SMUG1 DNA glycosylase modulates reward behaviour through regulation of olfactory receptor gene expression

Abstract

The SMUG1 DNA glycosylase is the primary enzyme responsible for excising 5-hydroxymethyluracil (5hmU) from genomic DNA. SMUG1 activity is high in the brain, suggesting its crucial role in controlling 5hmU levels in this organ. 5hmU has been proposed as an epigenetic mark, but it is still unclear whether, and in which contexts, it may have regulatory functions. Here we show that accumulation of 5hmU in Smug1-deficient mice leads to widespread gene expression alterations across multiple brain regions. Chromatin immunoprecipitation revealed that SMUG1 is recruited to AT-rich promoters of olfactory receptor genes, where it regulates 5hmU levels. The persistence of 5hmU in the olfactory receptor promoters leads to deficits in sensory perception of smell and influences reward-related behaviour. Thus, we identify a specific function for SMUG1 in promoting transcription of olfactory receptor genes.

Graphical Abstract

Graphical Abstract.

Introduction

DNA bases undergo continuous modifications induced by naturally occurring or exogenous metabolites, leading to changed chemical properties. While previously seen purely as DNA damage, there is a growing recognition that certain modified DNA bases can be introduced in a programmed manner as epigenetic modifications that add an extra layer of gene regulation.

Modified bases are selectively recognized by DNA glycosylases that initiate the base excision repair (BER) pathway that reinstalls a cognate base pair. In addition, the BER pathway may be employed to induce programmed DNA strand breaks. For example, thymine-DNA glycosylase (TDG)-initiated BER activates cell-fate specification programmes during neuronal and macrophage development by generating single-strand breaks at enhancers containing TET-mediated oxidized methylcytosines [1]. Programmed DNA breaks are also introduced in several types of neurons, e.g. during learning [2] and cell specification [3, 4].

The elusive base 5-hydroxymethyluracil (5hmU) is another base that may have different functions in cells depending on the context. 5hmU can be incorporated from modified 5hmdUTP during replication [5] or induced by oxidation of thymine, e.g. by ionizing radiation [6]. Additionally, two routes of enzymatically generated 5hmU have been proposed: either through the direct hydroxylation of thymine by TET enzymes [7, 8] or via active deamination of 5-hydroxymethylcytosine (5hmC) by activation-induced cytidine deaminase or apolipoprotein B messenger RNA (mRNA)-editing catalytic polypeptide-like (APOBEC) family enzymes [9, 10].

The possibility that 5hmU may act as an epigenetic mark was suggested based on the identification of several proteins binding preferentially to 5hmU-containing DNA [7]. While limited data exists on 5hmU dynamics, the available literature suggests that the levels fluctuate based on biological context and developmental stage. In mouse embryonic stem cells, 5hmU was detected at high levels and believed to arise from TET-mediated oxidation of thymine [7]. As mouse embryonic stem cells (mESCs) differentiate into somatic tissues, 5hmU levels decrease [7], which may be indicative of active regulation.

The SMUG1 DNA glycosylase is the main enzyme responsible for excising 5hmU from genomic DNA. Smug1^−/−mice are viable and appear to develop normally [11]. Smug1^−/− mice are not prone to tumour development but show a tendency to develop dysmetabolic phenotypes late in life [12]. Although SMUG1 is recognized for its action on various substrates, such as uracil and formyl uracil, its significance is particularly evident in its role regulating genomic 5hmU levels [11]. SMUG1 activity is high in the brain, where Smug1^−/− deficient mice exhibit more than a 20-fold increase in 5hmU levels compared to isogenic WT mice. While previous morphological examinations did not reveal gross abnormalities in SMUG1-deficient brains [11, 13], targeted analyses are required to determine the functional implications of elevated activity of SMUG1 in the brain.

Here, we describe a widespread transcriptional rewiring occurring across multiple brain regions in Smug1^−/− mice. This rewiring involves extensive downregulation of olfactory receptor (OR) genes coinciding with impaired sensory perception of smell. Chromatin immunoprecipitation (ChIP) revealed that SMUG1 is recruited to AT-rich promoters of the affected OR genes, where it regulates 5hmU levels. Further, we show that SMUG1 is required for formation of physiological DNA strand breaks in the olfactory bulb (OB). In summary, these data provide mechanistic evidence for a specialized role of SMUG1 in modulating 5hmU levels within AT-rich environments of OR genes, thus serving as transcriptional regulator fundamental for brain function.

Materials and methods

Mice

C57BL/6 Smug1^−/− mice were generated as previously reported [11]. C57BL/6 WT was used as a control. Animals were maintained at a regimen of 12 light/12 dark cycle given free access to food and water. All the mice included in the experiments were adult males from 6 to 9 months old. All animal experiments were approved by the Norwegian Animal Research Authority and conducted in accordance with the laws and regulations controlling experimental procedures in live animals in Norway and the European Union’s Directive 86/609/EEC.

Brain isolation and tissue processing

For analysis that required DNA and RNA material, adult male mice were sacrificed by cervical dislocation and had the brain immediately removed. The following brain regions were dissected on a cold and clean surface: OB, prefrontal cortex (PFC), striatum (STR), hippocampus (HPC), and ventral midbrain (vMB). Tissues were snapped frozen in liquid nitrogen and kept at −80°C until use. For samples used for histological examination, mice were deeply anesthetized with isoflurane and transcardially perfused with 10% formalin in phosphate buffered saline (PBS). Brains were removed and kept in 10% formalin for 24 h for optimal fixation before paraffin embedding. Sagittal brain sections of 4 μm were used for all related experiments.

Behavioural studies

Open field

The open field (OF) test was conducted to observe spontaneous activity and performed as previously reported [14]. The mice (WT n = 17, Smug1^−/−n = 16) were allowed to freely explore an arena measuring H35 cm × L40 cm × W40 cm with white walls and floor for 45 min. Movement of the mice was tracked and analysed by ANY-maze video tracking software (Stoelting, IL, USA).

