Background

The hypothalamus serves as the central hub for controlling energy homeostasis, stress response, temperature, learning, feeding, sleep, social behavior, sexual behavior, hormone secretion, reproduction, osmoregulation, blood pressure, visceral activities, emotion, and circadian rhythms [1]. The hypothalamic energy-sensing system, particularly the circuits that regulate food intake, plays a crucial role in life span extension [2]. Elevated metabolic activity in the aged hypothalamus has been reported in aged hypothalamus, including increased mTor signaling [3, 4]. Additionally, decreases in gonadotropin-releasing hormone (GnRH), Ghrh, Trh, monoamine neurotransmitters, and blood supply are hallmarks of aging hypothalamus [5].

Previous studies have demonstrated that 17α-estradiol extends the lifespan of male mice and has beneficial effects on metabolism and inflammation, similar to those of rapamycin and acarbose [6–8]. Recent study indicated that 17α-estradiol also extends the lifespan of male rats [9].Further investigations revealed certain unique features of 17α-estradiol in life extension distinct to rapamycin and acarbose [10, 11]. Moreover, it has been shown that 17α-estradiol targets hypothalamic POMC neurons to reduce metabolism by decreasing feeding behavior through anorexigenic pathways [12]. Interestingly, the lifespan extension effect has only been observed in male animals [13]. The safety of 17α-estradiol is key for translation into clinical treatment, and the potential side effects on reproduction and feminization by 17α-estradiol treatment must be considered. However, contradictory results have been reported regarding its side effects on reproduction and feminization [6, 14, 15]. Therefore, further investigation and verification are needed to understand the underlying mechanisms of lifespan extension and the safety of 17α-estradiol.

In this report, we utilized single-nucleus transcriptomic sequencing and performed supervised clustering of neurons based on neuropeptides, hormones, and their receptors. Supervised clustering offers better resolution in cell cluster screening compared to traditional unsupervised clustering. We assessed the effects of 17α-estradiol on metabolism, stress responses, ferroptosis, senescence, inflammation, and pathways involved in synaptic activity in each neuron subtype, ranking the most sensitive neurons. The effects of 17α-estradiol on reversing aging-related cellular processes were evaluated by two opposing regulatory networks involved in hypermetabolism, stress, inflammation, and synaptic activity. Several key endocrine factors from serum were examined, and the potential side effects of 17α-estradiol on specific neurons were also evaluated.

Materials and methods

Animals, treatment and tissues

Twelve Norway brown male rats (12 months old) were acquired from Charles River, including 8 12-months old and 4 1-month old (Beijing). Aged rats were randomly allocated into control and 17α-estradiol-treated groups. Four Aged rats treated with 17α-estradiol (Catalog #: E834897, Macklin Biochemical, Shanghai, China) were fed freely with regular diet mixed with 17α-estradiol at a dose of 14.4 mg/kg (14.4 ppm), starting at 24 months of age for 6 months according to prior reports [16, 17]. The young rats were fed a regular diet without 17α-estradiol continuously for 3 months until 4 months old. All rats had ad libitum access to food and water throughout the experiments. The rats were then euthanized via CO₂, hypothalami, testes and blood serum were collected for subsequent experimental procedures. All blood samples were collected at 9:00-9:30 a.m to minimize hormone fluctuation between animals. All animal procedures were reviewed and approved by the Institutional Animal Care and Use Committee at Nantong University.

Enzyme immunoassays

Enzyme immunoassays kits for rat Oxt (Catalog #: EIAR-OXT), Corticotropin Releasing Factor (Catalog #: EIAR-CRF), and gonadoliberin-1 (Catalog #: EIAR-GNRH) were obtained from Raybiotech (GA, USA). Enzyme immunoassay kits for rat serum total testosterone (Catalog #: ml002868), estradiol (Catalog #: ml002891), aldosterone (Catalog #: ml002876), and cortisol (Catalog #: ml002874) were obtained from Enzyme-linked Biotechnology (Shanghai, China). Sera from 3 animals per group were used and each was diluted 10 or 20 times for immunoassays.

Seminiferous tubule inflammation test

8 testes were obtained from each sample group and then subjected to fixation in 4% formalin for at least 1 week. Formalin-fixed paraffin-embedded mouse testis sections of 5 µm thickness were used for HE Staining. At least 30 seminiferous tubules in each slide were examined for inflammation test. Testis with at least 1 inflammatory seminiferous tubule was set as 1, and normal testis was set as 0 for inflammation index calculation.

