig. 3. LTA4H inhibition with AKST1220 improves cognition in aged mice.

   [102]Fig. 3.
   [103]Open in a new tab

   (A) Percent entries into the novel arm of the y-maze are significantly
   increased after 4 weeks of AKST1220 dosing compared to vehicle.
   Unpaired t test, **P = 0.0050 (n = 14 and 13). Percent entries into the
   novel arm of the y-maze are unchanged after 4 weeks of SC57461A dosing
   compared to vehicle. Unpaired t test, P = 0.4743 (n = 13 and 15). (B)
   Contextual fear conditioning (CFC) is significantly improved after 10
   days of AKST1220 treatment. Two-way analysis of variance (ANOVA) with
   repeated measures, **P = 0.007, F = 8.638 (n = 14 and 14). All data are
   shown as means ± SEM.

   Following a single oral dose of AKST1220 (10 mg/kg), aged mice have
   high concentrations found in total brain homogenate that persists up to
   24 hours ([104]Fig. 4A). Furthermore, AKST1220 was brain-penetrant at
   doses as low as 0.1 mg/kg ([105]Fig. 4A). Conversely, SC57461A was not
   found to be brain-permeant ([106]Fig. 4B) and, compared to AKST1220 at
   the same dose, was relatively undetectable ([107]Fig. 4C). To better
   understand whether AKST1220 was entering the brain parenchyma or was
   bound within the brain vasculature, mice were dosed with
   ^14C-radiolabeled AKST1220 and imaged using micro-radioautography
   followed by IHC for the blood vessel maker lectin. Imaging revealed
   that ^14C-AKST1220 was broadly distributed throughout the brain. Silver
   grain particles of ^14C-AKST1220 were found both within blood vessels
   and in the brain parenchyma of the hippocampus ([108]Fig. 4, D and E,
   and fig. S7). These results suggest that AKST1220 may have a direct
   impact not only on the peripheral LTA4H activity but also locally in
   the brain parenchyma, while SC57461A may function only by inhibiting
   LTA4H in the periphery.

Fig. 4. AKST1220 is a brain-penetrant molecule.

   [109]Fig. 4.
   [110]Open in a new tab

   (A) Brain homogenate concentration of AKST1220 (nanomolar) following a
   single oral dose of 10, 1, or 0.1 mg/kg. (B) Brain homogenate
   concentration of SC57461A (nanomolar) following a single oral dose of
   10 or 5 mg/kg. (C) Comparison of the brain homogenate concentrations of
   AKST1220 (A) and SC57461A (B) at 10 mg/kg. All data contain n = 3 mice
   per treatment group per time point at 20 months of age. All data are
   shown as means ± SEM. (D) Representative image of mouse midbrain after
   a single oral dose of ^14C-AKST1220 (small black particles) and
   subsequent immunohistochemical staining of biotinylated lectin (brown).
   Inset zoom in of the dentate gyrus (DG) with black arrows identifying
   blood vessels and arrowheads identifying AKST1220 in the brain
   parenchyma. H&E, with hematoxylin and eosin. (E) Dark-field image of
   (D) to visualize ^14C-AKST1220 silver grains as bright white particles
   on a dark background.

LTA4H inhibition targets pathways responsible for synaptic organization,
structure, and activity

   To investigate the mechanism of improved cognition following chronic
   LTA4H inhibition, single nuclear RNA sequencing (Nuc-seq) was performed
   on hippocampal neurons ([111]40) in aged mice following 1 month of
   LTA4H inhibitor treatment. Hippocampal nuclei were obtained from
   microdissected hippocampi that were processed following the 10x
   Genomics protocol for Nuc-seq, labeled with a phycoerythrin
   (PE)–conjugated anti-NeuN antibody, and enriched for neuronal nuclei
   using fluorescence-activated cell sorting (FACS; [112]Fig. 5A and fig.
   S8). A total of 98,570 nuclei were isolated and profiled from three
   animals per treatment (30,744 nuclei from three AKST1220 treated
   animals, 31,402 nuclei from three SC57461A-treated animals, and 36,424
   nuclei from three vehicle-treated animals). Low-quality nuclei (less
   than 1000 genes per nucleus) were removed, leaving a total of 93,683
   nuclei with a mean of 3235 genes per nucleus that were retained for
   further processing (fig. S9). Unbiased clustering was performed on the
   nuclei using the Seurat R package [v4; ([113]41, [114]42)] and manually
   curated cell types using known marker genes, identifying CA1 and CA3
   pyramidal neurons, dentate gyrus (DG) cells and a heterogeneous group
   of interneurons (IN) as shown by the uniform manifold approximation and
   projection (UMAP) plot ([115]Fig. 5B). A small number of unidentified
   neuronal and nonneuronal cell types were excluded after the initial
   cell classification, retaining a total of 92,161 neuronal nuclei. LTA4H
   inhibition does not lead to obvious segregation of cells within the
   major cell-type clusters ([116]Fig. 5C). Furthermore, the contribution
   from each treatment type to each of the major cell types is comparable,
   suggesting that LTA4H inhibition does not overtly affect cell-type
   distribution ([117]Fig. 5D). Differentially expressed genes (DEGs)
   between LTA4H inhibition and vehicle treatment were computed for each
   major cell type using Seurat with the model-based analysis of single
   cell transcriptomics (MAST) differential test framework ([118]43). DEGs
   with a fold change (FC) of ≥1.1 and an adjusted P value of <0.05 were
   considered statistically significant resulting in 425 DEGs for CA1
   pyramidal neurons after AKST1220 treatment and 511 significant DEGs
   after SC57461A treatment compared to vehicle ([119]Fig. 5E and table
   S3). To obtain an unbiased overview about DEGs between cell types and
   treatment comparisons, the top 2000 DEGs were selected from all major
   cell types across both comparisons, and hierarchical clustering was
   performed. The resulting heatmap reveals DEG fingerprints shared
   between treatments, most notably for CA1 pyramidal and DG cells
   ([120]Fig. 5F and tables S4 and S5). Notably, IN and CA3 pyramidal
   neurons had fewer shared DEGs than CA1 and DG cells.

Fig. 5. LTA4H inhibition leads to cell-type–specific changes in gene
expression in hippocampal neurons.

   [121]Fig. 5.
   [122]Open in a new tab

   (A) Schematic workflow for Nuc-seq: After microdissection, nuclei are
   stained against NeuN and sorted by fluorescence-activated cell sorting
   (FACS) for 10x Genomics single-nuclei processing. (B) Uniform manifold
   approximation and projection (UMAP) plot reveals robust separation of
   clusters identifying major neuronal cell types in the hippocampus. (C)
   UMAP plot colored by treatment paradigm reveals intermingling of cells
   within clusters. (D) Bar plot representing the relative contribution of
   cells from each treatment paradigm. (E) Volcano plot shows the results
   of differential gene expression changes between AKST1220 treatment and
   vehicle control in each cell type. Genes were considered differentially
   expressed if their FC of >1.1 or <−1.1 (dashed lines). (F)
   Hierarchically clustered top 2000 differentially expressed genes (DEGs)
   between LTA4H inhibitors AKST1220 against vehicle (mid-gray) and
   SC57461A against vehicle (dark gray) for the four major neuronal cell
   types. Red, increased expression after inhibitor treatment; blue,
   decreased expression after inhibitor treatment.