Elevated zero maze

The elevated zero maze task was used to study anxiety-related behaviour. Mice (WT n = 17, Smug1^−/−n = 16) were tested in a custom-made apparatus as previously described [14]. The animals were placed on the maze facing a closed area and allowed to explore the space for 5 min. Movement of the mice were tracked and analysed by ANY-maze video tracking software (Stoelting, IL, USA).

T-maze task-rewarded alternation

To measure spatial working memory and reward, we conducted the rewarded alternation task in a T-maze apparatus. The procedure was performed according to the method described by Deacon and Rawlins, 2006 [15] with some modifications. After habituation and training sessions with the reward (50% condensed sweetened milk), mice (WT n = 10, Smug1^−/−n = 9) were evaluated in three trials per day during four consecutive days. After a pause of 2 days, the trials resumed for another three consecutive days. The pause between trials intended to evaluate memory retention capacity of mice.

Buried food test

Mice (WT n = 17, Smug1^−/−n = 14) were food deprived for 16 h before the test. The animals were first habituated to a new cage with new bedding for 5 min. After this period, the mice were removed from the cage, and a food pellet was randomly hidden under the bedding before introducing the mice to the cage again. Latency to both finding the food and starting to eat the food was recorded manually by the observer.

RNA and DNA isolation

Total RNA and DNA were isolated from different brain regions using the ALLPrep isolation kit (Qiagen). Before extraction, frozen tissues were transferred to a 2 ml lysing tube containing 1.4 mm zirconium oxide beads (Precellys) and 600 μl of RLT buffer (provided by the Kit). Samples were homogenized using FastPrep (20 s, 5000 rpm) and centrifuged at full speed for 5 min. The supernatant was transferred to an ALLPrep spin column, and isolation proceeded according to the manufacturer’s protocol. To improve the quality of the RNA, β-mercapto-ethanol was added to the RLT buffer at the recommended concentration, and samples were further treated with DNase I to digest contaminating DNA. RNA quality and concentration were measured using Agilent Bioanalyser (Agilent Technologies) or Nanodrop. DNA parameters were acquired with Nanodrop.

RNA sequencing and bioinformatics

RNA samples from brain regions (OB, PFC, STR, HPC, and vMB) were isolated from 3 WT and 3 Smug1^−/− 6 months old male mice. Library construction and sequencing (paired-end sequencing with read length = 150) were performed by Novogene Ltd using the Illumina Hiseq platform. Alignment was done using HISAT2 [16] with the GRCm38 as the reference genome and Ensembl release 81 [17] for the gene annotation. Resulting SAM files were sorted and converted to BAM files using SAMtools [18]. StringTie [19] was used to assemble transcripts and quantify them into counts. Bioinformatic analyses were done using R version 4.1.0. Differential gene expression was performed using the DESeq2 R package [20], version 1.32.0. The standard DESeq2 pipeline is used to obtain the differential gene expression results. Selection of significant differentially expressed genes is made based on a P value below 0.05 and a log₂ fold change value >1 or <−1. The visEAGO package [21], version 1.6.0 was used to perform a functional enrichment and comparison of results for the differential gene expression experiments performed. The package requires that genes be in their entrezID format. These are obtained by using the mapIds function from the annotation Dbi package, version 1.54.1. The library used for the conversion is org.Mm.eg.d version 1.13.0. Genes were split into two categories, significantly upregulated and significantly downregulated, these groups were submitted for functional enrichment analysis of biological processes using visEAGO.

Digital droplet PCR

RNA isolated from OB and PFC of 6 months old mice (n = 3 per genotype) was retro-transcribed using the high-capacity complementary DNA (cDNA) reverse transcription kit (Life Technologies) according to the manufacturers’ instructions. Brain regions from three mice of the same genotype were pooled and subjected to pre-amplification using the TaqMan® PreAmp Master Mix (Life Technologies) following the manufacturers’ instructions. The pre-amplified cDNA was then assayed via digital droplet PCR (ddPCR) using Droplet Digital PCR QX system (Bio-Rad). Briefly, the pre-amplified cDNA was added to a 20 μl PCR mixture containing 10 μl 2× QX200 ddPCR Supermix for Probes (No dUTP) (Bio-Rad), 1 μl of TaqMan® MGB probes labeled with FAM for Olfr102, Olfr597, Olfr872, Olfr1115, Olfr524, Olfr727, Olfr1206, Olfr1509, Olfr527, and Olfr1000 (Life Technologies), and 1 μl of TaqMan® MGB probes labeled with VIC for Gapdh (Life Technologies) (Supplementary Table S1). Twenty microliter of PCR mixture and 70 μl Droplet generation oil for Probe (Bio-Rad) were mixed. Droplets were generated using a QX100 Droplet Generator (Bio-Rad). The following PCR conditions were used: after enzyme activation at 95°C for 10 min, 40 cycles at 94°C for 30 s, and 60°C for 1 min were followed by one step at 98°C for 10 min. Reactions were read in the QX200 Droplet Reader (Bio-Rad). Results are expressed as relative mRNA level to Gapdh.

LC-MS/MS quantification of DNA base modifications

The following DNA base modifications were quantified from WT and Smug1^−/− (n = 5 per genotype) brain regions: 5hm(dU), 5m(dC), 5hm(dC), and 8-oxo(dG). For quantification, 2 μg of genomic DNA was enzymatically hydrolysed to DNA nucleosides and analysed by LC-MS/MS as previously described [11]. Prior hydrolysation, RNA and free nucleotides were removed as reported.

