Skip to main content
NCBI home page
Endocrinology logoLink to Endocrinology
. 2025 Apr 30;166(6):bqaf084. doi: 10.1210/endocr/bqaf084

An Atlas of Thyroid Hormone Responsive Genes in Adult Mouse Hypothalamus

Shijia Wu 1,2, Julien Dellinger 3, Suzy Markossian 4, Yves Dusabyinema 5, Romain Guyot 6, Sandrine Hughes 7, Denise Aubert 8, Marie Fackeure 9, Karine Gauthier 10, Benjamin Gillet 11, Wenzheng Jiang 12, Frédéric Flamant 13,✉
PMCID: PMC12068221  PMID: 40302247

Abstract

We present an atlas of genes that respond to thyroid hormone in the adult mouse mediobasal hypothalamus. Based on droplet-based single nuclei RNA-seq method and batch transcriptome analyses, the atlas lists putative target genes of the hormone nuclear receptors in 20 different types of neuronal and glial cells. The transcriptional regulation of these genes varies extensively across neuronal and glial cell types. However, while astrocytes appear to be highly sensitive to thyroid hormone stimulation, differentiated oligodendrocytes are relatively insensitive. This atlas is expected to promote future investigations into the molecular and cellular mechanisms that underlie the numerous functions of thyroid hormone in the hypothalamic circuits.

Keywords: single nuclei analysis, thyroid hormone, hypothalamus, bulk RNA-seq, single nuclei RNA-seq

The hypothalamus represents a major interface between the brain and the periphery. It integrates external biochemical and sensory input to maintain homeostasis under various circumstances, controlling pituitary secretions, blood pressure, energy metabolism, sleep, appetite and reproduction. It is also the location of the central circadian and seasonal clocks. Its extreme neuroanatomical complexity, interconnecting 15 nuclei, enables a number of cross-talks between these functions (1).

The thyroid hormones (TH, including thyroxine, or T4, and its more active metabolite 3,3′,5-triiodo-L-thyronine or T3) exert a broad influence on hypothalamic functions. Several regulations have been documented, suggesting that the hypothalamus is one of the main targets of T3 action: (i) T3 downregulates thyrotropin-releasing hormone (TRH) secretion, which stimulates in turn the secretion of TSH by thyrotroph cells in the pituitary, and thus exerts a negative feedback regulation on the hypothalamic-pituitary-thyroid axis (2); (ii) T3 exerts local control on metabolism and energy balance by stimulating the sympathetic nervous system (3); (iii) T3 maintains the set point of body temperature (4); (iv) T3 inhibits food intake, refraining the activity of proopiomelanocortin (POMC) neurons (5) while activating neuropeptide Y (NPY) neurons (6); (v) T3 is a major component of the seasonal clock, coupling the length of the day with seasonal behavior (7); and (vi) T3 is also a relay to inflammatory response in pathological situations (8, 9).

The T3 content of the mediobasal hypothalamus is not directly coupled to the circulating level of TH. It can be enhanced by the conversion of circulating T4 by type 2 deiodinase, an enzyme that is present mainly in the tanycytes lining the wall of the lower part of the third ventricle. On the contrary, when T3 is in excess, the catabolic enzyme type 3 deiodinase is expressed in other cell types, including neurons and tanycytes located in the upper part of the ventricle, and the local level of T3 decreases. This metabolism allows rapid changes in T3 local content, while serum concentration remains stable. This also generates a local heterogeneous distribution of T3, which probably has a paracrine influence on specific neuronal cell populations in specific nuclei (10).

T3 exerts its action by binding to nuclear receptors, TRα1 and TRβ1/2 which are ligand-dependent transcription factors, collectively called TRs (11). They are encoded by the Thra and Thrb genes, which are expressed in many cell types, including virtually all hypothalamic cell types. Therefore, a complete understanding of T3 function in the hypothalamus represents a daunting task. The full characterization of TR target genes in each hypothalamic cell types would represent a major achievement and a significant progress in the understanding of many physiological functions. As a first step of this program, we establish here an atlas of T3-responsive genes in the mediobasal hypothalamus of adult mice at cellular resolution. This in-depth analysis identifies the most responsive cell types and lists putative TR target genes in each of them.