   To better understand the differences between LTA4H inhibition by
   AKST1220 and SC57461A in cognition and brain penetrance, DEGs
   significantly modulated by AKST1220 treatment but unchanged with
   SC57461A were examined (tables S6 to S9). This analysis revealed a
   handful of interesting genes that could contribute to the differences
   in cognition observed between the two compounds. For example, Ralyl
   expression was increased in CA1 and DG following treatment with
   AKST1220 and was unchanged following SC57461A treatment. Higher
   expression of this gene has been linked to greater AD reserve, a
   phenomenon where individuals have a high burden of disease pathology
   but limited cognitive decline ([123]44). In addition, decreased Ralyl
   expression is associated with AD ([124]45) and PD ([125]45, [126]46),
   and genetic mutations have been identified in amyotrophic lateral
   sclerosis (ALS) ([127]47). Conversely, Lingo1 was down-regulated by
   AKST1220 but unchanged with SC57416A treatment. Knockdown or inhibition
   of Lingo1 in neurodegenerative models protects animals from cognitive
   impairment in radial arm and y-mazes ([128]48, [129]49), possibly via
   enhancing neuronal survival ([130]50, [131]51) or decreasing amyloid-β
   production ([132]52). Together, increased Ralyl or decreased Lingo1
   expression following AKST1220 treatment could contribute to the
   enhancement of cognition in aged mice and provide protection in
   patients with AD.

   To determine the extent of similarities between DEGs from the two LTA4H
   inhibitors, the DEGs from both inhibitors were plotted as scatter
   plots: CA1 pyramidal neurons ([133]Fig. 6A), DG neurons ([134]Fig. 6B),
   and CA3 and IN (fig. S10). Consistent with the observations from the
   overview heatmap, the specific comparisons between the two inhibitors
   reveal strong correlations between AKST1220-treatment and SC57461A
   treatment for CA1 and DG (Pearson correlation coefficients of 0.92 and
   0.90, respectively). Pathway enrichment analysis using gProfiler
   ([135]Fig. 6C and fig. S11) ([136]53) revealed a majority of terms with
   the highest enrichment P values in the source ontology “biological
   compartment” with synapse-related terms ranking among the highest.
   Similar to high correlations between the DEGs detected following the
   two inhibitor treatments, high correlations of the gene ontology (GO)
   enrichments were identified for CA1 pyramidal neurons ([137]Fig. 6D)
   and DG neurons ([138]Fig. 6E). Furthermore, the GO terms on the scatter
   plots show a strong correlation (Pearson correlation coefficients of
   0.96 and 0.80, respectively) ([139]Fig. 6, D and E). Together, these
   data highlight some of the similarities and differences between
   potential mechanisms underlying LTA4H inhibition between the two small
   molecules.

Fig. 6. LTA4H inhibitors lead to similar changes in gene expression and
enriched pathways associated with synaptic organization, structure, and
activity.

   [140]Fig. 6.
   [141]Open in a new tab

   (A and B) Scatter plots of log[2] fold changes between LTA4H inhibitors
   AKST1220 (y axes) and SC57461A (x axes) for CA1 (A) and DG. (B) High
   correlation of DEGs between LTA4H inhibitors. (C) Gene ontology (GO)
   enrichment for top 500 genes in AKST1220 versus vehicle comparison for
   CA1 neurons reveals enrichment of synapse-associated terms. Y axis,
   −log[10] (adjusted P value) of enrichment, x axis, and color-coded,
   ontology sources. MF, molecular function; CC, cellular compartment; BP,
   biological process; KEGG, Kyoto Encyclopedia of Genes and Genomes;
   REAC, reactome; TF, transcription factor; MIRNA, microRNA; CORUM,
   comprehensive resource of mammalian protein complexes; HP, human
   phenotype; WP, Wiki pathways. (D and E) Scatter plots of GO terms
   associated with both LTA4H inhibitors in CA1 (D) and DG (E). X axis,
   −log10 (adjusted P value) of enrichment terms for SC57461A
   treatment–associated terms; y axis, −log10 (adjusted P value) of
   enrichment terms for AKST1220 treatment–associated terms. Fill colors
   encode term sizes of enriched GO terms.

   Inhibition of LTA4H in aged mice with AKST1220 resulted in improvements
   to hippocampus-dependent memory tasks, and GO pathway enrichment
   suggested a potential molecular mechanism for this improvement in
   synaptic organization, structure, and activity. To better understand
   these changes, some of the top DEGs up-regulated by LTA4H inhibition
   were first investigated (fig. S12 and tables S4 and S5). The CA1 and DG
   regions of the hippocampus were focused on because of their link to
   spatial and contextual learning and memory ([142]54–[143]59), and they
   had the strongest impact following LTA4H inhibition ([144]Fig. 5F). For
   example, in CA1, the important excitatory synaptic genes Cadm2
   (avg_log[2]FC AKST1220 = 0.15 and avg_log[2]FC SC57461A = 0.13) and
   Grin2a (avg_log[2]FC AKST1220 = 0.12 and avg_log[2]FC SC57461A = 0.12)
   were significantly up-regulated. Cadm2 encodes for synaptic cell
   adhesion molecule 2, which is responsible for forming homo- and
   heterodimeric complexes at synapses and is an important promoter of
   excitatory synaptic number ([145]60). Similarly, Grin2a encodes for the
   2A subunit of the ionotropic glutamate/N-methyl-d-aspartate receptor
   and is essential for the regulation of synaptic plasticity, synaptic
   strength, and memory ([146]61, [147]62).

   In the periphery LTA4H plays a proinflammatory role ([148]18, [149]63).
   To understand whether LTA4H inhibition could affect cell types
   important for inflammation in the brain, microglia activation was
   quantified using the histological markers ionized calcium binding
   adaptor molecule 1 (Iba1) and CD68. Inhibition of LTA4H did not affect
   microglia activation measured by percent area of Iba1 or CD68 (fig.
   S13). Together with the transcriptomic results, these findings suggest
   that LTA4H inhibition in aged mice may improve age-related cognitive
   impairment by modulating gene expression responsible for changes in
   synaptic structure and function and not by affecting microglia cells
   responsible for proinflammatory signaling. As LTB4R1 is increased in
   PSD-enriched fractions of aged animals ([150]Fig. 1E), these data link
   LTA4H inhibition with synaptic structure and function to improve
   cognitive impairment in aged mice.