AT content and motif analyses in OR promoters

DNA sequences surrounding ±1 Kb to transcription start site (TSS) of significantly downregulated OR genes were extracted from the mm10 mouse genome. The CG content of OR promoters was performed according to the method described in Mikkelsen et al. [22]. The AT content was extrapolated from the relative amount of CG detected in the promoters. An in-house python script was made to search for the motif AACTTTTTAATGA in both forward and reverse strands of the promoter (−500 bp) [23]. A mismatch of three base pairs was allowed and the total counts of the regulatory sequency included both forward and reverse strands.

Chromatin immunoprecipitation

ChIP was performed in the OB and PFC of 6 months old mice. Brain regions from three mice of the same genotype were pooled, homogenized, and subsequently washed three times in ice-cold PBS. Homogenized material was then suspended in 500 μl of 1% formaldehyde in PBS and cross-linked for 10 min at RT. The cross-linking reaction was quenched by adding 125 mM glycine, followed by a 5 min incubation at RT. After three washes in cold PBS, cells were resuspended in 300 μl freshly prepared sodium dodecyl sulphate (SDS) lysis buffer [50 mM Tris–HCl, pH 8, 10 mM ethylenediaminetetraacetic acid (EDTA), pH 8, 1% SDS] supplemented with protease inhibitors and incubated for 20 min on ice. Chromatin was sheared to 200–250 bp DNA fragments by sonication using Bioruptor (Diagenode) for 25 cycles, with each cycle consisting of 30 s of sonication followed by 30 s of rest, at 4°C. Chromatin was pre-cleared with Dynabeads protein G (Thermo Fisher) on rotation for 2 h at 4°C in ChIP buffer (16.7 mM Tris–HCl, pH 8, 167 mM NaCl, 1.2 mM EDTA, 0.01% SDS, and 1.1% Triton X-100). Twenty micrograms of precleared chromatin were transferred to a clean 1.5 ml tube and incubated with primary antibody against SMUG1 (Abcam, 192240) or IgG (Abcam, 172730) with rotation overnight at 4°C in ChIP buffer. Immunoprecipitation was done with Dynabeads Protein G with rotation for 4 h at 4°C. Beads were then washed for 5 min sequentially with low salt wash buffer (16.7 mM Tris–HCl, pH 8, 167 mM NaCl, 0.1% SDS, 1% Triton X) (two washes), high salt wash buffer (16.7 mM Tris–HCl, pH 8, 500 mM NaCl 0.1% SDS, 1% Triton X) (one wash), LiCl wash buffer (250 mM LiCl, 0.5% NP40, 0.5% Na-deoxycholate, 1 mM EDTA, 10 mM Tris–HCl, pH 8) (two washes), and TE buffer (10 mM Tris–HCl, pH 8, 5 mM EDTA) (two washes). Chromatin was eluted from the beads with 300 μl of elution buffer (1% SDS, 100 mM NaHCO₃) for 20 min at RT. Beads were immobilized on a magnetic rack, and supernatant was transferred to a clean 1.5 ml tube. For reverse-crosslinking, chromatin was incubated with 21.2 μg Proteinase K in 50 mM Tris–HCl, 250 mM NaCl, and 12.5 mM EDTA for 1 h at 50°C, then overnight at 65°C. DNA was purified via phenol-chloroform extraction, ethanol-precipitated, and resuspended in water. Quantification was performed with QuantStudio™ 7 Flex Real-Time PCR System (Applied Biosystems). Derived DNA (2 μl, 1:10) was mixed with 5 μl of SYBR Green real-time PCR Master Mix, 2.6 μl of water and 0.2 μl of each primer targeting promoters of the genes of interest (Supplementary Table S2). Levels of immunoprecipitated DNA are expressed as relative enrichment = (% input of promoter region of regulated gene)/(% input of promoter region of unaffected gene).

Focal detection of Smug1 substrates assay

Focal levels of SMUG1 substrates were assessed in the promoters of OR genes of the OB and PFC WT and Smug1^−/− (n = 7 per genotype). Initially, 4 μl of 50 ng/μl of freshly isolated genomic DNA was incubated with 10U of SMUG1 (NEB, M0336S) and NEBuffer™ 4 in a 10 μl reaction volume for 30 min in water bath at 37°C. Next, 10U of APE1 (NEB, M0282S) prepared in NEBuffer™ 4 (30 μl reaction volume) was added to the reaction, and a second incubation for 30 min was carried out. Each sample had a negative control reaction consisting of genomic DNA incubated with the respective storage buffer of each enzyme. To strengthen the DNA-breaks formation induced by SMUG1, a second control was tested. Genomic DNA was incubated with heat-inactivated SMUG1 (hiSMUG1, NEB M0336S) at 95°C for 20 min to ensure catalytic death and DNA binding capacity of the enzyme. Two-step reactions with 10U of hiSMUG1 and 10U of APE1 were carries out. After the reactions, samples were cooled at RT, dilutions of 1:20 were prepared, and 2 μl was added to 5 Power SYBR Green PCR Master Mix 2× (Applied Biosystems), 2.6 μl of water, and 0.2 μl of each primer targeting promoters of the genes of interest (Supplementary Table S2). Samples were amplified using QuantStudio™ 7 Flex Real-Time PCR System (Applied Biosystems). Levels of SMUG1 substrates were calculated as follow: ΔCt = Ct(+SMUG1 + APE1) – Ct(-SMUG1 buffer-APE1 buffer); and ΔCt = Ct(+SMUG1 + APE1) – Ct(+hiSMUG1 + APE1).

