Abstract

Objective

   Fibroblast growth factor 21 (FGF21) analogs have been tested as
   potential therapeutics for substance use disorders. Prior research
   suggests that FGF21 administration might affect alcohol consumption and
   reward behaviors. Our recent report showed that plasma FGF21 levels
   were positively correlated with alcohol use in patients with alcohol
   use disorder (AUD). FGF21 has a short half-life (0.5–2 h) and crosses
   the blood–brain barrier. Therefore, we set out to identify molecular
   mechanisms for both the naïve form of FGF21 and a long-acting FGF21
   molecule (PF-05231023) in induced pluripotent stem cell (iPSC)-derived
   forebrain neurons.

Methods

   We performed RNA-seq in iPSC-derived forebrain neurons treated with
   naïve FGF21 or PF-05231023 at physiologically relevant concentrations.
   We obtained plasma levels of FGF21 and GABA from our previous AUD
   clinical trial (n = 442). We performed ELISA for FGF21 in both
   iPSC-derived forebrain neurons and forebrain organoids. We determined
   protein interactions using co-immunoprecipitation. Finally, we applied
   ChIP assays to confirm the occupancy of REST, EZH2 and H3K27me3 by
   FGF21 using iPSC-derived forebrain neurons with and without drug
   exposure.

Results

   We identified 4701 and 1956 differentially expressed genes in response
   to naïve FGF21 or PF-05231023, respectively (FDR < 0.05). Notably, 974
   differentially expressed genes overlapped between treatment with naïve
   FGF21 and PF-05231023. REST was the most important upstream regulator
   of differentially expressed genes. The GABAergic synapse pathway was
   the most significant pathway identified using the overlapping genes. We
   also observed a significant positive correlation between plasma FGF21
   and GABA concentrations in AUD patients. In parallel, FGF21 and
   PF-05231023 significantly induced GABA levels in iPSC-derived neurons.
   Finally, functional genomics studies showed a drug-dependent occupancy
   of REST, EZH2, and H3K27me3 in the promoter regions of genes involved
   in GABA catabolism which resulted in transcriptional repression.

Conclusions

   Our results highlight a significant role in the epigenetic regulation
   of genes involved in GABA catabolism related to FGF21 action.

   (The ClinicalTrials.gov Identifier: [47]NCT00662571)

   Keywords: FGF21, GABA catabolism, Alcohol use disorder

Highlights

     * •
       FGF21 analogs may have utility as therapeutics for substance use
       disorders.
     * •
       FGF21 was positively correlated with GABA levels in patients with
       AUD.
     * •
       FGF21 increased GABA levels in iPSC-derived neurons and brain
       organoids.
     * •
       FGF21 mediated the occupancy of REST, EZH2 and H3K27me3.
     * •
       The REST-EZH2-H3K37me3 complex mediates gene repression in a drug
       dependent manner.

Abbreviations

   FGF21
          Fibroblast growth factor 21

   AUD
          Alcohol use disorder

   iPSC
          Induced pluripotent stem cell

   GABA
          γ-aminobutyric acid

   EB
          Embryoid body

   RIN
          RNA integrity numbers

   FDR
          False discovery rate

   GSEA
          Gene Set Enrichment Analysis

   DAVID
          The database for Annotation, Visualization and Integrated
          Discovery

   TF
          Transcription factor

   ELISA
          Enzyme-linked immunosorbent assay

   co-IP
          Co-immunoprecipitation

   ChIP
          Chromatin immunoprecipitation

   REST
          Repressor element 1 silencing transcription factor

   TAF2
          TATA-box binding protein associated factor 2

   H3K27me3
          Tri-methylation of lysine 27 on histone H3 protein

1. Introduction

   Alcohol use disorder (AUD) is the most prevalent substance use disorder
   worldwide [[48]1]. To date, acamprosate, naltrexone, and disulfiram
   have received United States Food and Drug Administration approval for
   the treatment of AUD. However, growing evidence has prompted the
   development of fibroblast growth factor 21 (FGF21) based therapy as a
   potential treatment for AUD [[49][2], [50][3], [51][4]]. A major goal
   of AUD research is the development and optimization of effective drugs
   to treat AUD. Over the past three decades, clinical trials for AUD have
   been conducted on more than 30 different drugs [[52]5]. However, most
   clinical trials showed no significant impact or the effect size was
   small [[53]5]. This may be true, at least in part due to 1) the
   heterogeneity of AUD phenotypes, and 2) the critical knowledge gaps
   underlying the mechanisms of action of the medications used to treat
   AUD and the pathophysiology of AUD.