snRNA-seq data processing, batch effect correction, and cell subset annotation

Intact hypothalami were cryopreserved in liquid nitrogen from sacrificed rats. Two (O) or three (Y and O.T) hypothalami were pooled within each group and homogenized in 500 µL ice-cold homogenization buffer (0.25 M sucrose, 5 mM CaCl₂, 3 mM MgAc₂, 10 mM Tris-HCl [pH 8.0], 1 mM DTT, 0.1 mM EDTA, 1× protease inhibitor, and 1 U/µL RiboLock RNase inhibitor) with Dounce homogenizer. Then, the homogenizer was washed with 700 µL ice-cold nuclei washing buffer (0.04% bovine serum albumin, 0.2 U/µL RiboLock RNase Inhibitor, 500 mM mannitol, 0.1 mM phenylmethanesulfonyl fluoride protease inhibitor in 1× phosphate buffer saline). Next, the homogenates were filtered through a 70-µm cell strainer to collect the nuclear fraction. The nuclear fraction was mixed with an equal volume of 50% iodixanol and added on top of a 30% and 33% iodixanol gradient. This solution was then centrifuged for 20 min at 10 000 ×g at 4 °C. After the myelin layer was removed from the top of the gradient, the nuclei were collected from the 30% and 33% iodixanol interface. The nuclei were resuspended in nuclear wash buffer and resuspension buffer and pelleted for 5 min at 500 ×g at 4 °C. The nuclei were filtered through a 40-µm cell strainer to remove cell debris and large clumps, and the nuclear concentration was manually assessed using trypan blue counterstaining and a hemocytometer. Finally, the nuclei were adjusted to 700–1200 nuclei/µL, and examined with a 10X Chromium platform.

Reverse transcription, cDNA amplification and library preparation were performed according to the protocol from 10X Genomics and Chromium Next GEM Single Cell 3′ Reagent Kits v3.1. Library sequencing was performed on the Illumina HiSeq™ 4000 by Gene Denovo Biotechnology Co., Ltd (Guangzhou, China). 10X Genomics Cell Ranger software (version 3.1.0) was used to convert raw BCL files to FASTQ files, and for alignment and counts quantification. Reads with low-quality barcodes and UMIs were filtered out and then mapped to the reference genome. Reads uniquely mapped to the transcriptome and intersecting an exon at least 50% were considered for UMI counting. Before quantification, the UMI sequences were corrected for sequencing errors, and valid barcodes were identified using the EmptyDrops method. The cell × gene matrices were produced via UMI counting and cell barcodes calling. Cells with an unusually high number of UMIs (≥8000) or mitochondrial gene percent (≥15%) were filtered out. Batch effect correction was performed by SCTransform function built in Seurat V4.4.0.

Pathways, gene signatures, TFs and TF cofactors, cell communication

Gene sets and pathways were derived from Hallmark gene sets of MSigDB collections, the KEGG pathway database, Reactome pathway database, and WikiPathways database, and some ontology terms derived from the Gene Ontology (GO) resource. Mitochondrial pathways were derived from MitoCarta3.0 [18]. Pathways, gene sets, and gene signatures were evaluated with the PercentageFeatureSet function built into R package Seurat. TFs and TF cofactors were obtained from AnimalTFDB 3.0 [19]. TFs and TF cofactors were further filtered with mean counts >0.1. The ligand–receptor pairs were calculated via R package CommPath [20].

Correlation analysis and ROC analysis

Pearson correlation coefficient was calculated with the linkET package (p < 0.05). Fast Wilcoxon rank sum test and auROC analysis was performed with the wilcoxauc function in R package presto. The minimal cell number in either one of the comparing pairs should be no less than 15. Ranks of area under the curve (AUC) values were in descending order. A total of 431 pathways from Hallmark, KEGG and PID databases were used for correlation analysis with MitoCarta OXPHOS subunits in neurons and non-neural cells (Figure 2—figure supplement 1). A total of 97 pathways related to synapse activity were derived from GO, including GO cellular components, GO biological processes and GO molecular functions (Table S1).

The division of expression level-dependent clusters in each pathway and their gene signatures

The quarters of the mixed cell populations from O, O.T, and Y hypothalamic neurons were equally divided using the R function fivenum from the R package stats, based on pathway expression levels. Thus, the total number of neurons was evenly divided into four clusters (c1-c4) in terms of cell number. The cell proportions from O.T, O, and Y neurons in each cluster were weighted against the total number of neurons in the three groups. The unique markers of each cluster were calculated using the FindAllMarkers function from the Seurat package. The intersection of the unique markers from the six pathways was obtained for heatmap plotting. Nineteen genes that were highly expressed in c1 were identified as c1.up.signature via the PercentageFeatureSet function in the Seurat package. Twelve genes that were highly expressed in c4 were identified as c4.up.signature. There were no intersecting unique markers in clusters c2 and c3 among the six selected pathways.

TF and pathway activities

The TF resources were derived from CollecTRI, the pathway resource was from PROGENy, and the enrichment scores of TFs and pathways were performed with the Univariate Linear Model (ulm) method according to the pipeline in R package decoupleR [21].

Subtypes of neurons generated by supervised clustering

Vast majority of these subtypes were clustered by neuropeptides, hormones, and their receptors within all the neurons with the subset function from R package Seurat (the target gene expression level > 0). A total of 121 neuron subtypes were obtained, comprising 80 neuropeptide-secreting neurons and 41 neurons expressing a unique neuropeptide receptor or hormone receptor. Further groupings may exist within the identified neuron subtypes, and the category of excitatory or inhibitory neurons was not discriminated further. The cell proportion of each neuron subtype was weighted according to the total number of neurons in O.T, O, and Y samples. The mean values ± standard deviation of pathways and gene signatures were performed for each subtype. The top 10 and the bottom 10 items were calculated, and the top 10 subtypes ranked within each of 16 pathways or gene signatures were collected for word clouds annotation via R package wordcloud2.