Immunofluorescence of brain sections

Sagittal brain sections from WT and Smug1^−/− mice were baked at 58°C for 15 min, deparaffinized in xylene substitute, and slowly hydrated in decreasing concentrations of ethanol. Heat-induced antigen retrieval was performed in citrate buffer (0.01 M, pH 6.0) using a pressure cooker for 3 min. Next, samples were permeabilized in PBS + 0.25% Triton X-100 for 15 min and blocked in PBS + 5% goat serum + 5% bovine serum albumin for 1 h. Incubation with primary antibodies was carried out overnight at 4°C in a humidified chamber. The following primary antibodies were used for immunofluorescence experiments: anti-pan-ADP-ribose reagent (rabbit Fc tag) (Merck, MABE 1016, 1:500), anti-phospho-histone H2A.X (Merck, 05-636, 1:50), anti-NeuN [EPR12763] (Abcam, ab279296, 1:250), anti-MAP2 (Abcam, ab5392, 1:500), and anti-Tyrosine Hydroxylase (Merck, MAB318, 1:200). Slides were washed 3× in PBS + 0.1% Tween and incubated with fluorescence-tagged secondary antibodies for 2 h at RT. The following secondary antibodies were used at dilution 1:500: Goat anti-Mouse IgG1 Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 (Thermo Fisher, A21121); Goat anti-Rabbit IgG (H + L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 594 (Thermo Fisher, A11012); Goat anti-Mouse IgG2a Cross-Adsorbed Secondary Antibody, Alexa Fluor 647 (Thermo Fisher, A21241); and Goat Anti-Chicken IgY H&L Alexa Fluor 647 preadsorbed antibody (Abcam, ab150175). Slides were mounted with ProLong Gold antifade reagent with DAPI (Molecular Probes). All images were acquired in a Zeiss LSM780 confocal microscope with 20×, 40×, or 63× objectives from five different Z planes. Quantification of fluorescent signals was carried out on threshold of maximum intensity projection images using ImageJ v1.53k. For nuclear staining of γH2AX the NeuN channel was used to define a region of interest. For poly-ADP ribose (PAR), the Dapi-counterstained nuclei was used and the mean grey value within each nucleus was calculated. MAP2 quantification is expressed as raw intensity (arbitrary units, A.U.) and TH signals are shown as percentage (%) of area fraction.

Suspension breaks labelling In Situ and sequencing and bioinformatics

Suspension breaks labelling In Situ and sequencing (sBLISS) protocol was performed as described in Bouwman et al. [24]. Nuclei from frozen OB tissue from 6-months-old WT and Smug1^−/− mice (n = 3 per genotype) were isolated and fixed before template and library preparation using sBLISS adaptors and indexed primers (Supplementary Tables S3–S5). Sequencing (paired-end sequencing with read length = 150) was performed by Novogene Ltd using the Illumina NovaSeq X Plus series platform. The quality of all fastq files and assessment of possible contamination were verified using FastQC (0.12.1) and FastScreen (0.16.0), respectively. Relevant barcodes were found and mapped. Originally, the authors use scan_for_matches; however, this tool is old and difficult to install while retaining compatibility with other tools. As a substitute, we wrote a bash script which performs the same task. To test the bash script, we used example data provided by Bouwman et al. [24]; the results with the bash script and with scan_for_matches were identical. Following barcode isolation, we trimmed away the unnecessary components of the barcodes and aligned to the mm10 genome, specifically the GRCm39 primary assembly genome. The alignment was done with bwa (0.7.18). Unique molecular identifiers (barcodes) are then filtered and counted using scripts provided by the protocol. The scripts are available on their github page (https://github.com/BiCroLab/blissNPanalysis/tree/master). From the resulting bam files peaks are calculated using MACS3 (3.3.0). Peaks were obtained using one file at a time, each file being set as the treatment (per MACS3’s wording). All other peak calling parameters were left to default.

Nissl staining

Sagittal brain sections (4 μm) of paraffin-embedded brains of WT and Smug1^−/− mice (n = 3 per genotype, 2–4 sections per animal) were used to evaluate the cytoarchitecture of the OB and PFC. The sections were baked for 15 min at 58°C, and the paraffin was removed in xylene. Samples were hydrated in decreasing alcohol gradient (2 × 100%, 1 × 95%, 1 × 70%), washed in ddH₂O and stained with 0.1% cresyl violet (Abcam). Dehydration was done in increasing alcohol gradient (1 × 70%, 1 × 90%, 1 × 95%, 2 × 100%) before the sections were mounted with Vectamount (Vector Laboratories). Images were acquired at NORBRAIN (The Norwegian Brain Initiative) using 20× objective in a slide scanning AxioScan Z1 microscope (Carl Zeiss).

Statistical analysis

Statistical analyses were performed using GraphPad software version 10.0.2 or 10.4.1. The comparison of the means between two independent groups was done using two-tailed unpaired t-test with alpha set <0.05 when values passed normality test (Shapiro–Wilk test). Non-normal distribution values were analysed using two-tailed unpaired Mann–Whitney U-test with alpha set <0.05. One-way ANOVA followed by Bonferroni’s multiple comparison test was applied for samples with more than two independent groups. Correlation analyses were performed using Spearman coefficient test. Independent groups with interaction factors (time, genotype, etc.) were analysed with two-way ANOVA with Sidak’s post hoctest or uncorrected Fisher’s LSD.

Results

SMUG1 regulates behaviour

To better characterize SMUG1 function in the brain, we first asked whether SMUG1 was universally expressed or confined to distinct brain regions. RNA scope and immunohistochemistry analyses showed that Smug1 mRNA and protein expression were found throughout the brain (Supplementary Fig. S1A–C), with the STR having the highest fraction of Smug1-positive cells (Supplementary Fig. S1B). As a first evaluation of the functional impact, we subjected adult male WT and Smug1^−/− mice to behavioural tasks (Fig. 1). Exploratory behaviour and general activity, measured using the open field test, showed that Smug1^−/− mice explored the environment more than WT mice, according to the number of entries in the central area and distance travelled (Fig. 1A). As increased exploratory behaviour might be an indicator of reduced anxiety, we evaluated anxiety-like behaviour in the elevated zero maze test. Confirming the OF test, Smug1^−/− mice spent significantly more time exploring the open areas (Fig. 1B, left plot). The total distance travelled did not change (Fig. 1B, right plot), and there was no indication of reduced motor function or motor coordination (Supplementary Fig. S1D and E), supporting that Smug1^−/− mice have a less anxious phenotype. Next, T-maze with rewarded alternation was used to evaluate cognition (Fig. 1C). Here, Smug1^−/− animals consistently required more time to find the reward than WT mice (Fig. 1C, left plot). Initially, only 20% of Smug1^−/− animals chose the correct arm compared to 60% of the WT mice (day 1, P ≤ .001). During the first four consecutive days of trials, the number of mice choosing the correct arm gradually increased, indicating that Smug1^−/− mice have the capacity to learn a task (Fig. 1C, right plot). However, after 2 days pause between trials, the percentage of Smug1^−/− animals choosing the rewarded arm was reduced, while the WT mice retained the memory (Fig. 1C, right plot).