Methods

Animals

The research project was approved by a local ethics committee for animal experimentation (C2EA015) and subsequently authorized by the French Ministry of Research (Project #39805-2022110119433980 v4). Two groups of C57Bl6/j mice were fed for 2 weeks with iodine-deficient food supplemented with 0.15% of propyl-thiouracyl (Envigo ref TD.95125) to cause deep hypothyroidism. The first group of mice, referenced THhigh, received a single intraperitoneal TH injection (40 μg T4 + 4 μg T3 dissolved in 200 µL of phosphate buffer saline). The second group of mice, referenced THlow, was injected with 200 µL of phosphate buffer saline as control. Mediobasal hypothalami were dissected the next day, flash frozen using liquid nitrogen and stored at −70 °C.

Bulk Whole Cell RNA-seq From Sorted Hypothalamus Astrocytes

Hypothalami were dissected from 2 groups of 5 wild-type mice prepared as above. Astrocytes were isolated by magnetic sorting using the Anti-GLAST (ACSA-1) MicroBead Kit (Miltenyi Biotec Cat# 130-123-555, https://www.antibodyregistry.org/AB_2811532). The cDNAs were directly prepared from individual whole cell lysates using the whole cell RNA SMART-Seq V4 Ultra Low Input Kit (Takara). Sequencing libraries were then prepared from 1 ng of cDNA using the Nextera XT DNA Library Preparation Kit (Illumina) with Unique Dual Indexes. Quantification and qualification of dual indexed libraries were performed with both Qubit 4.0 (HS DNA Kit, Thermo Fisher) and Tapestation 4150 equipment using the D5000 ScreenTape System (Agilent). Qualified libraries were pooled in an equimolar manner and then sequenced on a NextSeq 500 platform (Illumina) according to the manufacturer's instructions. Sequencing was performed with High Output reagents in single-end and dual indexing, for a total of 72, 10 and 10 cycles run for Read 1, i7 and i5 index respectively. Raw sequencing data were deposited at EMBL-EBI, accession no. E-MTAB-15049.

Bulk Nuclei RNA-seq From Sorted GABAergic Neurons

Transgenic Gad2Crex ROSA-GSL10gfplox mice expressing a green fluorescent protein (GFP) in the nuclei of GABAergic neurons were treated as above. In these mice, the Gad2Cre driver (IMSR Cat# JAX:010802, RRID: IMSR_JAX:010802) drives Cre/loxP recombination in GABAergic neurons. It is combined with the ROSA-GSL10gfplox reporter transgene (12) for the conditional expression of a fusion protein based on the ribosomal L10a protein (13). Hypothalami were frozen individually in liquid nitrogen immediately after dissection. Samples were thawed and resuspended in 1.5 mL of lysis buffer (Nuclei EZ Lysis Buffer, Sigma Aldrich) and transferred in a 2-mL Kimble Dounce tissue grinder (Sigma Aldrich) for cell lysis (25 strokes with A pestle and 25 strokes with B pestle, on ice). Suspensions were transferred into tubes containing 2.5 mL of cold lysis buffer, incubated 5 minutes on ice and centrifuged (5 minutes 500g 4 °C, swinging rotor). Supernatants were discarded and nuclei pellets were washed once with 4 mL Nuclei EZ Lysis Buffer and once with 4 mL of Nuclei Suspension Buffer (NSB: phosphate buffer saline without Mg2+ and Ca2+; 1% nuclease free BSA (Sigma); 200 U/mL RNAsin (Promega). Nuclei pellets were resuspended in 0.3 mL NSB and filtrated on a 35 μm cell strainer (Corning). Then 1 μg of DAPI (Sigma Aldrich) was added to nuclei suspensions for DNA staining, allowing FACS sorting of singlets and not of aggregates of several nuclei. All nuclei resuspensions were achieved by pipetting 10 times with 1 mL of buffer. All GFP positive single nuclei were sorted (FACS Aria IIµ cytometer; BD Biosciences) in low-binding Eppendorf tubes containing 0.5 mL NSB. Nuclei were pelleted (5 minutes 2.104 g). Supernatant were discarded and nuclei resuspended in 350 μL RNA lysis buffer containing β-mercaptoethanol. RNA was extracted (RNeasy Micro Kit, Qiagen) and quantified using Tapestation 4150 (Agilent). The cDNAs were prepared from 1 ng of RNA using the RNA SMART-Seq V4 Ultra Low Input Kit (Takara Bio). Libraries were constructed and sequenced as described above. More than 24 M reads by sample were obtained. Raw sequencing data were deposited at EMBL-EBI, accession no. E-MTAB-15050.