   FGF21 analogs have been tested in clinical trials for metabolic
   diseases, i.e. type 2 diabetes and non-alcoholic fatty liver disease,
   and they are generally well-tolerated [[54]6]. FGF21 is a peptide
   hormone that is primarily secreted by the liver, but it also crosses
   the blood–brain barrier [[55]7]. Prior literature suggests that FGF21
   administration could affect alcohol use and reward behaviors [[56]3].
   Our recent study showed that FGF21 levels were associated with recent
   alcohol consumption in patients with AUD [[57]8]. Specifically, we
   observed a positive correlation between plasma FGF21concentrations and
   recent alcohol consumption, ie the number of drinks per day in the past
   30 days and 90 days [[58]8]. Furthermore, we demonstrated that the
   effects of FGF21 on induced pluripotent stem cell (iPSC)-derived
   3D-brain organoids were associated, at least in part, with the
   regulation of catecholamines—neurotransmitters that are modulated by
   γ-aminobutyric acid (GABA) neurotransmission and which play a crucial
   role in reward pathways [[59]2,[60]8]. These observations potentially
   open the way to pharmacologic manipulation of FGF21, which in turn
   could have significant implications for the regulation of
   neurotransmitters and alcohol use. GABA is the primary inhibitory
   neurotransmitter in the brain, and it has been implicated in substance
   use disorders [[61]9]. Research has shown that changes in GABA
   concentrations in the prefrontal cortex can lead to deficits in
   decision-making and impulse control, which may contribute to the
   development of addiction [[62]10]. However, GABA may have a complex
   role in these pathways that is not fully understood at present [[63]9].
   Current in vitro models for studying drug action in the human brain are
   limited. Therefore, we used iPSC-derived neurons as a novel “cell-line
   based” approach to advance our understanding of drug action in the
   brain. Since the half-life of the naïve form of FGF21 is about 0.5–2 h
   [[64]11], it may not be suitable for clinical use due to poor
   pharmacokinetic and biophysical properties [[65]12]. Therefore, we
   designed the present study to investigate both the naïve form of FGF21
   and a long-acting FGF21 molecule (PF-05231023), which was created by
   conjugating two molecules of modified FGF21 (dHis/Ala129Cys) to an
   antibody scaffold [[66]13].

   We set out to identify molecular mechanisms for the naïve form of FGF21
   and a long-acting FGF21 molecule (PF-05231023) in iPSC-derived neurons.
   We began by determining the gene expression profiles of these two
   compounds in iPSC-derived forebrain neurons. We found remarkable
   overlap in both up-regulated and down-regulated genes between the naïve
   form of FGF21 and PF-05231023. We then further investigated these
   overlapping genes and identified pathways enriched for genes involved
   in GABA catabolism. Finally, we identified a possible epigenetic
   mechanism related to the regulation of GABA catabolism by FGF21. These
   studies have not only extended our original findings with regard to
   FGF21 and catecholamine metabolism [[67]8], but they have also revealed
   a novel mechanism related to a pharmacologic action of FGF21 in GABA
   catabolism.

2. Methods

2.1. Study subjects and ethics statements

   This study was conducted under protocols (20-00372 and 18-006428)
   reviewed and approved by the Mayo Clinic Institutional Review Board. We
   maintained confidentiality for all study participants. All participants
   whose samples were used in this study gave their consent for
   participation in the study and for publication of the study results in
   a peer-reviewed scientific journal.

2.2. Generation of iPSCs, iPSC-derived forebrain neurons, and 3-D brain
organoids

   Induced pluripotent stem cells (iPSCs) were generated from four
   subjects with alcohol use disorder (two males and two females), see
   [68]Supplementary Table 1. Specifically, we performed iPSC
   reprogramming using peripheral blood mononuclear cells (PBMCs) and the
   CytoTune™-iPS 2.0 Sendai Reprogramming Kit (A16517, Thermo Fisher, USA)
   as described previously [[69]14,[70]15]. All iPS cell lines showed
   normal karyotypes, and they all expressed pluripotency markers. Cell
   lines were regularly characterized and verified to be free from
   mycoplasma. We then generated iPSC-derived forebrain neurons as
   reported previously [[71]14,[72]16]. Briefly, iPS cells were cultured
   on Matrigel with mTeSR1 Plus media (STEMCELL technology, MA, USA). We
   suspended 3-D iPSC aggregates in embryoid body (EB) medium, consisting
   of FGF2-free iPS cell medium supplemented with 2 μM dorsomorphin
   (STEMCELL technology, MA, USA) and 2 μM A-83 (STEMCELL technology, MA,
   USA), in non-treated polystyrene plates for 6 days. EB medium was
   renewed every day. After 6 days, we replaced EB medium by neural
   induction medium (NPC medium) consisting of DMEM/F12, 1× N2 supplement,
   1× NEAA, 2 μg ml^−1 heparin (Tocris Bioscience) and 2 μM cyclopamine
   (Tocris Bioscience). The floating EBs were then transferred to
   Matrigel-coated plates at day 7 to form neural tube-like rosettes. The
   attached rosettes were kept for 15 days with NPC medium change every
   other day. On day 22, we transferred the rosettes to low attachment
   plates in NPC medium containing 1× B27 (Thermo Fisher). On day 24,
   neural progenitor spheres were dissociated and placed onto
   Matrigel-coated plates in neuron culture medium, consisting of
   Neurobasal medium supplemented with 1× glutamax, 1× B27, 0.2 mM
   l-Ascorbic Acid, and 0.5 mM cAMP (Sigma-Aldrich, St Louis, MO),
   10 ng/ml BDNF and 10 ng/ml GDNF. Medium was replaced every three days
   during continuous culture [[73]17]. The inhibitory/excitatory subtype
   classification was characterized by immunocytochemistry
   ([74]Supplementary Figure 1). Specifically, four weeks after
   differentiation from NPC, most neurons expressed a glutamatergic marker
   (VGLUT1). Approximately 5–10% of neurons are GABAergic neurons that
   express glutamate decarboxylase 1 (GAD1), and less than 1% of the
   neurons are dopaminergic neurons that express tyrosine hydroxylase
   (TH). The iPSC-derived neurons functionally mature four weeks after
   differentiation from NPCs. For example, they form synapses, fire action
   potentials, and have spontaneous synaptic activity [[75]17].
   Neurotransmitters including GABA were also detected in the iPSC-derived
   neurons [[76]18] ([77]Supplementary Figure 2A). Drug treatment for
   iPSC-derived forebrain neurons was initiated four weeks after
   differentiation from NPC.