Figure 1.

Smug1^−/− mice show decreased anxiety and reward behaviour. (A) Exploratory behaviour and locomotor activity were assessed using the open field test. Time spent in the central area and distance travelled of mice that moved freely in the open field for 30 min (WT n = 17, Smug1^−/−n = 16). (B) Elevated zero maze was used to evaluate anxiety-like behaviour. Mice were allowed to travel between open and closed areas for 5 min. The time spent in the open areas and the distance travelled were analysed (WT n = 17, Smug1^−/−n = 16). (C) T-maze with rewarded alternation using 50% condensed milk. The time to find reward and the percentage of correct choice were parameters used to evaluate the spatial memory and learning capacity of Smug1^−/− and WT mice (WT n = 10, Smug1^−/−n = 9). (D) Buried food test evaluated the olfaction (time to find the food) and reward (time to eat the food) levels (WT n = 17, Smug1^−/−n = 14). Behavioural tests were compared using two-tailed unpaired t-test: *P ≤ .05; ** P ≤ .01; *** P ≤ .001; **** P ≤ .0001. Statistical analysis for T-maze was performed using two-way ANOVA with Sidak’s post-test, P< .0001.

As reward behaviour in the T-maze critically depends on olfaction, we employed the buried food test (Fig. 1D), which not only measures the general olfaction function but also uncovers motivation and ability to eat food. The latency to find the food was significantly higher for Smug1^−/−, indicating impaired olfaction capacity (Fig. 1D, left plot). Furthermore, once the food was found, Smug1^−/− mice exhibited reduced motivation to eat, corroborating the observed impaired reward behaviour (Fig. 1D, right plot). We did not detect obvious structural changes in reward-related regions, such as the STR, or the tyrosine hydroxylase-positive dopaminergic neurons in the substantia nigra (SN) (Supplementary Fig. S2A–D). Thus, Smug1^−/− mice exhibited a complex behavioural phenotype characterized by decreased anxiety and reward-related behaviour, delayed learning capacity, and reduced memory retention.

SMUG1 modulates gene expression in different brain regions

Multiple brain regions coordinate reward responses. Therefore, to better understand the mechanisms underlying the behavioural phenotype of Smug1^−/− mice, we performed RNA sequencing analyses. The brain regions were selected based on the behavioural tests (Fig. 2): the olfactory bulb (OB), as central for olfaction; the prefrontal cortex (PFC), crucial for high order cognitive processes [25]; the striatum (STR), known for its involvement in multiple sensory functions and as a critical part of the reward circuit [26]; the ventral midbrain (vMB), comprising subregions such as the SN and the ventral tegmental area which are important modulators of motor movement and reward functions [27]; and the hippocampus (HPC), which has a major role in memory, learning, and anxiety regulation [28] (Fig. 2A).

Figure 2.

RNA sequencing analysis of Smug1^−/− versus WT brains. (A) Brain regions with critical role in behaviour were isolated and prepared for RNA sequencing analysis. (B) Venn diagram depicting the number of individual and overlapped differentially expressed genes (DEGs) in each brain region. (C) Complex heatmap showing the read counts of the significantly DEGs [57] in each brain region evaluated (P ≤ .05). (D, F) UpSet plots indicate the set size of significantly upregulated and downregulated Gene Ontology (GO) terms per brain region and their intersections (P ≤ .01). (E, G) Heatmap showing significantly enriched GO terms clustered based on Wang distance measurement. Upregulated terms are shown in panel (E), while downregulated terms are shown in panel (G). Columns show the results for the different brain regions studied. Magnification of selected areas highlight GO terms intersecting between OB and PFC. Experiments were performed in 6-months-old WT and Smug1^−/− male mice (n = 3 per genotype).

Loss of SMUG1 modulated gene expression programs in all brain regions tested (Fig. 2B). The highest number of DEGs were found in the OB and PFC (Fig. 2C, P ≤ .05), of which ∼5% (31 genes) were shared (Fig. 2B). Strikingly, the OB showed a very high proportion of downregulated genes (∼80%) (Fig. 2C).

GO analyses were performed for all regions individually. UpSet plots (Fig. 2D and E) show that most of the enriched GO terms were unique for each region. Among the upregulated genes, the PFC shows the higher number of enriched GO terms. Many of the impacted GO terms in this brain region are related to calcium-mediated signalling, import, and intracellular transport (Supplementary Table S6). Calcium ion transport is similarly upregulated, but to a lesser extent in the OB. The only significant intersection observed between the OB and PFC is a group of transcripts related to calcium ion transport (GO:0051298, P ≤ .01) (Fig. 2D and E). Among the downregulated GO terms (Supplementary Table S7), two intersections were observed between the OB and PFC datasets: sensory perception of smell (GO:0007608, P ≤ .01) and G protein-coupled receptor signalling (GO:0007186, P ≤ .01) (Fig. 2G). Concept network plots illustrate that these clusters were characterized by an extensive downregulation of OR genes (Fig. 3A and B and Supplementary Tables S8–S11). The number of downregulated OR genes was much higher in OB than in the PFC. A subset of OR genes was selected to demonstrate the reduced expression levels in Smug1^−/−mice (Fig. 3C and D; Supplementary Tables S8 and S10; Supplementary Fig. S3). ORs are G protein-coupled receptors expressed in the olfactory epithelium and various brain regions, comprising the largest multi-gene family associated with perception of smell in mammals [29]. Consequently, the behavioural deficits in Smug1^−/−mice coincided with downregulation of genes implicated in olfactory sensory perception in the two brain regions.