Differential expression and pathway enrichment analysis

DEGs between groups were identified via FindMarkers (test.use = bimod, min.pct = 0.1, logfc.Threshold = 0.25, avg_diff > 0.1 or < −0.1). DEGs were then enriched in redundant GO terms via WebGestalt and filtered with false discovery rate < 0.05 [22].

Bidirectional Mendelian randomization (MR) study

The protein quantitative trait locus (pQTL) GWAS summary data of 204 human endocrine-related GWAS summary data with European ancestry were obtained from open-access MRC Integrative Epidemiology Unit (IEU) (Table S4) [23, 24]. Independent genome-wide significant SNPs for exposure OXT (id:prot-a-2159) or GNRH1 (id: prot-a-1233) were used as instrumental variables with genome-wide significance (P < 1 × 10⁻⁵), independence inheritance (r2 < 0.001) without linkage disequilibrium (LD) with each other for MR. For the reverse MR, independent genome-wide significant SNPs from 203 endocrine-related GWAS summary data (P < 1 × 10⁻⁵, r2 < 0.001) without LD with each other were obtained as exposures and human GWAS summary data of OXT (id:prot-a-2159) or GNRH1 (id:prot-a-1233) were used as outcomes. Weak instruments less than 10 were discarded via F-statistics.

MR and reverse MR analysis were conducted with method inverse-variance weighting (IVW), MR Egger, Weighted median, Simple mode, and Weighted mode. The screening criteria: all of the odds ratio (OR) values of the 5 methods should be simultaneously either >1 or <1 and the significant p value of IVW was <0.05. The heterogeneity via IVW method and the horizontal pleiotropy were also evaluated with R package TwoSampleMR [25].

Results

The overall changes in aged hypothalamus with or without long-term 17α-estradiol treatment via snRNA-seq profiling

To investigate the hypothalamus as a potential key target of 17α-estradiol’s effects on life extension, we performed snRNA-seq on the entire hypothalamus of aged and 17α-estradiol-treated aged Norway brown rats, using the hypothalamus from young adult male rats as a control. We identified 10 major cell types based on specific cell markers of the hypothalamus (Figure 1A-B). Notably, the proportions of all non-neural cells changed in O versus Y (Figure 1C). For instance, Oligo, OPC, and Micro were found to be increased, while Astro, Tany, Fibro, PTC, and Endo were decreased in O compared to Y. The proportions of Oligo, OPC, and Micro were also increased in 17α-estradiol-treated aged group (O.T) compared to those in Y. Furthermore, Endo was increased in O.T compared to both Y and O. The proportions of Astro, Tany, Epen, and PTC decreased more in O.T than those in O when compared to Y. These results indicated that 17α-estradiol treatment had extensive effects on the proportions of non-neural cells in hypothalamus.

Figure 1.

Cell communication analysis revealed significant changes in the ligand-receptor pairs between neurons and other cell types, particularly in Endo, Fibro, Tany, and Astro (Figure 1D). Significant ligand–receptor interactions among neurons also changed in O.T and O groups, especially in O (Figure 1E). Notably, among the significant ligand–receptor pairs in neurons, Bmp2–Acvr1/Acvr2a/Acvr2b/Bmpr, Gdf11–Acvr2a/Acvr2b, Inhba–Acvr1/Acvr2a/Acvr2b, Nrg1/Nrg2/Nrg4–Erbb4, Rspo1–Lgr5/Lrp6, and Rspo3–Lgr5 were exclusively and significantly increased in neurons of the O group compared to those in O.T and Y, suggesting enhanced TGF superfamily-mediated signaling activity and canonical Wnt signaling during aging. The significantly changed ligand–receptor pairs Nlgn1–Nrxn1/Nrxn2, Nlgn2–Nrxn1/Nrxn2/Nrxn3, Nlgn3–Nrxn1/Nrxn2/Nrxn3, Nxph1–Nrxn1/Nrxn2/Nrxn3, Nxph3–Nrxn1/Nrxn2/Nrxn3, Pomc–Oprd1/Oprk1/Oprm1, and Vip–Adcyap1r1/Avpr1a/Vipr2 were exclusively increased in neurons of O.T compared to O and Y (Figure 1E). These ligand–receptor pairs were associated with synaptic activity, cellular adhesion, the opioid system, and vasodilation, indicating unique roles of 17α-estradiol in restoring certain physiological functions in the aging hypothalamus. The increased Pomc signal in O.T neurons aligns with previous reports suggesting that 17α-estradiol treatment decreases food uptake in mice, potentially correlated with Pomc neurons (Figure 1E) [12].

Gene set enrichment analysis (GSEA) based on DEGs also corroborated the expression profiles related to stress responses and synapse-associated cellular processes in neurons (Figure 1F). ROC analysis of significantly differently expressed pathways related to neural synapses, manually selected from Gene Ontology databases, indicated that most top-ranked pathways related to synapses, according to AUC values, were downregulated in aged neurons, while 17α-estradiol treatment reversed this trend (Figure 1G, Table S1).