   We also generated 3-D iPSC-derived forebrain organoids. Briefly, EBs
   were embedded in matrigel and cultured with 1× N2, 1× NEAA and 1×
   Glutamax (Invitrogen, Grand Island, NY), 1 μM SB431542, and 1 μM
   CHIR99021 (Selleckchem, Carlsbad, CA) for seven days. On day 14,
   organoids were mechanically dissociated from the matrigel and cultured
   in a bioreactor [[78]19]. Culture medium from days 14–70 consisted of
   DMEM/F12 medium supplemented with 1× N2, 1× B27, 1× NEAA and 1×
   Glutamax, 1× 2-metabptoethanol, 100× penicillin–streptomycin solution,
   and 2.5 μg/ml insulin (Sigma-Aldrich, St Louis, MO). The medium was
   changed every other day. From day 70 onward, supplementing media with
   20 ng/ml BDNF, 20 ng/ml GDNF (Peprotech, Rocky Hill, NJ), 0.2 mM
   l-Ascorbic Acid, and 0.5 mM cAMP (Sigma-Aldrich, St Louis, MO) was used
   [[79]20]. Neurotransmitters i.e. GABA, serotonin, and glutamate, were
   also detected in the iPSC-derived forebrain organoids [[80]18]
   ([81]Supplementary Figure 2B). Our recent single cell RNA-seq in
   iPSC-derived forebrain organoids suggested that the number of GABAergic
   cells is ∼28% of the total neuronal population at day 90 [[82]16]. Drug
   treatment for iPSC-derived forebrain organoids was initiated from day
   83 to day 90 [[83]16].

2.3. Drug treatment

   The naïve form of FGF21 (100 ng/ml, PeproTech, cat: 100-42) at a
   physiologically relevant concentration was used to perform the proposed
   experiments [[84]21]. Concentrations of the naïve form of FGF21
   (Peprotech, cat#100-42) [[85]22,[86]23] and PF-05231023
   (MedChemExpress, cat: HY-113697) [[87]24,[88]25] were selected to fall
   within the range of blood concentrations for these drugs observed
   during preclinical and clinical studies. We treated cells with either
   the naïve form of FGF21 or PF-05231023 for 7 days, with a daily medium
   change.

2.4. Immunostaining and confocal microscopy

   Cells were fixed in paraformaldehyde (4%) at room temperature for
   15 min. Cells were permeabilized with Triton X-100 (0.2%) in PBS. After
   blocking, cells were incubated with the primary antibody (see
   [89]Supplementary Table 1) in 5% BSA overnight. After a washing step,
   they were then incubated for 1 h at room temperature with secondary
   antibodies. Antifade mounting media with dapi (VECTOR laboratory,
   Burlingame, CA, USA) was used to stain the nuclei. Fluorescence
   microscopy (Olympus, FV1200) was used to visualize slides.

2.5. RNA sequencing and data analysis

   Total RNA was extracted using Trizol and the RNeasy mini kit (Qiagen,
   Valencia, CA, USA). The RNA integrity numbers (RIN) were above 9.5 for
   RNA samples. We performed RNA-seq with four AUD patients and each
   subject included three treatment conditions (vehicle, naïve FGF21 and
   PF-05231023) and two technical replicates. RNA-seq experiments were
   conducted by GENEWIZ using an Illumina HiSeq 6000 with eight samples in
   each lane using 100 bp paired end index reads ([90]Supplementary
   Table 2). Fastq files containing paired RNA-Seq reads were aligned with
   STAR [[91]26] against the UCSC human reference genome (hg19). RNA-seq
   differential expression analysis was performed using the DESeq2 package
   with default parameters [[92]27]. The Wald test was used to compare the
   beta estimate divided by its estimated standard error to a standard
   normal distribution to derive p-values. P values were then adjusted for
   multiple testing using the Benjamini and Hochberg method to produce
   false discovery rate (FDR) values. A false discovery rate less than
   0.05 was considered statistically significant. Gene Set Enrichment
   Analysis (GSEA) software [[93]28,[94]29], and the Database for
   Annotation, Visualization and Integrated Discovery (DAVID)
   [[95]30,[96]31] were used to perform pathway enrichment analysis. We
   then performed over-representation analysis [[97]32] for transcription
   regulatory targets of individual transcription factor (TF) using the
   Fisher's exact test method for selected gene lists i.e. the top 500 or
   200 genes against the transcription factor target gene sets.
   Specifically, target genes of transcription factors determined from
   transcription factor binding site profiles were downloaded from ENCODE
   [[98]33,[99]34], and MSigDB [[100]35]. The R function fisher.test with
   parameter alternative = “greater” was used to determine whether
   selected gene sets were significantly over-represented in the gene
   lists.