Figure 3.

OR genes are overrepresented in downregulated GO terms intersecting between OB and PFC. (A) Sensory perception of smell and (B) G protein-coupled receptors. Olfr: OR genes. (C) Relative mRNA levels of selected ORs are significantly lower in the OB of Smug1^−/− mice. (D) ddPCR analysis showing the significant downregulation of OR mRNAs in the OB (a) and in the PFC (b). Olfr1000 and Olfr527 were found not significantly affected in the RNA seq analysis and served as control genes for the biological validation of mRNA levels in both brain regions. Two-way ANOVA, Sidak’s multiple comparison test, *P ≤ .05; ** P ≤ .01, **** P< .0001.

Accumulation of 5hmU is associated with downregulated OR genes

Quantification of base modifications in genomic DNA via mass spectrometry revealed higher levels of 5hmU in all brain regions of Smug1^−/− mice compared to isogenic WT mice (Fig. 4A). Interestingly, 5hmU levels in the OB and PFC of WT mice were below the detection limit. If 5hmU was generated by deamination of 5hmC, as suggested previously [9, 10], a negative correlation between these base modifications would be expected. However, the global levels of 5mC and 5hmC remained unchanged in all brain regions of Smug1^−/− mice (Fig. 4B and C). Nevertheless, a correlation was observed between the levels of 5hmU and 5hmC in Smug1^−/− brain regions (Fig. 4E and F), which might point to a shared mechanism of biogenesis. Given that 8oxoG levels, a biomarker of oxidative stress, remained unchanged and showed no correlation with 5hmU (Fig. 4D and Supplementary Fig. S4A and B), it is improbable that oxidation of thymine by reactive oxygen species is a major source of 5hmU in the Smug1^−/− brain.

Figure 4.

Smug1^−/− mouse brain accumulates 5hmU and results in downregulation of OR genes in the OB and PFC. (A–D) Global levels of 5hmU, 5mC, 5hmC, and 8oxoG were measured using mass spectrometry in different brain regions. Results are presented as the number of modifications (mods) residues per million nucleosidese. Analysis was performed in five mice per genotype (E, F) Spearman correlation showing the relationship between 5hmU and 5hmC levels in WT and Smug1^−/− brain regions, respectively. A significant (p< .0001) and positive correlation was detected in Smug1^−/− mice. (G, H) Quantification of the AT content (%) in downregulated OR genes in ± 1 kb regions from TSS in the OB (n = 159) and PFC (n = 20). Two-tailed unpaired t-test: *P ≤ .05; **P ≤ .01; ***P ≤ .001. Basal levels of 5hmU in the OB and PFC of the WT mice were below the limit of detection (0.05 nM 5hmU∼0.8 mods/10⁶ dNs); therefore, unpaired t-test could not be performed.

The promoters of OR genes are characterized by their AT-rich nature [30] and, given the previous suggestion that 5hmU might be generated by enzymatic oxidation of thymine, it was interesting to note that the AT content of the significantly downregulated OR genes in Smug1^−/− mice ranged from 55% to 70%, as measured within a distance of ±1 Kb from the TSS (Fig. 4G and H). Several regulatory regions control the expression of OR genes [31, 32]. Among these, the AT-rich motif 5′-AACTTTTTAATGA-3′ is often found within regulatory sequences that control OR gene expression [32]. Given that this T-rich motif may serve as a hotspot for 5hmU formation [22], we reasoned that its occurrence in the promoter region of downregulated OR genes could be indicative of 5hmU-mediated regulation. Allowing for three mismatches, we assessed its occurrence in the promoter (−500 bp from TSS) of the downregulated OR genes [23, 32, 33] and found that ∼70% of them contained at least one occurrence of this motif (Supplementary Fig. S4C). Taken together, these data suggest that accumulation of 5hmU is associated with reduced expression of OR genes in Smug1^−/− mice.

SMUG1 localizes at the promoters of OR genes

A mechanistic involvement of SMUG1 in expression regulation would require its binding to 5hmU in the OR promoter regions. To address this, we performed ChIP focusing on the OB as this region had the highest number of downregulated AT-rich OR genes. We found significant SMUG1 enrichment at the promoters of the downregulated Olfr102, Olfr1019, Olfr1115 genes, and a positive trend for Olfr872 (Fig. 5A) compared to the non-regulated gene Olfr527. SMUG1 also showed higher localization in the Olfr1115 promoter compared to Gapdh promoter (Fig. 5B), a non-regulated gene. Next, we asked whether SMUG1 substrates accumulated in these promoter regions in absence of the DNA glycosylase. Using a qPCR-based method which relies on DNA lesion excision [34] (Fig. 5C), we found that SMUG1 substrate levels were consistently higher in the promoters of differentially expressed OR genes in the OB of Smug1^−/− mice (Fig. 5D and Supplementary Fig. S5A). The specificity of SMUG1 substrate accumulation in the regulated OR is evidenced by the absence of such accumulation in non-regulated genes (Olfr1000,Gapdh,andDnph1). SMUG1 was also detected at the promoters of OR genes in the PFC (Olfr524,Olfr727,andOlfr1509) (Supplementary Fig. S6A and B) and its substrates were found in some OR genes in the PFC of Smug1^−/− mice (Supplementary Figs S5B and S6C). Furthermore, the same OR genes do not show SMUG1 substrate accumulation in the HPC—a brain region where we have not detected read counts for these genes (Supplementary Fig. S5C). Collectively, these results show that SMUG1 recognizes and regulates modified DNA bases specifically within AT-rich promoters of OR genes.