LTA4H is increased with age and correlated with cognitive impairment in
humans

   To understand how LTA4H is further relevant to human biology, levels of
   LTA4H were measured from pools of plasma from young (18 years old) and
   aged (65+ years old) healthy human donors using two independent
   methods, enzyme-linked immunosorbent assay (ELISA) and an aptamer-based
   approach from SomaLogic, and found that LTA4H levels were significantly
   elevated in aged plasma ([151]Fig. 7A and fig. S14). To determine the
   granularity of LTA4H plasma changes across different age ranges,
   healthy human plasma samples were quantified from donors in age
   brackets between 18 and 65+ years of age. The largest increase in LTA4H
   occurred between 18 and 30 years of age with only a more gradual
   increase in years after ([152]Fig. 7B and fig. S14B).

Fig. 7. LTA4H is increased in age and correlated with cognitive impairment in
humans.

   [153]Fig. 7.
   [154]Open in a new tab

   (A) LTA4H levels (nanograms per milliliter) in pools of healthy human
   plasma measured by enzyme-linked immunosorbent assay (ELISA) are
   increased between 18 and 65 years of age. Five human plasma samples
   were pooled per age. Mann-Whitney test, ****P < 0.0001. n = 9 and 8.
   (B) LTA4H levels in individual healthy human plasma samples were
   measured by SomaLogic, and relative fluorescence units (RFU) show an
   increase with age. Kruskal-Wallis test, P = 0.0023 with Dunn’s multiple
   comparisons: 18 versus 45, P = 0.0198; 18 versus 55, P = 0.004; and18
   versus 66, P = 0.0099; n = 50 per age. (C) LTA4H levels in individual
   human plasma from patients diagnosed with subjective cognitive decline
   (SCD), mild cognitive impairment (MCI), or possible/probable AD
   measured by SomaLogic and RFU show an increase with higher cognitive
   impairment diagnosis. Kruskal-Wallis test, P = 0.0023 with Dunn’s
   multiple comparisons: SCD versus MCI, P = 0.0037; and SCD versus AD, P
   < 0.0001; n = 122 (SCD), 106 (MCI), and 104 (AD). (D) Plasma levels of
   LTA4H measured in RFU at SomaLogic are correlated with mini-mental
   state exam (MMSE) score following correction for confounding factor of
   age. Age was found to not be a confounding factor in the comparison of
   LTA4H levels between AD, MCI, and SCD (fig. S9). Spearman R =
   −0.1832712, **P = 0.0034; n = 254. (E) Plasma levels of LTA4H are
   correlated with plasma levels of total tau measured in RFU at
   SomaLogic. Spearman R = 0.1978, **P = 0.0032; n = 221. All data are
   shown as means ± SEM.

   To determine whether LTA4H levels are relevant to age-related human
   disease, the plasma from patient donors with different diagnoses of
   cognitive impairment: subjective cognitive decline (SCD), mild
   cognitive impairment (MCI), and AD diagnosis probable or possible (AD),
   was evaluated. A stepwise increase in LTA4H plasma levels was observed
   as the diagnosis of cognitive impairment worsened ([155]Fig. 7C).
   Cognition is assessed in humans using the mini-mental state exam
   (MMSE), where a lower score indicates a higher level of cognitive
   impairment and a score of 24 or higher is considered to have no
   impairment. When plotting LTA4H plasma levels against MMSE test score
   at the time of collection, a strong negative correlation was found
   (Spearman R = −0.1832712, P = 0.0034) ([156]Fig. 7D and fig. S15). Age
   was not identified as a confounding factor in this analysis, and there
   was no correlation with LTA4H plasma levels and age in this population
   of patients (fig. S15B), suggesting that, while LTA4H levels increase
   with healthy aging ([157]Fig. 7), they can be further exacerbated with
   disease that is not age-dependent. Last, LTA4H plasma levels were also
   correlated with plasma levels of tau (Spearman R = 0.1978, P = 0.0032)
   ([158]Fig. 7E), a plasma biomarker in AD ([159]64), but LTA4H is not
   correlated with amyloid precursor protein plasma levels and was not
   changed across APOE genetic status (fig. S15, C and D). Together, this
   suggests that LTA4H has relevance in human biology and could be used as
   a potential plasma biomarker for AD or, more broadly, for cognitive
   impairment in patients.

DISCUSSION

   In this study, we identify that LTA4H has a fundamental effect on the
   CNS in aging and disease related cognitive decline. We show that LTA4H
   and the LTB4 receptor LTB4R1 are expressed in neurons and that LTA4H
   inhibition improves cognition in aged mice. Using Nuc-seq and
   pharmacokinetic profiling, we propose that the mechanism of cognitive
   improvement may be by modulating important pathways for synaptic
   structure, organization, and function directly within neurons. As
   AKST1220 is brain permeable, we hypothesize that it may function by
   directly inhibiting neuronally expressed LTA4H within the CNS. However,
   AKST1220 and SC57461A also inhibit peripheral LTA4H and LTB4
   production. Furthermore, SC57461A is not brain-penetrant while still
   affecting neuronal gene expression pathways. Therefore, it is possible
   that reduction in peripheral LTB4 may also provide indirect
   improvements to neuronal health by reducing peripheral inflammation and
   increasing beneficial signaling cascades at the blood brain barrier
   ([160]65). The LT signaling upstream and downstream of LTA4H have
   previously been shown to be important for brain capillary health
   ([161]66–[162]68). Therefore, future studies should explore the impacts
   of peripheral versus central LTA4H inhibition on brain endothelial
   cells and other neurovascular unit cell types. In addition, brain
   levels of LTB4 were unable to be measured from these studies using
   ELISA, likely due to a high matrix effect from brain lysate. Future
   studies should assess brain levels of LTB4 following LTA4H inhibition
   using more sensitive assays, such as purification of LTs by
   high-performance liquid chromatography (HPLC) followed by mass
   spectrometry ([163]69).

   We used molecular modeling of AKST1220 and SC57461A with the human
   LTA4H enzyme crystal structure and pharmacokinetic profiling to
   highlight the similarity of these two small-molecule inhibitors. We
   believe that these results provide an important path toward the aim of
   developing therapeutics to treat cognitive decline in aging and disease
   and provide a broad understanding of the potential molecular mechanisms
   of LTA4H inhibition in the brain. In addition to its hydrolase
   activity, LTA4H also has peptidase activity ([164]63, [165]70). The
   active sites for the peptidase and hydrolase activities are in close
   proximity within the structure of the enzyme ([166]39), and both small
   molecules used in this study are predicted to inhibit the peptidase
   activity of LTA4H. However, the consequences of this additional
   activity were not explored in this study.