2.6. Enzyme-linked immunosorbent assay (ELISA) for GABA

   We measured GABA levels in the culture medium of iPSC-derived neurons
   and iPSC-derived brain organoids with the human GABA ELISA kit (Abcam:
   [101]AB287792) in accordance with the manufacturer's instructions.
   Specifically, we used 100 μL of culture medium for ELISA. We made an
   8-point standard concentration curve using the calibration standard
   solution to quantify GABA concentrations. ELISA was performed in
   duplicate and absorbances were read on a microplate reader at 450 nm.

2.7. Co-immunoprecipitation (co-IP) and Western blot analysis

   We used iPSC-derived neurons for co-IP. Cells were resuspended in
   500 μL IP lysis buffer containing 5 μL protease inhibitor cocktail and
   were incubated on ice for 30 min. Anti-REST (Proteintech, Rosemount,
   IL, USA) and anti-EZH2 antibodies and control IgG (Cell Signaling
   Technology, Danvers, MA, USA) were used to perform IP. Specifically, IP
   Samples containing protein A/G agarose beads (ThermoFisher, Pierce™
   Protein A agarose Beads, cat#:20333, USA) were rotated at 4 °C
   overnight. Immunoprecipitates were washed three times with ice cold
   lysis buffer. Protein samples were eluted with 100 μL 2× Laemmli
   loading buffer and boiled at 95 °C for 10 min. Proteins were separated
   on 4–12% SDS-PAGE gels and transferred onto PDVF membranes. After
   blocking, membranes were incubated with primary antibodies against REST
   or EZH2 at 4 °C overnight. The washed membrane was then incubated with
   secondary antibody for an hour at room temperature. The membranes were
   then incubated with super signal ECL substrate (Thermo Scientific,
   Madison, WI, USA) and were visualized with the Geldoc Go Gel imaging
   system (Bio-rad, USA). Antibodies for co-IP and Western blot analysis
   are listed in [102]Supplementary Table 1.

2.8. Chromatin immunoprecipitation (ChIP) assay

   ChIP assays were performed using iPSC-derived neurons and the MAGnify™
   Chromatin Immunoprecipitation System (ThermoFisher, cat#492024, USA).
   REST, EZH2, and H3K27me3 complexes were immunoprecipitated using
   antibodies against REST, EZH2, and H3K27me3, respectively. Control IgG
   was used (Cell Signaling Technology, Danvers, MA, USA) as a negative
   control. Real time PCR was used to quantify DNA-protein binding.
   Antibodies and primers for ChIP assays are listed in [103]Supplementary
   Table 1.

2.9. Statistical analysis

   Statistical analyses were performed using R Statistical Software
   (version 4.2.2; R Foundation for Statistical Computing, Vienna,
   Austria). The correlation between plasma FGF21 levels and plasma GABA
   levels was tested using the Spearman rho correlation coefficient. ELISA
   and ChIP assays were analyzed using ANOVA, followed by Tukey's multiple
   comparison tests for individual comparisons when significant effects
   were detected. P < 0.05 was considered statistically significant.

2.10. Data availability

   All data supporting our findings can be found in the main paper or in
   supplementary files. Sequencing data are available via the GEO
   accession number: [104]GSE223173.

3. Results

3.1. Gene expression profiles in iPSC-derived neurons treated with FGF21 or a
long-acting FGF21 molecule (PF-05231023)

   The present study investigated the molecular mechanisms of the naïve
   form of FGF21 and PF-05231023 using iPSC-derived forebrain neurons.
   Previous preclinical and clinical studies reported vital involvement of
   the prefrontal cortex in substance use disorders [[105]36,[106]37]. As
   a result, we generated iPSCs using peripheral blood mononuclear cells
   from four unrelated individuals (2 men and 2 women) and differentiated
   them into forebrain neurons ([107]Figure 1). We treated these
   iPSC-derived forebrain neurons with the naïve form of FGF21 (100 ng/ml)
   [[108]22,[109]23] and PF-05231023 (20 μM) [[110]24,[111]25] with
   concentrations falling within the range of blood concentrations for
   these drugs observed during preclinical and clinical studies. We
   identified 4701 and 1956 genes that displayed significant changes in
   expression after the cells were treated with either naïve FGF21 or
   PF-05231023, respectively (FDR < 0.05), as determined by RNA-seq
   ([112]Figure 2A). The effects of FGF21 on the central nervous system
   have been associated with the regulation of catecholamines
   [[113]2,[114]8]. In line with those observations, we identified a
   series of biological pathways i.e., neurotransmitter receptor activity
   and catecholamine secretion, that were the most common and most highly
   affected pathways in response to naïve FGF21 or PF-05231023
   ([115]Supplementary Tables 3 and 4). Furthermore, we found that a large
   number of differentially expressed genes in response to naïve FGF21 or
   PF-05231023 exposure overlapped ([116]Figure 2B). Notably, those
   overlapping genes displayed identical gene expression patterns after
   drug treatment ([117]Figure 2C and [118]Supplementary Table 5).
   Furthermore, when we used those overlapping genes for pathway analysis,
   we found that the GABAergic synapse pathway was the most significant
   biological pathway identified ([119]Supplementary Table 6). We then
   performed pathway analysis for up-regulated or down-regulated genes
   separately ([120]Supplementary Tables 7 and 8). Notably, the GABAergic
   synapse pathway was the most significant pathway affected by the
   down-regulated genes ([121]Supplementary Table 7). These results
   significantly expanded our original observations with regard to the
   role of FGF21 in catecholamine metabolism [[122]8], and served to
   highlight a possible novel mechanism of FGF21 or the long-acting FGF21
   molecule (PF-05231023). Therefore, the next series of studies was
   designed to 1) determine the association of FGF21 and GABA in the
   plasma samples from patients with AUD, and 2) to identify possible
   molecular mechanisms related to this relationship.