Overall, these findings suggest that 17α-estradiol broadly reshapes cell populations, cellular communication, neuropeptide secretion, and synapse-related cellular processes in the aging hypothalamus, distinguishing it from both the young hypothalamus and the untreated aged hypothalamus.

The two opposing signaling networks in regulating metabolism and synapse activity, which can be balanced effectively by 17α-estradiol

To monitor the metabolism and neural status affected by 17α-estradiol, we utilized the energy metabolism pathway MitoCarta OXPHOS subunits to calculate the positively or negatively correlated pathways in hypothalamic neurons (Figure 2—figure supplement 1). Our findings revealed that energy metabolism and synapse activity represent two opposing regulatory signaling networks in hypothalamic neurons, with 17α-estradiol strongly playing a significant role in balancing these networks (Figure 2A). At the core of these opposing signaling pathways are two categories of contrasting TFs (Figure 2B). For example, Calr, Clu, Peg3, Prnp, Ndufa13, Actb, Ywhab, Nfe2l1, Mtdh, Npm1, Bex2, Aft4, and Maged1 were positively correlated with pathways involved in OXPHOS subunits, lysosome function, protein export, mTorc1 signaling, and the unfolded protein response (UPR) in O, O.T and Y neurons, while showing negative correlations with pathways related to ubiquitin-mediated proteolysis, endocytosis, tight junctions, focal adhesion, axon guidance, and MAPK signaling. Additionally, TFs Myt1l, Ctnnd2, Tenm4, Camta1, Med12l, Rere, Csrnp3, Erbb4, Jazf1, Dscam, Klf12, and Kdm4c exhibited opposite correlation patterns with these selected pathways in O, O.T and Y neurons. These TFs may take conserved roles in regulating the two opposing biological processes in hypothalamic neurons.

Figure 2.

We then attempted to establish gene signatures to represent these two opposing signaling networks, thereby displaying the cell status of aging and evaluating the effects exerted by 17α-estradiol. To achieve this, we evenly divided the expression levels of each of the six selected pathways from the two opposing signaling networks into four quarters (c1-c4) among the mixed neurons from O, O.T, and Y, calculating the shared unique markers in each quarter (Figure 2C, D). From the distribution patterns, we observed that the proportion of neurons in O decreased from c1 to c4 in metabolic pathways (MitoCarta OXPHOS subunits and Hallmark mTorc1 signaling), while this trend was reversed in the opposing signaling pathways (GOBP synapse organization and KEGG MAPK signaling pathway) (Figure 2C). In contrast, in Y, this trend was opposite, suggesting the expression levels from the 4 quarters (c1-c4) of the two opposing signaling networks can be used to monitor aging status. Treatment with 17α-estradiol alleviated this trend or even reversed it in O.

We then screened the shared unique markers of each quarter from the six selected pathways in an attempt to establish the gene signatures representing the two opposing signaling networks. Unique markers in c1 (19 genes, c1.up.signature) and c4 (12 genes, c4.up.signature) were identified; however, c2 and c3 lacked unique markers shared by the 6 pathways (Figure 2D). Consequently, the 19 genes in c1.up.signature displayed an inversed correlation pattern with the 12 genes in c4.up.signature, indicating the two opposing gene signatures are capable of reflecting the two opposing signaling networks in hypothalamic neurons (Figure 2E). Conversely, the balance of the two opposing signaling networks affected by 17α-estradiol in non-neural cell types was less pronounced than in neurons, showing variable effects on non-neural cells (Figure 2—figure supplement 2). GOBP pathway enrichment analysis revealed that Micro exhibited lower levels of synapse-related cellular processes in O.T compared to O, which was distinct from the observations in neurons (Figure 2—figure supplement 3). Therefore, in this report, we primarily focused on hypothalamic neurons and their responses to aging and 17α-estradiol.

Supervised clustering revealed distinct responses of different subtypes of hypothalamic neurons to aging and 17α-estradiol

The hypothalamus contains numerous neuron subtypes that release various neuropeptides and hormones to regulate fundamental body functions. To differentiate the changes occurring during aging and the effect of 17α-estradiol on each neuron subtype, we performed supervised clustering based on neuropeptides, hormones, or their receptors (Figure 3A). Most of the top 10 neuron subtypes involved in UPR, ferroptosis, mTorc1 signaling, insulin signaling pathway, immune response pathways, OXPHOS subunits, and senescence signature during aging were attenuated by 17α-estradiol treatment (Figure 3B, Figure 3—figure supplement 1). For example, Serpine1-secreting neurons exhibited very high expression levels of UPR, ferroptosis signature, mTorc1 signaling, Tnfα signaling via NF-kB, and OXPHOS subunits during aging, all of which were attenuated by 17α-estradiol.

Figure 3.

To further identify the most sensitive neurons in response to aging and 17α-estradiol, we calculated the frequency of each neuron subtype by combining the top 10 neurons across each of the 16 pathways and gene signatures (Figure 3C, Table S2). Serpine1-secreting neurons were exclusively enriched in aged hypothalamus, suggesting they are effective targets of 17α-estradiol.