   Single-cell nuclear RNA sequencing of hippocampal neurons in aged mice
   following LTA4H inhibition revealed changes to important genes for
   synaptic organization, structure, and activity. For example, the
   Nuc-seq results identify changes to genes like Cadm2 and Grin2a that
   suggest an overall improvement of synapse health, structure, and
   function. In addition, a handful of genes were identified to be
   significantly modulated by AKST1220 treatment but unchanged with
   SC57461A. Differences in gene expression of Ralyl and Lingo1, along
   with distinct brain penetrance profiles, could also contribute to the
   enhancement of cognition in aged mice observed with AKST1220, but not
   SC57461A. However, it is possible that the different doses of AKST1220
   and SC57461A used in this study could explain the different effects
   observed in behavior. The dosing for each compound was chosen on the
   basis of the published literature ([167]34, [168]35, [169]71). While
   the levels of LTA4H inhibition, measured by plasma LTB4, were
   comparable between each compound, the pharmacokinetic profile
   identified that SC57461A had slightly lower plasma levels than AKST1220
   at the same dose. This finding suggests that SC57461A may be
   metabolized or cleared faster than AKST1220. Together, the differences
   in brain penetrance, differential gene expression, and pharmacokinetics
   likely contribute to the different behavioral outcomes between the two
   compounds.

   For these findings to be applied to a potential therapeutic approach,
   LTA4H inhibition should be tested within a CNS model of
   neurodegeneration and neuroinflammation. These results would determine
   whether LTA4H inhibition can overcome the cognitive decline caused by
   aggressive disease pathologies, in addition to the improvement in
   age-related cognitive decline that we have identified here. Last, we
   highlight the potential of LTA4H as a plasma biomarker in AD and
   cognitive decline in humans. LTA4H plasma levels are increased in both
   aging and AD and correlated with cognitive impairment measured using
   MMSE, as well as plasma tau, which has been explored as a
   neurodegenerative protein driver ([170]64, [171]72). Further testing in
   clinical settings is necessary to apply LTA4H as a possible future
   biomarker for human disease. Overall, our study identifies a role for
   LTA4H in the brain and highlights the potential for the development of
   therapeutics and biomarkers for age-related cognitive decline and
   disease.

MATERIALS AND METHODS

Experimental design

   The objective of this study was to characterize and understand the
   observation that LTA4H and LTB4R1 are expressed in mouse hippocampal
   neurons. We hypothesized that LTA4H and LTB4 could play a role in
   age-related cognitive decline and that inhibition of LTA4H may be a
   beneficial therapeutic strategy.

Animals

   All animal handling and use was in accordance with Institutional Animal
   Care and Use Committee approved protocols (ALK-005). Male C57BL/6J mice
   were obtained from the Jackson Laboratory (Sacramento, CA) and shipped
   to Alkahest Inc. before the start of each study. Aged mice arrived at
   17.5 months of aged and were further aged in house to between 20 and
   22.5 months of age. Young mice arrived at 2 months and were further
   acclimated in house for at least 2 weeks before the start of dosing.
   Upon arrival, all mice were single-housed at standard temperature (22°
   ± 1°C) and in a light-controlled environment (lights on from 7:00 a.m.
   to 7:00 p.m.) with ad libitum access to food and water. Aged animals
   were subjected to a pretreatment open-field test before the initiation
   of dosing. Mouse weight, total distance traveled, and average velocity
   were used to evenly group animals between different treatment groups.
   Young mice were evenly distributed on the basis of weight alone.
   Different cohorts used in these findings are described in [172]Table 1.
   All cohorts were on the background strain C57BL/6J, with the exception
   of cohort 9 used for the micro-autoradiography (mARG) study, which was
   performed using CD-1 mice.

Table 1. Description of mouse cohorts.

   NA, not applicable; po, per os (oral dosing); iv, intravenous; bid, two
   times a day; PK, pharmacokinetic; mARG, micro-autoradiography; IHC,
   immunohistochemistry.
   Cohort Treatment groups Age (months) Dose and treatment window Figures
   Cohort 1: young versus aged Untreated 3 NA [173]Figs. 1B and [174]2A
   22.5
   Cohort 2: AKST1220 Vehicle and AKST1220 22.5 10 mg/kg, 4 weeks once
   daily po [175]Figs. 2B and [176]3 (A and B) and figs. S2 (E and F),
   S3A, S5D, and S13 (A and B)
   Cohort 3: AKST1220 Vehicle and AKST1220 22.5 10 mg/kg, 10 days once
   daily po [177]Figs. 2B and [178]3A and figs. S2D and S5E
   Cohort 4: SC57461A Vehicle and SC57461A 22.5 5 mg/kg, 4 weeks once
   daily po [179]Fig. 3A and figs. S2G, S3B, S5 (A to D), and S13 (C and
   D)
   Cohort 5: recombinant LTA4H Vehicle and recombinant human LTA4H 3.5 4.6
   μg/day for 7 days iv fig. S6
   Cohort 6: transcriptomics Vehicle, AKST1220, and SC57461A 22.5
   AKST1220, 1 mg/kg; SC57461A, 2.5 mg/kg bid, 4 weeks po [180]Figs. 5 and
   [181]6 and figs. S1 and S8 to S12
   Cohort 7: PK AKST1220 Vehicle and AKST1220 20 Single dose of 10, 1, and
   0.1 mg/kg po [182]Fig. 4 (A to C) and figs. S4 (A, C, D, and F) and S6
   Cohort 8: PK SC57461A Vehicle and SC57461A 20 Single dose of 10 and 5
   mg/kg po fig. S4 (B, C, E, and F)
   Cohort 9: mARG + IHC Vehicle and ^14C-AKST1220 3 Single dose of 10
   mg/kg po [183]Fig. 4 (D and E) and fig. S7
   Cohort 10: calcimycin stimulation assay Untreated 18 NA fig. S2A
   Cohort 11: pilot dose response AKST1220 Vehicle and AKST1220 18 Single
   dose of 0.03, 0.1, and 0.3 mg/kg po fig. S2B
   Cohort 12: pilot dose response SC57461A Vehicle and SC57461A 3 Single
   dose of 1.25, 2.5, and 5 mg/kg po fig. S2C
   [184]Open in a new tab

Materials

   Recombinant human LTA4H was purchased from R&D Systems (Minneapolis,
   MN) and underwent buffer exchange using a 3000-Da cutoff spin column
   (Thermo Fisher Scientific, Hampton, NH) to remove buffer components
   incompatible with in vivo dosing. The final concentration of LTA4H was
   determined by bicinchoninic acid (BCA) assay (Thermo Fisher Scientific,
   Waltham, MA). Vehicle and LTA4H were aliquoted in single use 1.5-ml
   O-ring tubes and stored at −80°C until needed for injections.