Figure 1.

   [123]Figure 1
   [124]Open in a new tab

   Generation of iPSC-derived neurons. (A) iPSC generation using
   peripheral blood mononuclear cells (PBMC). All iPSC cell lines
   displayed normal karyotypes. The panel below the schematic displays
   representative examples of staining for iPSC pluripotency markers:
   SOX2, and TRA-1-81. All iPSC lines were positive for pluripotency
   markers. (B) A schematic outline of procedures used during the
   differentiation of iPSC-derived forebrain neurons. The panel below the
   schematic displays representative examples of staining for TUJ1 and
   MAP1/2.

Figure 2.

   [125]Figure 2
   [126]Open in a new tab

   Bulk RNA-seq using iPSC-derived forebrain neurons treated with FGF21,
   PF-05231023 and vehicle. (A) Volcano plots indicate differentially
   expressed genes with FDR 0.05. (B) Venn diagram showing expression
   profiles for the 974 genes with expression that could be affected by
   both FGF21 and PF-05231023 as determined by RNA-seq (FDR < 0.05). The
   middle and bottom Venn diagrams illustrate overlapping up-regulated and
   down-regulated genes in response to FGF21 and the long-acting FGF21
   molecule. Those results indicate that those overlapping genes displayed
   identical directionality. (C) Heatmap showing expression profiles for
   the 974 genes affected by FGF21 and PF-05231023.

3.2. FGF21 was positively correlated with GABA concentrations in patients
with AUD

   We previously conducted a clinical trial with a total of 442 patients
   with AUD. FGF21 and GABA were assayed using plasma samples as described
   previously [[127]8,[128]20]. We collected plasma samples from AUD
   patients prior to acamprosate treatment. We observed a significant
   positive correlation of FGF21 and GABA concentrations in the plasma
   samples of patients with AUD (r: 0.13, p: 0.008) ([129]Figure 3A).
   Since FGF21 can cross the blood brain barrier [[130]7], this led us to
   test whether FGF21 might modulate levels of GABA in iPSC-derived brain
   cells. As anticipated, the concentrations of GABA in the culture medium
   increased significantly in response to FGF21 treatment
   ([131]Figure 3B). In line with this finding, when iPSC-derived
   forebrain neurons were exposed to PF-05231023, the concentrations of
   GABA in the culture medium also increased substantially
   ([132]Figure 3B). We next generated iPSC-derived forebrain organoids to
   verify those results ([133]Figure 4A). Brain organoids derived from
   iPSCs are three-dimensional self-assembled structures containing
   multiple brain cell types such as neurons, astrocytes, and microglia
   [[134]38,[135]39], which can mimic human brain structures and which
   might help to bridge the gap between preclinical and clinical trials.
   Consistently, the levels of GABA in the culture medium of iPSC-derived
   forebrain organoids increased significantly in response to either naïve
   FGF21 or PF-05231023 treatment ([136]Figure 4B). Our findings suggest
   that both naïve FGF21 and PF-05231023 might play a role in the
   regulation of GABA catabolism. Therefore, we designed the next series
   of studies to determine the molecular mechanism mediating this
   relationship.

Figure 3.

   [137]Figure 3
   [138]Open in a new tab

   Plasma FGF21 concentrations were positively correlated with plasma GABA
   concentrations. (A) Plasma FGF21 concentrations were associated with
   plasma GABA levels in patients with alcohol use disorder. (B) both
   FGF21, and PF-05231023 could substantially increase the level of GABA
   in the culture medium of iPSC-derived forebrain neurons.

Figure 4.

   [139]Figure 4
   [140]Open in a new tab

   FGF21 and PF-05231023 induced GABA concentrations in iPSC-derived
   forebrain organoids. (A) A schematic outline of procedures used during
   the differentiation of iPSC-derived forebrain organoids. The middle
   panel displays images of iPSC-derived forebrain organoids during
   differentiation. The bottom panel of the schematic displays
   representative examples of staining for neural progenitors (SOX2) and
   neurons (TUJ1). (B) FGF21 and PF-05231023 substantially induced the
   concentrations of GABA in the culture medium of iPSC-derived forebrain
   organoids.