Additionally, Mlnr, Nmb, Pomc, Ednra, Serpine3, Gast, and Pcsk6 neurons were all affected by 17α-estradiol (Figure 3D). Galp, Glp1r, Serping1, Sstr2, Sstr3, and Vip neurons were enriched in O, O.T and Y hypothalamus, indicating a lack of effect of 17α-estradiol on these neuron types. The unique neurons in O.T (Crh, Serpinb9, Cckbr, Pth2r, Kiss1, Prpr, Hcrtr1, Npb, Nppa, Nxph3, and Npff) may be a consequence of compensatory effects from 17α-estradiol treatment.

We then evaluated the most and least sensitive neurons to aging by ranking the neuron subtypes according to the levels of c1.up.signature and c4.up.signature (Figure 3E). Serpine1, Galp, Calca neurons were among the most affected subtypes in the aged hypothalamus, consistent with the combined ranks of several pathways (Figure 3—figure supplement 1). Npr3, Crh, Ar, Esr1, and Esr2 neurons were among the least affected subtypes in the aged hypothalamus (Figure 3E). In O.T, neuron subtypes ranked by c1.up.signature and c4.up.signature indicated that Gast, Npb, Nppa, and Crh were affected most by 17α-estradiol treatment (Figure 3F). Neurons Oxtr, Glp1r, Gnrh1, and Crh also showed the lowest levels of c4.up.signature, indicative of an aging phenotype. Meanwhile, neurons Glp2r, Ar and Esr1 ranked among the top neurons based on c4.up.signature, suggesting higher expression levels of synapse-related processes and relatively less stress status in these types of neurons in O.T.

These results indicate that supervised clustering of each neuron subtype of neurons facilitated the visualization of the different responses from distinct subtypes of neurons in response to aging and medical treatments. Furthermore, it strongly suggests that Crh, Oxtr, Glp1r, Gnrh1, Glp2r, Ar, and Esr1 neuron subtypes are particularly sensitive to long-term 17α-estradiol treatment, which is associated with appetite regulation, glucose metabolism, stress response, and, as expected, sex hormone secretion and signaling.

The potential side effects on hypothalamic–pituitary–adrenal (HPA) axis by long-term 17α-estradiol treatment in the males

To further investigate the potential side effects or compensatory effects of 17α-estradiol treatment, we performed stricter screening by intersecting the top 10 ranks in the UPR, OXPHOS subunits, and ferroptosis signature (Figure 4A). Crh and Nppa were the only two intersected neuron subtypes identified through this strict screening. The treatment with 17α-estradiol elevated several key metabolic pathways in Crh neurons compared to those in Y and O (Figure 4B). Additionally, 17α-estradiol treatment increased the c1-up-signature while simultaneously reducing many pathways associated with synapse activity and the c4-up-signature in Crh neurons of O.T, indicating a potent stressed phenotype in Crh neuron. In contrast, in Nppa and Nppc neurons, the decreased c1-up-signature in O.T implied a lesser extent of stressed phenotype in these neurons compared to Crh neurons. The aberrant changes in Crh neurons were also evidenced by the increased number of DEGs related to mitochondria-expressed genes in O.T, along with a reduced number of DEGs in the adherens junction pathway (Figure 4C). The status of Crh neurons in O.T may be associated with elevated TF activities of Esr2, Usf2, Hdac5, Creb3l1, Tfam, Preb, Pou3f2, and Hoxb5 (Figure 4D).

Figure 4.

Notably, the HPA axis was altered by 17α-estradiol treatment, as evidenced by the elevated cortisol levels in O.T compared to O (p = 0.078) (Figure 4E). The correlation between elevated cortisol production and the heightened stress in Crh neurons by 17α-estradiol treatment needs further investigation. Additionally, as a crucial component of the renin-angiotensin-aldosterone system, the significantly increased serum aldosterone in O.T and its potential role in sodium reabsorption and cardiovascular health also warrant more detailed investigation (Figure 4E). In summary, 17α-estradiol treatment altered the activity of HPA axis in male BN rats and introduced increased potential side effects in Crh neurons.

17α-estradiol increased Oxt neuron proportion and secretion and its possible role in mediating the effect of 17α-estradiol on endocrine system