   AKST1220 was synthesized at Boehringer Ingelheim (Ingelheim, Germany),
   previously having been explored in a rabbit atherosclerosis model
   ([185]71) and was stored protected from light before experimentation.
   For behavioral studies, AKST1220 was formulated in 0.5% methylcellulose
   with 0.015% Tween 80 in water or 10% dimethyl sulfoxide (DMSO). For RNA
   sequencing and pharmacokinetic studies, AKST1220 was formulated in 45%
   hydroxypropyl-ß-cyclodextrin (pH 7; Sigma-Aldrich, St. Louis, MO).
   AKST1220 was made fresh every 5 to 7 days and stored at 4°C in between
   dosing sessions. SC57461A was purchased from Cayman Chemicals (Ann
   Arbor, MI) and was stored at −20°C in powdered form. A stock solution
   of 10 mg/ml was made in 100% DMSO and stored at −20°C until day of
   dosing. Stock solutions were thawed and diluted into sterile water
   immediately before each daily po (per os) dosing (1:9). If an entire
   vial of working solution was not used, then the vial was returned to
   −20°C until the next dosing day, where it was combined with new freshly
   thawed stock and made working solution vial for that day.

Dosing paradigms

   For young mouse behavioral studies, 2-month-old mice were divided into
   two treatment groups of 14 to 15 mice each. Mice received a pulse
   dosing paradigm of 7 days of daily intravenous injections of 4.6 μg of
   recombinant human LTA4H or vehicle followed by behavioral testing and
   sacrifice 6 weeks later. For aged mouse behavioral studies, 19.5 to
   20.8 months of age underwent 10 days or 4.7 weeks of daily po dosing
   with vehicle, AKST1220 (10 mg/kg), or SC57461A (5 mg/kg). Mice were
   euthanized 1 day following the conclusion of the last behavior test and
   2 hours after the last dose.

   For transcriptomic profiling, aged mice were divided into groups of 3
   and administered either AKST1220 (1 mg/kg), SC57461A (2.5 mg/kg), or
   vehicle po two times a day for 4 weeks before sequencing. Mice were
   euthanized 2 hours after the last dose.

   For pharmacokinetic analysis of AKST1220, aged mice were dosed by po
   with a single dose of AKST1220 at 10, 1, or 0.1 mg/kg. Mice were
   euthanized at times 0, 0.25, 0.5, 1, 2, 4, 8, and 24 hours after dose.
   Blood, brain, and plasma were collected, and pharmacokinetic analysis
   was performed at Quintara Discovery Inc. (Hayward, CA).

   For pharmacokinetic analysis of SC57461A, aged mice were dosed by po
   with a single dose of SC57461A at 10 or 5 mg/kg. Mice were euthanized
   at times 0.25, 1, and 4 hours after dose. Blood, brain, and plasma were
   collected, and pharmacokinetic analysis was performed at Quintara
   Discovery Inc. (Hayward, CA).

Behavioral paradigms

Open field

   The open-field test was used before dosing in aged mice to randomize
   mice across treatment groups. Mice were brought into the experimental
   room for at least 30 min of acclimation to the experimental room
   conditions (dim lighting) before testing. Mice were placed individually
   into the center of a 16-in–by–16-in (0.40-m–by–0.40-m) box and allowed
   to explore for 15 min. The movements of the mice were recorded and
   tracked within a predefined peripheral and center (65% of total area)
   space using automated software (CleverSys, Reston, VA). In addition,
   the total distance and velocity were measured for duration of the test.

Y-maze

   The y-maze was constructed of acrylic plastic; each arm is 15 in by 3
   in, and the walls of the arms are 6 in high (0.38 m by 0.08 m by 0.15
   m). Unique cues in the form of black shapes were adhered to walls at
   the ends of two of the arms (rising sun and swirl), while the third was
   un-cued and designated as the starting point for the mice. First, mice
   were individually placed in the starting arm and allowed to explore
   only one of the other two arms for 5 min; the second arm was blocked
   off with an acrylic plastic wall identical to that of the rest of the
   apparatus. After a 3-hour delay, the mouse was then returned to the
   maze with now both arms open to explore. All movements were recorded
   and tracked for analysis. The number of entries into and the time spent
   in each of the two arms, familiar and novel, was measured, as well as
   within the hub and start arm.

Contextual fear conditioning

   Mice were brought into the testing room immediately before their trial
   to avoid exposure to sounds and scents from testing. Day 1: For
   training, mice were placed in the chambers, bright house light and fan
   on, for 2 min. Then, an auditory cue [2000 Hz, 70 dB; conditioned
   stimulus (CS)] was presented for 30 s. A 2-s foot shock (0.6 mA;
   unconditioned stimulus) was administered for the final 2 s of the CS.
   This procedure was repeated once, each after a 2-min interval, and the
   mouse was removed from the chamber 30 s after the second shock. The
   pans, chamber walls, and grid floors were cleaned with 70% ethanol
   between trials. Day 2: 24 hours after the training for aged mice and 3
   days after the training for young mice, the mouse was returned to the
   same chamber in which the training occurred (memory for context), and
   freezing behavior was recorded for 3 min. The mouse was returned to its
   home cage. The pans, chamber walls, and grid floors were cleaned with
   70% EtOH between trials.

Radial arm water maze

   The water maze was filled with water at least 24 hours before the test
   to equilibrate to 23°C. The water was dyed with white latex paint to
   make the animals visible for tracking and to allow for the use of a
   hidden platform. Eight distinct visual cues were placed at the end of
   each arm of the RAWM inserts. On day-1 animals were given five trials
   each with a visible platform and a 30- to 60-min intertrial interval.
   Animals had 60 s to reach the platform. If they did not reach the
   platform in that time, they were guided to it and allowed to remain for
   15 s before there were removed from the tank. The goal arm remained
   constant, and a different start arm was randomly assigned for each of
   the five trials so that each mouse started in every arm once except for
   the two arms directly across from the platform. The goal arm was
   switched after every two to three mice and balanced between all
   treatment groups. After each trial, the mice were placed in an empty
   cage with blue pads and allowed to dry off under a heat lamp before
   being placed back into their home cage. Day 2 was the testing day, 48
   hours after training, where animals were subjected to the same test of
   five trials each and a 30- to 60-min intertrial interval, but with a
   hidden platform. Animals were scored for number of errors (entry into a
   non-goal arm) and for latency to reach the platform. All trials were
   recorded using ANY-maze (Wood Dale, IL).

Histology

   Mice were perfused with 0.9% saline transcardially. The brains were
   subsequently dissected and cut sagittally in two even halves. One-half
   was snap-frozen for later use in dry ice, and the other was fixed in 4%
   paraformaldehyde in phosphate-buffered saline (PBS) for use in IHC.
   After 2 days of fixation, the hemibrains were transferred to a 30%
   sucrose in PBS solution. Hemibrains were sectioned sagittally at 30 μm
   on a microtome at −22°C. Blocking was done on free floating sections in
   the appropriate serum at 10% serum in PBS with 0.5% Triton X-100.
   Primary antibodies were incubated overnight at 4°C at appropriate
   dilutions. Fluorescent secondary antibodies were applied the next day
   at a concentration of 1:300 for 1 hour at room temperature. Once
   mounted on slides, ProLong Gold Mounting Media (Thermo Fisher
   Scientific, Waltham, MA) was used to coverslip. The following primary
   antibodies were used in the IHC experiments: rabbit anti-LTA4H at 1:100
   (Abcam, Cambridge, UK), rabbit anti-BLT1 (LTB4R1) at 1:200 (Cayman
   Chemical, Ann Arbor, MI), mouse anti-NeuN (Millipore, Burlington, MA)
   at 1:500, guinea pig anti-MAP2 at 1:200 (Synaptic Systems, Göttingen,
   Germany), mouse anti-CD68 (Bio-Rad, Hercules, California) at 1:1000,
   and rat anti-Iba1 (Wako Chemicals, Richmond, VA) at 1:2500. Images were
   acquired using the Axio Scan.Z1 (Zeiss, Oberkochen, Germany) at 20× or
   a Zeiss LSM800 with Airyscan at 63× and Airyscan processed using Zen
   Blue 2.5 (Zeiss, Oberkochen, Germany).