3.3. REST, EZH2 and H3K27me3 co-occupied the promoter regions of genes
associated with GABA catabolism

   We performed upstream regulator analysis using selected gene sets,
   i.e., the threshold of the top 500 down-regulated or up-regulated genes
   in the presence of naïve FGF21 or PF-05231023 ([141]Figure 2A)
   [[142]32]. That analysis allowed us to identify the cascade of upstream
   transcription regulators that might contribute to gene expression
   pattern changes in response to drug treatment. Repressor element 1
   silencing transcription factor (REST) was predicted to be the most
   important upstream regulator of the genes down-regulated in response to
   naïve FGF21 or PF-05231023 treatment with p values of 9.14E-36 and
   1.04E-47 respectively ([143]Supplementary Table 9). This trend remained
   when we set the threshold of the top 200 down-regulated genes
   ([144]Supplementary Table 9). In contrast, TATA-box binding protein
   associated factor 2 (TAF2) was predicted to be the most important
   upstream regulator of genes up-regulated in response to FGF21 or
   PF-05231023 treatment, with p values of 8.21E-17 and 6.06E-09,
   respectively ([145]Supplementary Table 9). Notably, 14 out of 15 genes
   that were associated with the GABAergic synapse pathway
   ([146]Supplementary Table 6) were down-regulated in the presence of
   naïve FGF21 or PF-05231023 as determined by RNA-seq ([147]Figure 5A).
   This observation led us to ask whether REST might play a role in
   transcription regulation for the 15 genes associated with GABA
   catabolism. We began by exploring the ENCODE database and found that
   REST could bind to the promoter regions of all of those genes except
   GABRA3 and CACNA1C. As a result, we performed ChIP assays to confirm
   the occupancy of REST on the promotors of those genes in iPSC-derived
   forebrain neurons ([148]Figure 5B). Previous studies have shown that
   EZH2 is a co-factor for REST [[149]40]. Of importance, EZH2 was one of
   the top upstream regulators of genes that were down-regulated in the
   presence of naïve FGF21 and PF-05231023 ([150]Supplementary Table 9).
   We also performed co-IP to demonstrate whether REST could physically
   interact with EZH2 in iPSC-derived neurons ([151]Figure 5C). EZH2 is a
   catalytic subunit of polycomb repressive complex 2 (PRC2) which
   catalyzes the tri-methylation of lysine 27 on the histone H3 protein
   (H3K27me3) to regulate gene expression through epigenetic machinery
   [[152]41]. Specifically, H3K27me3 is a transcriptionally repressive
   histone mark, and it can function as a silencer to repress gene
   expression [[153]42]. We then performed ChIP assays to show that REST,
   EZH2 and H3K27me3 were able to bind to the same DNA regions in
   iPSC-derived forebrain neurons ([154]Figure 5B). These results
   confirmed the co-occupancy of REST, EZH2 and H3K27me3. However, they
   did not explain drug-induced inhibition of gene expression when the
   cells were treated with FGF21 and PF-05231023 ([155]Figure 5A).

Figure 5.

   [156]Figure 5
   [157]Open in a new tab

   REST, EZH2 and H3K27me3 co-occupied the promoter regions of genes
   associated with GABA catabolism. (A) Heatmap showing expression
   profiles for the 15 genes involved in the GABAergic synapse pathway.
   (B) ChIP assays showing that REST, EZH2 and H3K27me3 co-occupied the
   promoter regions of the 13 selected genes as shown in [158]Figure 5A
   that were associated with the GABAergic synapse pathway in response to
   FGF21 and PF-05231023 treatment of iPSC-derived forebrain neurons
   (n = 4). Percentages of ChIP DNA/input were determined by qPCR. Data
   are represented as % input, (enrichment relative to IgG control) = %
   input (antibody against REST, EZH2 and H3K27me3) – % input (IgG).
   ChIP-qPCR was performed in triplicate. Three independent experiments
   were performed. This figure represents results from one of the three
   experiments. (C) Co-Immunoprecipitation (IP) was used to determine
   whether REST protein could interact with EZH2 in iPSC-derived forebrain
   neurons. Whole cell lysates from iPSC-derived forebrain neurons were
   immunoprecipitated with anti-REST antibodies or anti-IgG antibodies and
   protein samples were immunoblotted and probed with antibodies against
   REST or EZH2.

3.4. REST-EZH2-H3K37me3 complex mediated gene repression in a drug-dependent
fashion

   It is well-documented that transcription factors (TF) bind to specific
   DNA binding motifs to regulate gene expression in a drug-dependent
   fashion [[159][43], [160][44], [161][45]]. We next performed ChIP
   assays using iPSC-derived forebrain neurons after exposure to naïve
   FGF21 or PF-05231023. When cells were treated with naïve FGF21 or
   PF-05231023, the vast majority of the genes tested, including ABAT, a
   gene encoding a key enzyme responsible for GABA catabolism, GRBBR1
   (GABA B receptor 1), and SLC32A1 (a GABA transporter), displayed
   significantly increased REST binding ([162]Figure 6). As mentioned
   previously, REST is also known as neuron-restrictive silencer factor, a
   protein that is encoded by the REST gene and which acts as a
   transcriptional repressor. Thus, changes in downstream gene expression
   could be mediated by the binding affinity of REST [[163]46]. We had
   demonstrated that REST co-exists with EZH2 and H3K27me3 in the target
   promoters in iPSC-derived forebrain neurons ([164]Figure 5B). EZH2 is
   the catalytic unit of the polycomb repressive complex 2 (PRC2) with
   methyltransferase activity. PRC2 is essential for transcription
   silencing by specifically catalyzing the formation of H3K27me3, a
   marker of transcriptional repression [[165]42]. We next attempted to
   determine whether the recruitment of EZH2 and H3K27me3 could be altered
   in a drug-dependent manner. Those results might explain repression of
   the list of genes involved in GABA catabolism, as shown in
   [166]Figure 5A. As anticipated, for the same genes (see [167]Figure 6),
   the recruitment of EZH2 and H3K27me3 was significantly induced by naïve
   FGF21 or PF-05231023.