Aging and 17α-estradiol treatment also altered the proportions of various neuron subtypes among O, O.T, and Y (Figure 5A, B, Table S3). The proportions of Grp, Pmch, Npb, Serpinb9, Sstr2, Agrp, Sstr3, Mlnr, and Hcrt neurons ranked in the top 10 in O, while Oxt, Vip, Avp, Calca, Glp2r, Tacr1, Trh, Serping1, Npff, and Npy1r were in the top 10 in O.T. Galp, Calcrl, Ednra, Oxt, Serpinh1, Pomc, Cck, Crh, Tacr1, and Kiss1 neurons were among the bottom 10 in O, whereas Oxtr, Galp, Agrp, Serpinb9, Npvf, Serpinh1, Ednrb, Agt, Gipr, and Pomc neurons ranked among the bottom 10 in O.T. Agrp, Pomc, Oxt, Oxtr, Gipr, and Glp2r neurons are well-known for their roles in regulating food intake and energy homeostasis. Agrp neurons are activated by hunger, while Pomc neurons are activated by satiety in the ARC of the hypothalamus. 17α-estradiol treatment effectively elevated the expression levels of the c4.up.signature and synapse-associated processes in neuron subtypes Agrp, Pomc, Oxt and Glp2r in O.T compared to O. This may mitigate the adverse effects of reduced cell populations in Pomc and Agrp neurons in aging hypothalamus (Figure 5C). This finding indicates a potential role of 17α-estradiol in appetite control, as previously reported [12]. Additionally, 17α-estradiol treatment resulted in elevated proportions of Vip, Avp, Npff, Calca, and Tacr1 neurons, which ranked in the top 10, while the proportions of Agt neurons decreased. These are all associated with blood pressure regulation (Figure 5—figure supplement 1). Notably, the proportions of Oxt and Glp2r neurons, both of which have anorexigenic effects [26, 27], increased in O.T. In addition to the increased number of Oxt-positive neurons, the expression level of Oxt also rose in O.T. The elevated expression of synapse-related pathways was supported by the increased DEGs in the enriched synaptic membrane pathway in Oxt neurons (Figure 5D).

Figure 5.

More importantly, the serum level of Oxt was significantly elevated in O.T compared to O (p=0.04), although it remained lower than that in Y (Figure 5E). Notably, the top TF activities in O.T and O differed markedly from those in Y (Figure 5F). The elevated levels of Hopx and Xbp1 may be associated with the response to 17α-estradiol treatment.

Due to the intricate regulatory networks among various endocrine factors, elucidating the causal effect of Oxt on other endocrine factors is quite complex using traditional methods. MR analysis, employing variant SNPs as genetic tools, is advantageous for such task. We performed a bidirectional MR analysis of the GWAS summary data of human plasma OXT and 203 endocrine-related and hypothalamus-related factors, most of which are protein quantitative trait loci (pQTL) data from the IEU (Table S4). As an exposure, OXT revealed a significant causal effect on POMC/beta-endorphin (id:prot-a-2327, id:prot-a-2325), glucagon (id:prot-a-1181), GNRH1/Progonadoliberin-1 (id:prot-a-1233), and total testosterone levels (id:ebi−a−GCST90012112, id:ieu−b−4864) (Figure 5G). NPW and CBLN1 were found to be negatively associated with OXT, but the significance of these associations was not found in the reverse MR analysis (Figure 5—figure supplement 2A, B).

In contrast, we could not identify significant associations between OXT and estradiol levels (id:ebi-a-GCST90012105, id:ebi-a-GCST90020092, id:ebi-a-GCST90020091, id:ieu-b-4872, id:ieu-b-4873, id:ukb-e-30800_AFR, id:ukb-e-30800_CSA). Interestingly, QRFP, IGF1, AGRP, TAC4, GRP, CLU, BNF, PCSK7, PACAP, ANP, TAC3, CRH, INSL6, and PRL displayed significant associations with OXT in both MR and reverse MR analysis, indicative of their complex causal effects (Figure 5—figure supplement 2A, B).

The results suggest that elevated Oxt levels induced by 17α-estradiol may have positive associations with endocrine factors governing feeding behavior, glucose metabolism, male reproduction, and sex hormones. Therefore, OXT may serve as a potential mediator of 17α-estradiol.

17α-estradiol activated HPG axis and the elevated Gnrh also took important roles in mediating the effect of 17α-estradiol on other endocrine factors

Since Gnrh and sex hormone expressing neuron subtypes were sensitive to 17α-estradiol treatment (Figure 3F, Figure 3—figure supplement 1), we examined their expression profiles alongside representative pathways of two opposing signaling networks related to metabolism and synapses, including c1-up-signature and c4-up-signature (Figure 6A). However, neither the c1-up-signature nor the c4-up-signature was up-regulated in Gnrh1 neuron in the O.T, similar to the observations in Esr2 neuron. Only in Esr1 neuron was the c1-up-signature found to be up-regulated. Meanwhile, both Ar and Esr neurons displayed high level of c4-up-signature, indicating a relatively healthy status. The data suggest that most of the 4 types of sex hormone-related neurons did not experience severe aging in O.T. However, based on these expression profiles, it’s difficult to define the precise physiological status of these subtypes of neurons, particularly regarding neuroendocrine activities. Consequently, we performed enzyme immunoassays of hormones from the serum of O, O.T and Y. The treatment with 17α-estradiol significantly increased the plasma level of Gnrh compared to Y (p = 0.0099) and approached significance when compared to O (p = 0.096) (Figure 6B). More intriguingly, testosterone levels in serum were significantly increased in O.T compared to O (p = 0.018) and approached significance when compared to Y (p = 0.052).

Figure 6.

Additionally, the serum estradiol levels were significantly increased in O compared to Y (p = 0.011) and significantly decreased in O.T compared to O (p = 0.019), suggesting that 17α-estradiol treatment markedly altered the homeostasis of testosterone and estradiol.