Plasma sample collection and processing

Mouse plasma samples and calcimycin stimulation

   For calcimycin-stimulated samples, whole mouse blood was collected by
   cardiac puncture using heparin as an anti-coagulant and incubated for
   15 min in 37C/5% CO[2]. During this time, a 0.1 mM calcimycin stock was
   made by diluting 10 mM calcimycin (Sigma-Aldrich, A23187) formulated in
   DMSO into PBS. Control solution was made by diluting the same volume of
   DMSO in PBS. Both solutions were sonicated for 10 min in a 37°C water
   bath. Calcimycin (0.1 mM; stimulated) or control solution
   (unstimulated) was added to each well of whole blood and incubated at
   37C/5% CO[2] for 30 min. Whole blood was spun for 10 min at 1000g to
   separate out plasma. Calcimycin stimulation is used to increase the
   dynamic range of the assay by releasing the intracellular pools of LTB4
   ([186]73, [187]74).

Human plasma samples aging cohort

   The first set of plasma samples was collected from healthy male donors
   of 18, 30, 45, 55, and 65 to 68 (average 66) years of age by
   plasmapheresis as part of standard plasma collection for Grifols Biomat
   USA source plasma. “Retain tubes” of individual donations containing 3
   ml of citrate plasma were stored at −30°C. Only samples from
   “infrequent donations” (collected >60 days after preceding plasma
   donation) were used. Plasma samples of 40 individuals per age group
   were pooled into eight pools of five samples each for proteomic
   analysis. A second set of plasma samples was collected from healthy
   male and female plasma donors of 18- to 22-year-old subjects and 55- to
   65-year-old subjects by venipuncture as part of plasma donation at
   Access Biologicals (Vista, CA). Samples containing either EDTA or
   citrate plasma were frozen at −30°C and subsequently were aliquoted and
   stored at −80°C until use. Aging plasma samples were acquired at
   Grifols or Access Biologicals for commercial purposes. All donors gave
   written consent before plasma donation.

Human AD plasma samples

   Plasma samples, demographic, clinical, and cognitive information,
   including MMSE scores, were used for analysis from patients with SCD (n
   = 122), MCI (n = 106), and AD dementia (n = 104) from the Memory Clinic
   at Ace Alzheimer Center Barcelona ([188]75). EDTA plasma samples were
   shipped frozen to Alkahest, aliquoted, and stored at −80°C until use.
   Aliquoting for proteomic analysis of all three sets of plasma samples
   was carried out as follows: Original frozen plasma samples were thawed
   on ice, centrifuged at 3200g for 30 min at 0°C, and filtered through
   0.22-μm MillexGV filter (MilliporeSigma, Burlington, MA) to remove
   cryoprecipitate. Filtrate was aliquoted into cryotubes and stored at
   −80°C until use. Cognitive assessment using the MMSE was performed at
   the Memory Clinic at Ace Alzheimer Center, and data were provided to
   Alkahest as individual MMSE values. All participants gave written
   consent, and the protocol was approved by the Ethics Committee of the
   Hospital Clinic i Provincial (Barcelona, Spain) (EudraCT:
   2014–000798-38) ([189]75).

Enzyme-linked immunosorbent assays

LTB4 plasma quantifications

   Mouse LTB4 levels were detected by running an ELISA on the collected
   stimulated and unstimulated plasma diluted 1:10 (Enzo Life Sciences,
   ADI-901-068; R&D Systems, KGE006B). The ELISA plates were read on a BMG
   LABTECH CLARIOstar plate reader at 405 nm.

LTA4H plasma quantifications

   Human plasma was diluted 1:500 and analyzed for LTA4H using LSBio Human
   LTA4H Sandwich ELISA Kit (LS-F22077-1). The ELISA was read on a BMG
   LABTECH CLARIOstar plate reader at 450 nm.

AA plasma quantifications

   Mouse AA plasma levels were detected by running an ELISA on the
   collected unstimulated plasma diluted 1:10 (Abcam, ab287798). The ELISA
   plates were read on a BMG LABTECH CLARIOstar plate reader at 450 nm.

Western blot

   After perfusion, brains were taken out, and cortex was subdissected
   into hippocampus and cortical cap before being frozen on dry ice.
   Tissue was homogenized first in PBS using the Bead Ruptor, homogenates
   were centrifuged at max speed (~21,330g) for 10 min at 4°C, and then
   supernatants were collected for subsequent analysis of the soluble
   fraction. A second radioimmunoprecipitation assay (Thermo Fisher
   Scientific, Waltham, MA) homogenization of the PBS pellet was performed
   with the same procedure described above. All steps were performed with
   the addition of a protease inhibitor cocktail (Thermo Fisher
   Scientific, Waltham, MA). The protein concentration was quantified
   using a Pierce BCA protein assay kit (Thermo Fisher Scientific,
   Waltham, MA). Samples were then normalized to 25 μg per well and
   reduced with lithium dodecyl sulfate (LDS) before being electrophoresed
   onto Bolt 4 to 12% Bis-Tris Plus Gel (Invitrogen, Waltham, MA). Gels
   were turbo-transferred to Trans-Blot Turbo Mini Nitrocellulose
   membranes (Bio-Rad, Hercules, CA). Blocking was done in a 5% milk
   solution (Bio-Rad, Hercules, CA) in Tris-buffered saline with 0.5%
   Tween (TBST) and incubated with 1:5000 rabbit anti-LTA4H (Abcam,
   Cambridge, UK) overnight at 4°C. Blots were imaged following incubation
   with horseradish peroxidase (HRP)–conjugated secondary antibodies at
   1:5000 (Thermo Fisher Scientific, Hampton, NH) for 1 hour at room
   temperature and subsequently with SuperSignal West Pico PLUS
   Chemiluminescent Substrate (Thermo Fisher Scientific, Waltham, MA).
   Blots were imaged on a Bio-Rad Chemidoc and quantified using Image Lab
   6.0 (Bio-Rad, Hercules, CA) software. Samples were randomized across
   gels and run blinded in single replicates. Blots were stripped and
   reprobed for ACTIN-HRP (Cell Signaling Technology, Danvers, MA) at
   1:5000 overnight at 4°C as a loading control.