Figure 5.

SMUG1 localizes to promoters of OR genes in the OB and induces the formation of physiological DNA strand breaks. (A) ChIP analysis showing preferential localization of SMUG1 in the promoters of significantly downregulated OR genes in OB. Results are from three independent experiments, each using OB from three animals pooled together (n = 9). Olfr527 was used as a control gene. Two-tailed unpaired t-test: *P ≤ .05; **P ≤ .01; ****P ≤ .0001. (B) SMUG1 preferentially localizes to promoter of OR genes, here exemplified by Olfr1115. Gapdh represents a non-regulated gene. Results are from OB pooled together (n = 3). (C) Illustration of the region-specific assessment of SMUG1 substrates in promoters of OR genes. DNA isolated from OB was incubated with SMUG1 and APE1 enzymes in a two-step reaction. Primers targeting the AT-rich promoters of OR genes were used to amplify the DNA after enzymatic reactions. (D) Regions with high accumulation of SMUG1 substrates result in DNA breaks and higher Ct values when quantified by qPCR. Ct values obtained from Smug1^−/− mice are significantly higher than WT indicating accumulation of modified bases in the promoters of the affected OR genes in the OB. Olfr1000, Gapdh,and Dnph1 represent promoters of non-regulated genes. Two-tailed unpaired U-test: *P ≤ .05, **P ≤ .01, ***P ≤ .001; (E) Immunofluorescent images of brain sections showing reduced γH2AX signal in NeuN-positive neurons of the mitral cell layer (ML) of the OB in Smug1^−/−. The ML is indicated by white dashed lines. (F) Quantification of γH2AX signal. (G) Representative images of reduced PARylation levels in the ML of the OB of Smug1^−/− mice compared to WT mice. (H) Quantification of PARylation images. The median of individual mean grey value per nucleus was used to represent the signal intensity of γH2AX and PARylation. Two-tailed unpaired U-test: ***P ≤ .001. Scale bars: 50 μm. (I) Quantification of DSB peaks in the genome of OB and PFC from WT and Smug1^−/− mice.

SMUG1 generates physiological DNA strand breaks

The BER pathway is emerging as a regulator of neurodevelopment due to its role in generating physiological DNA strand breaks [1]. To test whether SMUG1-initiated BER could be a source of physiological DNA strand breaks, we performed γH2AX staining in the OB and PFC. Due to the structural complexity of the OB, the analysis was limited to the ML. The ML contains mitral cells, which are the primary population of projection neurons in the OB that receive input from olfactory sensory neurons and transmit the information to several areas of the brain, including the olfactory cortex [35]. The intensity of γH2AX immunosignal per nucleus was markedly reduced in NeuN-positive neurons in the ML of Smug1^−/− mice (Fig. 5E and G). Similarly, the mean intensity of PAR signal per cell was reduced in Smug1^−/− mice (Fig. 5F and H). Thus, lack of SMUG1 suppresses BER-intermediate formation in ML neurons. PFC contrasted the ML in the OB; lack of SMUG1 induced a significant increase in γH2AX (Supplementary Fig. S6D and F) and PAR intensity (Supplementary Fig. S6E and G), suggesting that the bulk of physiological DNA strand breaks in the PFC is not solely SMUG1 dependent. To verify that SMUG1 deficiency results in a reduction of endogenous DNA strand breaks in the OB, we employed in-sBLISS, a method that enables genome-wide profiling of DNA double-strand breaks (DSB). The sBLISS analysis indicated a reduced amplitude of DSB peaks in the OB and an increased in the PFC in Smug1^−/− mice compared to WT controls (Fig. 5I), even though the medians between Smug1^−/− and WT are not significantly different. These findings are consistent with the observed γH2AX signals in the OB and PFC. In both brain regions, the global feature distribution of DSB is similar between the genotypes (Supplementary Fig. S7A and B).

As DNA damage and repair mechanisms are increasingly recognized as essential regulators of neuronal development and function [3, 36–38], we hypothesized that SMUG1 deficiency might affect the cytoarchitecture of the brain.

Nissl staining in Smug1^−/− brains revealed changes in cell layer organization in the OB, with thicker external plexiform layer compared to the WT (Fig. 6A, top panel). Additional subregions of the OB, such as the ML, internal plexiform layer, and granule layer showed changes in cell distribution (Fig. 6A, middle and bottom panels, respectively). In the neocortex, we observed clear dysplasia in the most superficial layers of the PFC; layers I–III were thicker in some areas, and the typical layered structure was lost in Smug1^−/− mice compared to WT mice (Fig. 6B). Additionally, the borders between layers II and III were highly diffuse. Another interesting observation regarding the PFC was that the cellular size was quite homogeneous within the same layer in WT mice, whereas the cellular homogeneity was lost in Smug1^−/− mice.

Figure 6.

The cytoarchitecture of the OB and PFC is altered in the absence of SMUG1. (A) Nissl staining showing cytoarchitectonic disorganization in the OB of Smug1^−/− mice. GL: glomerular layer; EPL: external plexiform layer; ML: mitral layer; IPL: internal plexiform layer; and GR: granular layer. (B) Nissl staining of the neuronal population in the neocortex showing atypical cell size and dysplasia in layers I–III of Smug1^−/− mice. Scale bars: 100 μm.

Taken together, our data suggest that SMUG1 is required for proper development and function of the OB and PFC by modulating strand-break formation through the timely processing of regulatory 5hmU marks, which are present in the AT-rich promoters of OR genes.

Discussion

Here we unveil a role of DNA glycosylase SMUG1 as a regulator of OR gene expression in the brain. Our findings reveal that loss of SMUG1 in mice leads to a complex behavioural phenotype characterized by impaired sensory perception of smell.