Furthermore, most testes from 30-month-old male BN rats exhibited severe age-related inflammation and epithelial collapse of seminiferous tubules (Figure 6C). The testes without inflammation in O.T displayed normal morphology. 17α-estradiol treatment slightly decreased the testis inflammation in O.T compared to that in O (p = 0.15), indicating a potential positive role of 17α-estradiol treatment in male reproductive system. The elevated TFs such as Sf1, Pparg, Litaf, Nupr1, Rxrg, E2f2, and Zfp42 may be involved in the transcriptional regulation by 17α-estradiol in O.T (Figure 6D). Importantly, the activities of androgen and estrogen pathways were decreased in Gnrh1 neurons in O.T compared to O, and were distinct from those in Ar, Esr1, and Esr2 neurons (Figure 6E). These signaling pathways are important for the feedback control of sex hormone secretion in Gnrh neurons, and these results may also reflect the strong effect of 17α-estradiol on Gnrh neurons.

To decipher the potential effects of elevated serum Gnrh levels on endocrine system, we performed bidirectional MR analysis of the GWAS summary data of human GNRH1 (id: prot-a-1233) and 203 endocrine-related factors with genetic variants SNPs. We found strong causal effects of GNRH1 on GAL/Galanin (id:prot−a−1166), POMC/Beta−endorphin (id:prot−a−2327, id:prot−a−2325), Adrenomedullin (id:prot−a−48), BDNF (id:prot−a−242), and LPR (id:prot−a−1724), which are involved in feeding, energy homeostasis, osmotic regulation, and neuron plasticity (Figure 6F). Notably, CRH/Corticotropin (id:prot−a−2326), PRLH/Prolactin−releasing peptide (id:prot−a−2376), NPW/Neuropeptide W (id:prot−a−2082), Glucagon (id:prot−a−1181), Chromogranin−A (id:prot−a−538) displayed bidirectional significance, indicating close and complex causal effects between GNRH1 and these endocrine factors (Figure 6G, Figure 6—figure supplement 1A, B). These results also suggest that the role of 17α-estradiol treatment in feeding, energy homeostasis, reproduction, osmotic regulation, stress response, and neuronal plasticity may be mediated, at least in part, by elevated Gnrh secretion.

Discussion

The most striking role of 17α-estradiol treatment revealed in this study showed that HPG axis was substantially improved in the levels of serum Gnrh and testosterone. The underlying molecular mechanism remains unclear; however, prior reports have indicated that 17α-estradiol can bind to ESR1 [9]. In our findings, 17α-estradiol treatment significantly decreased serum estradiol levels while elevating serum testosterone. Based on this evidence, we propose that 17α-estradiol may function similarly to estrogen receptor antagonists or aromatase inhibitors, potentially preventing the conversion of androgens to estrogens [28, 29]. These actions could alleviate the feedback inhibition exerted by estrogen on hypothalamus and pituitary, thereby facilitating the secretion of Gnrh, FSH, and LH [30].

The testosterone levels in men gradually decline beginning in the third decade of life [31]. Age-related deterioration of the gonadotropic axis, particularly in older males with low serum testosterone, is often linked to numerous aging symptoms, including loss of skeletal muscle mass, reduced muscle strength and power, low bone mineral density, frailty, impaired physical performance, mobility limitations, increased risk of diabetes, elevated all-cause cardiovascular mortality, cognitive decline, and heightened risk of Alzheimer’s disease [32]. Consequently, testosterone supplementation in older men is beneficial. Additionally, Gnrh supplementation may help mitigate age-related declines in neurogenesis and slow aging processes [33]. Importantly, treatment with 17α-estradiol did not result in feminization or adversely affect the sperm parameters and fertility in male animals [6, 14]. Thus, the observed increases in Gnrh and serum testosterone levels due to 17α-estradiol treatment are likely advantageous for older males, particularly those experiencing late-onset hypogonadism.

Postmenopausal women with low estrogen experience aging-related syndromes similar to those of older males with low serum testosterone. Those women also face increased mortality, cardiovascular disease, osteoporosis fracture, urogenital atrophy, and dementia, all of which may benefit from hormone therapy [34]. However, a prior report indicates that 17α-estradiol treatment does not provide positive life extension effects in aged females [13]. The discrepancy may stem from the inhibitory effects of estrogens associated with 17α-estradiol treatment, as evidenced by its inability to enhance female fertility [35]. Nonetheless, due to the lack of parallel data in aged female BN rats treated with 17α-estradiol, further research is needed to definitely address this question in the female subjects in the future.

Another notable effect of 17α-estradiol is its ability to reduce the overall expression levels of energy metabolism in hypothalamic neurons of aged male BN rats. The nutrient-sensing network, mediated by MTORC1 complex, is a central regulator of mRNA and ribosome biogenesis, protein synthesis, glucose metabolism, autophagy, lipid metabolism, mitochondrial biosynthesis, and proteasomal activity [36]. Downregulation of this nutrient-sensing network has been associated with increased lifespan and healthspan [37]. Notably, 17α-estradiol treatment diminished nutrient-sensing network activity in most hypothalamic neurons, which may be a contributing factor in promoting lifespan extension.