   For PSD-enrichment preparation, protocol was adapted from Wiggins et
   al. ([190]76). Hippocampus was subdissected before being homogenized in
   0.32 M sucrose buffer containing 10 mM Hepes and 1 mM EDTA. Some of the
   sample was saved for “crude fraction.” Samples were then centrifuged
   three times to remove unhomogenized, insoluble material and then spun
   again to separate cytosolic and pellet fractions. Supernatant from this
   step was collected and labeled as “cytosolic fraction.” Pellet was
   washed twice with 4 mM Hepes/1 mM EDTA, resuspended, and then spun.
   Pellet was then resuspended in sucrose buffer with 0.5% Triton X-100
   and incubated for 15 min to obtain “membrane fraction.” Then, samples
   were centrifuged, and supernatant was collected as “extrasynaptic
   membrane fraction.” Samples were resuspended in sucrose buffer with 1%
   Triton X-100, 1% SDS, and 1% sodium deoxycholate; incubated for 1 hour;
   and then centrifuged to yield “PSD-enriched fraction.” Protein
   quantification and western blotting were performed as stated above with
   the following antibodies: SYNAPSIN1 (Cell Signaling Technology,
   Danvers, MA) at 1:1000, PSD95 (Cell Signaling Technology, Danvers, MA)
   at 1:1000, LTA4H (Abcam, Cambridge, UK) at 1:5000, and LTB4R1 (Cayman
   Chemical, Ann Arbor, MI) at 1:200.

RNA sequencing

   After 4 weeks of dosing, aged mice were anesthetized 2 hours after the
   final dosing using isoflurane and transcardially perfused with 20 ml of
   ice-cold artificial cerebrospinal fluid ([191]77). Hippocampi were
   microdissected from three mice per treatment group and processed for
   nuclei isolation for 10x Genomics single-nuclei sequencing following
   the manufacturer’s guidelines (10x Genomics Sample Preparation
   Demonstrated Protocol [192]CG000124; 10x Genomics, Pleasanton, CA) with
   the following modification: Lysis incubation time was increased to 4
   min. Nuclei were stained for NeuN using a PE-conjugated anti-NeuN
   antibody (1:1000; EMD Millipore, FCMAB317PE) and counterstained with
   DRAQ5 (1:5000; Thermo Fisher Scientific, PI62251) for 30 min on ice to
   allow separation of nuclei from debris particles using a Sony MA900
   FACS sorter (Sony Biotechnology, San Jose, CA). Eight samples of 8000
   nuclei each were collected per treatment group and immediately
   processed for 10x single-nuclei sequencing using the 10x Genomics
   Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 Dual Index
   ([193]CG000315 Rev. B). Single-nuclei libraries were sequenced on an
   Illumina NovaSeq 6000 to a target depth of ~50,000 reads per cell. Raw
   sequencing reads were processed using 10x Genomics Cell Ranger (v6.0;
   with the option –include-introns) and aligned against Gencode GRCm39,
   release M26. Libraries were processed using Seurat [v4; ([194]41,
   [195]42, [196]78)], and nuclei were classified using known marker
   genes. Briefly, the following marker genes were used for cell
   classification: CA1 (Npcd, Stxbp1, Tcf4, Wfs1, Dcn, Calb1, Nov, and
   Rorb), CA3 (Tmem150a, Plagl1, Fmo1, Prkcd, Socs2, Serpinf1, Kctd4, and
   Dkk3), DG (Iqgap2, Gpr161, Prox1, Tdo2, Slc39a6, Slc26a10, Crlf1,
   Sema5a, Abcb10, Gpc4, and Stxbp6), and IN (Gad1, Gad2, Pvalb, Sst, Vip,
   Calb2, Ndnf, Smad3, Reln, and Npy). Cells with ambiguous signatures or
   low signal (<500 genes per cell), and nonneuronal cell types were
   filtered out. Differential gene expression was computed using Seurat’s
   FindMarker function with MAST ([197]43), and GO enrichment was
   performed using gProfiler2 ([198]53); briefly, for each comparison,
   DEGs were ranked [abs(log[2]FC) × −log[10] (P val)], and the top 500
   ordered genes were submitted to gProfiler2. All genes with an average
   expression above threshold (0.01 in Seurat’s “data” slot) per cell type
   were used as custom background list for GO queries. Volcano, scatter,
   and bar plots were generated using ggplot2 ([199]79)
   ([200]https://ggplot2.tidyverse.org), and heatmaps were generated using
   the R package ComplexHeatmap ([201]80).

Pharmacokinetic analysis

   Brain, blood, and plasma samples for bioanalysis of AKST1220 or
   SC57461A were collected and prepared at Alkahest. Brain samples were
   homogenized with two volumes of ice-cold water. Heparin blood was
   collected by cardiac puncture. An aliquot of blood was collected, and
   an aliquot was spun for 10 min at 1000g to separate out plasma.
   Aliquots of brain lysate, whole blood, and plasma were frozen and
   shipped on dry ice to Quintara Discovery Inc. (Hayward, CA) for
   bioanalysis.

   At Quintara, an aliquot of 20 μl of each plasma sample or
   plasma-diluted tissue homogenate was extracted with 140 μl of
   acetonitrile containing internal standard (dexamethasone at 50 ng/ml).
   The mixtures were vortexed on a shaker for 15 min and subsequently
   centrifuged at 3200g for 15 min. A 110-μl aliquot was then transferred
   to a fresh plate for injection into the liquid chromatography–tandem
   mass spectrometry. Calibration standards and quality control samples
   were prepared by spiking the test compound into control mouse plasma or
   brain homogenate and then processed with the unknown samples.

Molecular docking methodology

   The molecular docking studies were carried out using the computational
   software Discovery Studio (DS) Structure-Based-Design software from
   Dassault Systems Biovia Corp., San Diego, USA (v20.1.0.19295). The
   three-dimensional (3D) structure of AKST1220 and SC57461A was prepared
   using the Small Molecules module in DS using CHARMm force field at pH
   7.4 and was energy-minimized using the Smart Minimizer with 2000
   minimization steps (root mean square gradient = 0.01 kcal/mol) and
   distance-dependent dielectric constant at pH 7.4 ([202]81, [203]82).
   The solved x-ray structure of human LTA4H was obtained from Research
   Collaboratory for Structural Bioinformatics (RCSB) data bank (Protein
   Data Bank ID: 6ENC) ([204]39). The protein was prepared after removing
   water using the Prepare Proteins option under the Macromolecules module
   in DS using CHARMm force field. A binding sphere of 10 Å was prepared
   using the Receptor-Ligand Interactions module in DS after selecting the
   bound compound 11. The compound 11 was deleted, and molecular docking
   studies were carried out using the 3D structure of AKST1220 or SC57461A
   prepared earlier. The CDOCKER algorithm^2 in the Receptor-Ligand
   Interactions module was used for molecular docking studies. The CDOCKER
   algorithm is based on a simulated annealing protocol and included 2000
   heating steps to reach the target temperature of 700 K followed by 5000
   cooling steps and a target temperature of 300 K. The root mean square
   deviation was calculated using the Structure module in DS. The top 10
   docked binding modes of AKST1220 and SC57461A in human LTA4 hydrolase
   obtained were evaluated on the basis of the CDOCKER energy and CDOCKER
   interaction energy scores (kilocalories per mole). The ligand-enzyme
   interactions for the top ranked binding mode were evaluated by
   investigating the type of polar and nonpolar interactions and their
   distance parameters ([205]81, [206]82). The binding affinities of
   AKST1220 and SC57461A in kilocalories per mole was calculated using the
   equation E[binding] = energy of complex (E[ligand-receptor]) − energy
   of ligand (E[ligand]) – energy of receptor (E[receptor]) and the
   implicit solvent function generalized Born with a simple switching.