Figure 6.

   [168]Figure 6
   [169]Open in a new tab

   REST-EZH2-H3K37me3 complex-mediated gene repression in a drug dependent
   fashion. ChIP assays were performed in iPSC-derived forebrain neurons
   treated with FGF21, PF-05231023 and vehicle. We confirmed that changes
   in REST binding were correlated with changes in mRNA expression for all
   13 genes. Specifically, in the presence of FGF21, all 13 binding sites
   except GNAI1 displayed significantly increased REST binding. In
   parallel, the recruitment of EZH2 and H3K27me3 also increased
   substantially, thus confirming that changes in REST-EZH2-H3K37me3
   occupancy were highly correlated with transcription.

   Taken together, our results showed that FGF21 could increase GABA
   concentrations in both iPSC-derived forebrain neurons and forebrain
   organoids ([170]Figure 3, [171]Figure 4B). GABA is synthesized
   primarily from glutamate by glutamate decarboxylases, GAD1 and GAD2, in
   presynaptic neurons. The GABA synthesized is loaded into synaptic
   vesicles via a vesicular GABA transporter (VGAT also known as SLC32A1).
   GABA activity is terminated at the synapse by uptake into both nerve
   terminals and surrounding cells via plasma member GABA transporters.
   GABA is then catabolized by GABA transaminase (ABAT) [[172]47]. Most of
   the genes related to GABA catabolism can be suppressed by naïve FGF21
   or PF-05231023. This might be, at least in part, because FGF21 and
   PF-05231023 can facilitate the recruitment of REST-EZH2-H3K37me3 to the
   promotors of target genes, which in turn, represses gene expression
   ([173]Figure 7).

Figure 7.

   [174]Figure 7
   [175]Open in a new tab

   Schematic diagram illustrating the effects of FGF21 on GABA catabolism.
   Specifically, FGF21 and PF-05231023 significantly induced GABA levels
   in iPSC-derived forebrain neurons. We identified a drug-dependent
   occupancy of REST, EZH2, H3K27me3 in the promoter regions of genes
   associated with GABA catabolism i.e., ABAT (a key enzyme responsible
   for GABA catabolism), GRBBR1 (GABA B receptor 1), and SLC32A1 (GABA
   transporter), which resulted in transcriptional repression.

4. Discussion

   Substance use disorders such as AUD are complex and multifaceted
   disorders. Current FDA-approved treatment options are limited and have
   limited efficacy [[176]48]. FGF21 has been identified as a potential
   therapeutic target for substance use disorders due to its ability to
   regulate reward pathways in the brain [177][3], [178][49]. As a result,
   studying FGF21's mechanism of action is critical as it potentially
   opens a new avenue for developing more effective and safer treatments
   for addiction. Like most neuropsychiatric disorders, addiction is
   highly heterogeneous. This makes it challenging to develop models or
   tools that can capture the full range of individual differences in
   addiction. However, breakthroughs in iPSC technology offer a powerful
   research tool for studying neuropsychiatric disorders and can
   potentially revolutionize the field of neuroscience by providing new
   insights into disease mechanisms and by facilitating the development of
   new drugs.

   Our recent study demonstrated that plasma FGF21 levels were positively
   associated with recent alcohol use in patients with AUD, and that
   ethanol can induce the concentrations of FGF21 and catecholamines in
   iPSC-derived brain organoids [[179]8]. FGF21 has been suggested as a
   potential therapeutic target for the treatment of addictions. The role
   of FGF21 in metabolism has been studied extensively, however, molecular
   mechanism(s) underlying FGF21's mechanism of action in the brain remain
   unclear. As a result, the present study was designed to investigate the
   molecular mechanisms of naïve FGF21 and the long-acting FGF21 compound
   (PF-05231023) using iPSC-derived forebrain neurons, and the omics data
   from our previous AUD clinical study.

   The present study identified a large number of differentially expressed
   genes in response to exposure to naïve FGF21 or PF-05231023.
   Specifically, 974 differentially expressed genes overlapped between
   treatment with naïve FGF21 and with PF-05231023. REST was predicted as
   the most significant upstream regulator of differentially expressed
   genes in the presence of naïve FGF21 or PF-05231023. The GABAergic
   synapse pathway was the most significant biological pathway identified
   using those overlapping genes. GABA is the most common inhibitory
   neurotransmitter, and it plays a significant role in the development of
   addiction [[180]50]. We observed a significant positive correlation
   between plasma FGF21 and GABA concentrations in AUD patients. However,
   the effect size is small which might be due, at least in part, to
   heterogeneity of alcohol use disorder and biological variation among
   patients suffering from this disorder. In parallel, naïve FGF21 and
   PF-05231023 significantly induced GABA levels in iPSC-derived forebrain
   neurons and forebrain organoids. Opoku et al. reported that FGF21
   regulates GABA concentrations in the brain, liver, and serum of mice
   with hepatic encephalopathy [[181]51]. However, it is not fully
   understood how FGF21 regulates GABA.