In this report, we demonstrated significant changes in neuron populations involved in appetite control, including Agrp, Pomc, Oxt, Oxtr, Gipr, and Glp2r neurons. Among the identified subtypes, the proportion of Oxt neurons saw the most considerable increase due to 17α-estradiol treatment (Figure 5A). Oxt plays versatile roles in social behavior, stress response, satiety, energy balance, reproduction, and inflammation [38]. Most Oxt neurons originate from the paraventricular nucleus (PVN) and supraoptic nuclei (SON) in the hypothalamus, exhibiting high plasticity during development and intricate circuitry [39, 40]. The PVN, arcuate nucleus (ARC), and ventromedial hypothalamic nucleus together form a neural hub in the hypothalamus that integrates peripheral, nutritional, and metabolic signals to regulate food intake and energy balance [41]. Many effects of Oxt are sex-specific [42]; for instance, females are less sensitive to exogenous Oxt than males regarding social recognition [43]. Interestingly, Oxt injections, facilitated by nanoparticles that enhance blood-brain barrier penetration, reduced body mass while increasing social investigation and the number of Oxt-positive cells in the SON, particularly in male rats [44]. Additionally, intracerebroventricular injections of Oxt in rats showed a reduction in food intake in both sexes, with a more pronounced effect in males [45].Therefore, we propose that Oxt’s role in systemic aging and feeding behavior may contribute to the sex-biased effects of 17α-estradiol, warranting further verification.

Furthermore, 17α-estradiol treatment appears to have enhanced stress in HPA axis. One evidence was the increased levels of ferroptosis-signature and UPR in Crh neurons. The other evidence was the elevated serum cortisol, which is also a potential hallmark of aging HPA axis [46, 47]. Therefore, more attentions should be paid to the potential side effects of 17α-estradiol especially in its clinical application.

In summary, our findings suggest that 17α-estradiol treatment positively influences the HPG axis and neurons associated with appetite and energy balance. This may be closely linked to the life-extension effects of 17α-estradiol in aged males. Additionally, employing supervised clustering based on neuropeptides, hormones, and their receptors proves to be a valuable strategy for examining pharmacological, pathological, and physiological processes in different neuronal subtypes within the hypothalamus.

Abbreviations

• Ar: androgen receptor

• ARC: arcuate nucleus

• Astro: astrocyte

• AUC: area under the curve

• Cga: glycoprotein hormones, alpha Polypeptide

• Crh: corticotropin releasing hormone

• DEG: differentially-expressed gene

• Endo: endothelial cell

• Esr1: estrogen receptor 1

• Esr2: estrogen receptor 2

• Fibro: fibroblast

• Glp2r: glucagon-like peptide 2 receptor

• GnRH: gonadotropin releasing hormone

• GSEA: gene set enrichment analysis

• HPA: hypothalamic–pituitary–adrenal

• HPG: hypothalamic-pituitary-gonadal

• IEU: MRC Integrative Epidemiology Unit

• IVW: inverse-variance weighting

• KEGG: Kyoto Encyclopedia of Genes and Genomes

• Micro: microglia

• MR: Mendelian randomization

• MTORC1: mechanistic target of rapamycin kinase 1

• Nppa: natriuretic peptide A

• Oligo: oligodendrocyte

• OPC: oligodendrocyte precursor cell

• OR: odds ratio

• OXPHOS: oxidative phosphorylation

• PID: Pathway Interaction Database

• POMC: proopiomelanocortin

• pQTL: protein quantitative trait loci

• PTC: pars tuberalis cell

• PVN: paraventricular nucleus

• ROC: receiver operating characteristics

• snRNA-seq: single-nucleus transcriptomic sequencing

• Tany: tanycyte

• TF: transcription factor

• UPR: unfolded protein response.

Acknowledgements

We appreciate Dr. Qinghua Wang and Hongyun Shi from animal facility of Nantong University in helping with the animal experiments. Special thanks to Professor Ken-ichiro Fukuchi from University of Illinois College of Medicine for constructive comments and suggestions in manuscript preparation.

Additional information

Funding

This work was supported by the National Natural Science Foundation of China grant 31271448 (YL), 82171621 (LL), 82172566 (ZY), 82150107 (RL) and the National High Level Hospital Clinical Research Funding (2022-PUMCH-A-231) (LL).

Author contributions

Conceptualization, YL, LL; Methodology, YL, GW, XX; Animal operation: YL, LL; EIA assays: YL, LY; Inflammation test: YL, ZY, LL, YS. Validation, LL, JY, ZY; Formal Analysis, YL, JY; Resources, GW, YS, ZY, ZM, LX; Visualization, YL; Writing – Original Draft, YL, JY; Writing – Review & Editing, LL, YS; Supervision, YS, ZY, YL; Project Administration, YL, ZY, LL; Funding Acquisition, ZY, LL. All authors have read and approved the final manuscript.

Declaration of competing interest

The authors have no conflict of interests to declare.

Availability of data and materials

All data are available at GEO accession number GSE248413.

Competing interests

All authors declare no competing interests.