SomaScan assay (SomaLogic) methodology

   Plasma samples were analyzed by the Soma Scan multiplex proteomic
   profiling platform measuring 1305 protein analytes at SomaLogic Inc.
   (Boulder, CO) as described ([207]83). Briefly, test samples were
   incubated with a mixture of proprietary aptamer-based affinity reagents
   called SOMAmers. Two sequential bead-based immobilization and washing
   steps eliminated unbound or nonspecifically bound proteins and the
   unbound SOMAmers, leaving only protein target-bound SOMAmers. These
   remaining SOMAmers were isolated, and each reagent quantified
   simultaneously on a custom Agilent hybridization array.

Proximity extension assay (O-link) methodology

   Relative concentrations of 981 unique proteins (all analytes of 12
   O-link Proteomics panels: Cardiometabolic, Cell Regulation, CVD II, CVD
   III, Development, Immune Response, Immuno-Oncology, Inflammation,
   Metabolism, Neurology, Oncology, Organ Damage) were measured in the
   plasma samples by proximity extension assay (PEA) technology
   (Assarsson) at Olink Analysis Service (Watertown, MA). The PEA
   immunoassay relies on simultaneous dual target recognition by matched
   antibody pairs that are labeled by complementary oligonucleotide
   sequences. When both antibodies bind to the target protein, the
   oligonucleotide tags hybridize, and extension results in a DNA reporter
   that is quantified by real-time quantitative polymerase chain reaction.
   Protein levels are reported in NPX (Normalized Proteine eXpression),
   Olink’s arbitrary unit that is in log[2] scale ([208]84). Assay
   characteristics and validation data, like detection limit, precision,
   and validated sample types, are available from the Olink website:
   [209]www.olink.com/resources-support/document-download-center/#validati
   ondata.

mARG and IHC

   Brain penetrance of AKST1220 was measured using mARG followed by IHC at
   Pharmaron UK Limited. AKST1220 was radiolabled with ^14C with a
   specific activity of 58 mCi/mmol quantified by liquid scintillation
   counting. The radiochemical purity of the ^14C-AKST1220 was determined
   using radioHPLC and was ≥95% before use in the study. Male CD-1 mice
   were ordered from Charles River Laboratories and subjected to health
   examination before enrollment in the study. All animals were in the
   weight range of 27 to 29 g and approximately 8 to 12 weeks old at the
   time of dose administration.

   Five minutes before administration of ^14C-AKST1220, a single
   intravenous dose of biotinylated lectin (Vector Laboratories) was
   administered at 1.5 mg/kg. Mice received a single oral dose of
   ^14C-AKST1220 at 10 mg/kg (radioactive dose of 1 mCi/kg) in 45%
   HP-β-cyclodextrin (w/v; MilliporeSigma, product number H5784-10ML) in
   H[2]O and pH adjusted to pH 6 with 1 M NaOH. The required quantities of
   each radiolabeled test item were combined with the nonradiolabeled test
   item as appropriate to achieve a radioactive concentration. All dosing
   solutions were prepared fresh the same day as the dosing.

   Two male CD-1 mice each received a single oral administration of
   ^14C-AKST1220, following predosing with biotinylated lectin at a dose
   level detailed in section. An additional animal also received a dose of
   dose vehicle to act as a control for background radioactivity following
   a biotinylated lectin predose administration. Mice were euthanized at 2
   hours after dose, by overdose of carbon dioxide gas, followed by
   cervical dislocation. Immediately after sacrifice of each animal, the
   brain was removed and cut transversely into three separate portions to
   allow analysis of the forebrain, midbrain, and cerebellum. Each brain
   portion was rapidly frozen in isopentane and cooled to around −35°C
   with solid CO[2] in an appropriately sized plastic mold containing
   embedding medium. Blocked samples were stored in precooled, labeled
   bags at −80°C until required for sectioning. A frozen tissue block was
   removed from storage and attached to a precooled specimen disc using
   embedding medium. The resulting block was trimmed using a Leica CM3050S
   cryomicrotome, until a good cross section of tissue was observed. Under
   safe-light conditions, sections (5 μm in thickness) were transferred
   onto glass microscope slides that were precoated with nuclear emulsion.
   Sufficient sections were cut to allow processing of slides at different
   exposure intervals. Ten slides were prepared for each of four exposure
   time points (1, 2, 4, and 6 weeks). Slides were exposed in light-tight
   boxes at −20°C for the duration of the exposure period. Each box
   contained a desiccant to prevent accumulation of moisture. At each
   specified exposure time, slides were removed from storage and allowed
   to equilibrate to room temperature in the sealed box for approximately
   1 hour. Under safe-light conditions, slides were developed and fixed,
   after which time they were able to be handled under normal-light
   conditions.

   Following development for mARG, slides were rinsed in distilled water
   and fixed in 10% neutral buffered formalin for 3 min. Following
   washing, slides were incubated with a Vectastain Elite ABC kit for 45
   min. Slides were washed again and incubated with 3,3′-diaminobenzidene
   and substrate for 10 min. Slides were washed, stained with hematoxylin
   and eosin, and then coverslipped for imaging. Digital images were
   generated to show the localization of the radioactivity at cellular
   resolution. Separate sections collected onto uncoated microscope slides
   were exposed against phosphor-imaging plates for 7 days. At completion
   of the exposure period, imaging plates were scanned using an Amersham
   Typhoon IP Phosphor imager, and images were obtained to show the gross
   distribution of radioactivity within the brain sections. Slides
   processed for mARG were visually assessed using a Zeiss Axio Imager.A2
   light microscope. Representative digital images were captured using a
   Retiga 2000R CCD camera (QImaging) with associated Image-Pro Premier
   v9.1 software (MediCybernetics).

Statistical analysis

   Values represented in bar graphs are shown as means, and error bars
   represent SEM. Graphs were generated using GraphPad Prism. All
   statistical tests are represented in the figure legends along with n
   and P values. For n > 10 with normally distributed data, parametric
   tests were used; otherwise, nonparametric tests were used.

Acknowledgments