   We observed that a series of genes related to GABA catabolism were
   down-regulated in response to naïve FGF21 or PF-05231023
   ([182]Figure 5A). We also demonstrated a drug-dependent occupancy of
   REST, EZH2, and H3K27me3 in the promoter regions of genes associated
   with GABA catabolism, which resulted in transcriptional repression in
   iPSC-derived forebrain neurons ([183]Figure 6). GABA regulation has
   been associated with histone monomethylation or trimethylation, and
   histone demethylation [[184]52]. Chromatin remodeling at GABAergic gene
   promoters, including GAD1, and GAD2 histone methylation, has been
   observed throughout an extended period of normal human pre-frontal
   cortex development, and plays a role in the neurobiology of
   schizophrenia [[185]53]. H3K27me3, a repressive chromatin marker, was
   present at these promoter sites [[186]53]. This is consistent with our
   findings ([187]Figure 6). Furthermore, our results demonstrate that
   FGF21 mediated the occupancy of REST, EZH2 and H3K27me3. As expected,
   we observed a negative correlation between mRNA expression of GABAergic
   gene promoters and the occupancy of the REST-EZH2-H3K37me3 complex in
   iPSC-derived forebrain neurons ([188]Figure 5, [189]Figure 6). These
   data, taken together, support our hypothesis that the drug-dependent
   occupancy of the REST-EZH2-H3K37me3 complex could explain the
   difference in gene expression of GABAergic genes in response to drug
   treatment. Collectively, our findings suggest that a previously
   uncharacterized epigenetic mechanism regulates GABAergic gene promoters
   by FGF21. These results also demonstrate that iPSC-derived brain-like
   cells could help to generate testable hypotheses for functional
   mechanistic studies, and, ultimately, add mechanistic rigor to future
   clinical trials of AUD therapeutics.

   We demonstrated that FGF21 increases GABA production by
   REST-EZH2-H3K27me3 mediated transcriptional repression of genes
   associated with GABA catabolism. GAD2 is responsible for catalyzing the
   production of GABA ([190]Figure 7). Our results showed that GABA
   concentrations were induced by FGF21, however, mRNA expression of GAD2
   was repressed by FGF21. We also observed that ABAT, a crucial enzyme
   responsible for GABA catabolism, and SLC32A1 (an influx transporter for
   GABA), also displayed decreased expression in response to FGF21
   treatment. As a result, we observed an increase of GABA levels within
   the culture medium of iPSC-derived neurons and brain organoids as a
   result of decreased degradation in the face of decreased biosynthesis.
   These results may suggest a potential negative feedback loop because of
   the induction of GABA by FGF21.

   Our study also has limitations. We began by performing RNA-seq using
   iPSC-derived forebrain neurons. Our sample size is small, which limits
   the generalizability of our findings. The human brain is complex, and
   one cell type in a specific brain region may not represent the complex
   human central nervous system. Future studies that include different
   brain regions will be required to pursue the results reported here, as
   well as the application of single-cell sequencing in iPSC-derived brain
   organoids in an attempt to identify cell-type specific, and
   drug-specific effects. In addition, further study is also warranted to
   explore the links between GABAergic chromatin remodeling and
   electrophysiological changes in response to naïve FGF21 and
   PF-05231023. Those results could provide additional functional insights
   into both of the compounds studied. Despite these limitations, our
   study provides novel mechanistic insights into the drug actions of
   naïve FGF21 and PF-05231023.

5. Conclusions

   In summary, this study demonstrates a novel mechanism of FGF21. Our
   results show that FGF21 can induce GABA concentrations in both
   iPSC-derived forebrain neurons and forebrain organoids. GABA is
   synthesized primarily from glutamate. GABA is then stored in synaptic
   vesicles via a vesicular GABA transporter (SLC32A1). GABA has complex
   regulatory functions for monoamine neurotransmission. GABA activity is
   terminated at synapses by uptake into both nerve terminals and
   surrounding cells via plasma member GABA transporters. GABA is then
   catabolized by GABA transaminase (ABAT) [[191]47]. Those genes related
   to GABA catabolism can be suppressed by naïve FGF21 or PF-05231023.
   This might be, at least in part, because FGF21 and PF-05231023 can
   facilitate the recruitment of REST-EZH2-H3K37me3 to GABAergic gene
   promoters, which, in turn, represses gene expression. Taken together,
   these results may provide novel mechanistic insights into FGF21 action
   in the CNS.

Funding

   This work was supported in part by National Institutes of Health [grant
   numbers R01 AA27486, P20 AA17830, DA57928, and K01 AA28050]; Brain &
   Behavior Research Foundation, the Mayo Clinic Research Pipeline K2R
   Program, the Terrance and Bette Noble Foundation, and the Mayo Clinic
   Center for Individualized Medicine. The content of this manuscript is
   solely the responsibility of the authors and does not necessarily
   represent the official views of the National Institutes of Health.

Author contributions

   MH wrote the first draft of the manuscript and supervised the study. MH
   designed the research; MH, CZ, and IM performed the research. MH, and
   CZ analyzed the data; CZ, BC. JB, and HL contributed analytical tools.
   TO, MS, PC, VK, QN, CS, and MS recruited study subjects, and provided
   administrative support. MH and RW obtained funding. All authors have
   given final approval of the version to be published.

Acknowledgments