Our findings highlight a regulatory role of SMUG1 in transcriptional processes, evidenced by SMUG1 binding to OR promoters and the focal accumulation of SMUG1-specific substrates in these promoters in Smug1^−/− mice. Despite the unavailability of methods to simultaneously map different SMUG1 substrates, our data strongly suggest 5hmU to be the driver of the observed phenotypes, given its global accumulation in the brain regions of Smug1^−/− mice.

OR genes are organized in clusters, but individual genes are expressed in a monogenic and monoallelic manner [39]. The molecular mechanisms governing their expression are delineated by two major models: one stochastic, in which OR genes are randomly and independently activated, and the other deterministic, postulating a regulated process. The observed phenotypes are likely the result of a combination of both models, where some elements are described while others remain to be elucidated [40, 41]. The choice and de-silencing of a naive OR is driven by the transient expression and activity of the histone demethylase LSD1 [42, 43], which induces a local oxidative burst, resulting in oxidation of DNA bases like 8oxoG.

A noteworthy feature of OR promoters is their high AT content, but as thymine has higher oxidation potential than guanine, it is not known whether a physiological oxidative burst would be sufficient to generate 5hmU in the AT-enriched environment of the OR gene. An enzymatic route of 5hmU generation from T was suggested to depend on TET enzymes [7]. Irrespective of its source, however, our findings suggest that presence of 5hmU in OR promoters may add an additional layer of transcriptional regulation for this class of genes. It was previously demonstrated that DNA methylation in CpG and CpA dinucleotides does not influence the expression level of OR genes [44]. Thus, this is a first report where a DNA base modification in a regulatory region of OR genes was found to modulate gene expression. Further, we propose that SMUG1-initiated BER removes physiologically generated 5hmU prompting the restoration of T bases and that this process is essential for fidelity in the olfactory function. This model of SMUG1 erasing physiological 5hmU aligns with our observations of fewer DNA breaks in the OB of Smug1^−/− and the lower 5hmU levels in the OB compared to other brain regions of adult mice.

DNA breaks dynamically generated during BER may have a role in epigenetic reprogramming and cellular differentiation [1, 3]. Physiological DNA breaks are also necessary for subsequent consolidation of neuronal identity [45], learning [2, 46], and memory [47]. Adding to data describing roles for topoisomerase II [48] and active demethylation through the TET-TDG axis, our findings reveal that SMUG1 impacts the generation of physiological DNA strand breaks in the brain. In Smug1^−/− mice we detected reduced formation of DNA breaks in ML neurons, suggesting that SMUG1 contributes to a portion of the physiological DNA breaks in this region. This observation aligns with recent reports of reduced DNA strand breaks in the brain of alkyladenine DNA glycosylase-knockout mice (Aag^−/−), which were accompanied by lower levels of 5hmC and altered gene expression [49]. The observed increase in DNA strand breaks in the PFC, on the other hand, suggests the involvement of additional DNA glycosylases capable of recognizing and excising 5hmU, such as NEIL1 and NTHL1 [50].

While previous reports [11, 13] did not reveal structural abnormalities in the brains of Smug1^−/− mice, the more comprehensive analysis presented here of the brain regions impacted by OR gene downregulation revealed substantial differences from WT animals. The disorganization observed in cell layers and variances in cell sizes strongly suggest that SMUG1 plays a pivotal role in neurodevelopment. Dysregulation or dysfunction of DNA glycosylases has been associated with changes in neuronal function and cognition. This is evident from the behavioural consequences frequently seen in mice deficient in several DNA glycosylases [14, 49, 51–53]. A common characteristic observed in mouse models lacking DNA glycosylases, including Smug1^−/− mice, is a decrease in anxiety levels [49, 52, 53]. Reports have also indicated examples of delayed learning and memory retention in mice lacking DNA glycosylases, as observed in Smug1^−/− mice. Although the function of DNA glycosylases in olfaction impairment has not yet received much attention, it has been identified in Neil1^−/− mice, who, similarly to Smug1^−/− mice, displayed increased latency to find the food in the BF test [54]. However, while SMUG1-deficient mice exhibit some of the same behavioural abnormalities as other DNA glycosylase-deficient mice, diminished motivation to eat in the BF test has not been previously reported. Olfaction capacity is not only strongly linked to memory and learning, but also significantly influences odour-associated reward responses [55]. Smug1^−/− mice performed poorly in the rewarded-T maze, indicating deficits in rewarded-learning and memory. Reward-related learning is orchestrated by dopamine circuits connecting dopaminergic neurons in the midbrain to the striatal neuronal population. Despite the behavioural impairment, we did not find evidence of neurodegeneration of dopaminergic neurons in the SN or gross morphological alterations in the STR of Smug1^−/− mice. Importantly, the olfactory tuberculum was recently identified as the mediator of the connection between the OB and STR, providing evidence on how reward-associated behaviour is driven by olfaction capacity [56].

Given the influential role of olfaction in shaping emotional and cognitive responses, we propose that the behavioural responses observed in Smug1^−/− mice primarily stem from impaired olfactory function, which, under physiological circumstances, is regulated by the effective repair of 5hmU and subsequent formation of strand breaks in promoters of OR genes.

These findings shed light on unique processes involving SMUG1 in the brain and greatly broaden the biological relevance of 5hmU in the genome and neurodevelopment. Specifically, 5hmU functions as a transcriptional regulator of OR genes, and the impact of its persistence in the genome results in olfaction dysfunction.

Supplementary data

Supplementary data is available at NAR online.

Conflict of interest

None declared.

Funding

This work has been supported by grants funded to HN from the South-Eastern Norway Regional Health Authority (grant no. 2019098), The Research Council of Norway (grant no. 14477), and the Norwegian Cancer Society (grant no. 295901). Funding to pay the Open Access publication charges for this article was provided by The Research Council of Norway.

Data availability

The RNA sequencing data reported in this paper have been registered in the Gene Expression Omnibus repository under the accession code GSE274703.