Bulk RNA-seq Data Analysis

The 2 bulk RNA-seq datasets were analyzed using the European Galaxy platform (14) (https://usegalaxy.eu/ is maintained by the Freiburg Galaxy Team). Raw reads were mapped on the mouse genome (GRCm38 mm10 assembly), using Bowtie2 (Galaxy Version 2.5.3 + galaxy1) (15). Htseq-count (Galaxy Version 2.0.5 + galaxy0) (16) was then used to assign reads to genes (feature type: genes for GABAergic neurons nuclei, or exons for astrocytes). Differential gene expression analysis was performed with Deseq2 (Galaxy Version 2.11.40.8 + galaxy0) (17). Unsupervised principal component analysis, performed on gene count table of all samples from each project, identified one outlier library for astrocyte RNA-seq, one of the THlow samples, which was excluded from further analysis. Hierarchical clustering was performed using Euclidian distance and Ward's method with Clustvis (18).

Nuclei Isolation From Hypothalamus, Single Nuclei RNA Library Preparation, and Sequencing

Three hypothalamus samples (2 males, 1 female) were pooled from each group of mice, and nuclei were extracted as above except that the final centrifugation step was modified (5 minutes 300 g 4 °C swinging rotor). This step was optimized to minimize nuclei aggregation, which might produce clustering artifacts. For each condition, 2.105 nuclei were sorted and centrifuged at low speed to prevent nuclei aggregation. Supernatant volume was reduced to 100 μL before nuclei resuspension. Concentrations (1300 nuclei/μL) were determined with a Malassez hemometer. Quality control was later performed with ImageStream X cytometer (ISX, Luminex) to define the distribution of nuclei area (52 ± 2 μm2), diameter (0.8 ± 0.02 μm) and circularity (score 10.8 ± 0.5). Immediately after concentration determination, around 2.104 nuclei were partitioned to produce gel beads-in-emulsion (GEMs) using the Chromium Controller (10 × Genomics). Then, reverse transcription, cDNA recovery, cDNA amplification, and library construction were performed using the Chromium Next GEM Single Cell 3′ Reagent Kits v3.1 (10 × Genomics) following the manufacturer's protocols. Quantification and qualification of dual indexed libraries, referenced LibTHlow and LibTHhigh, were done with both Qubit 4.0 (High Sensitivity DNA Kit, Thermo Fisher) and Tapestation 4150 assays (D5000 ScreenTape reagents, Agilent). Qualified libraries were pooled in an equimolar manner and then sequenced on the NextSeq 500 platform (Illumina) using High Output reagents, according to the manufacturer's instructions (Illumina), with paired-end sequencing (28 and 120 cycles for reads R1 and R2, respectively) and dual indexing (10 cycles for each index).

Single Nuclei RNAseq Data Analysis

The Illumina raw sequencing data were analyzed with the dedicated Genomics set of pipelines, Cell Ranger (v7.1.0). Bcl files were first demultiplexed and converted to fastq files with the “mkfastq” pipeline. Single-cell feature-barcode matrices were then generated from fastq files with the count pipeline using the mouse genome (GRCm38 release 102) retrieved from Ensembl. Because single nuclei sequencing data are enriched in intronic reads, the include-introns argument was added to count reads falling in the intronic regions. More than 104 nuclei were identified by Cell Ranger in each library. It should be noted that over 30% of raw reads were mapped in antisense to genes and were therefore not included in the analyses. However, this high proportion is expected for Chromium Single Cell 3' Gene Expression v3.1 libraries when applied to suspensions of mouse brain nuclei (see Technical Note 10 × CG000376), with multiple mechanisms leading to these artifacts.

For each library, the filtered-feature_bc_matrix data (barcodes, features and matrix) were analyzed further with the Seurat toolkit v5 (19) considering a min.cells parameter equal to 3. Various filters were applied to reduce noise (nfeature_RNA parameter ranging from 200 to 5 000 for LibTHlow, 200 to 4 000 for LibTHhigh), and exclude doublets (DoubletFinder) (20) and cells with high mitochondrial content (>5% of reads). After these cleaning steps, 9504 nuclei and 9154 nuclei were retained for LibTHlow and LibTHhigh, respectively.

The 2 libraries were merged and normalized using the SCT normalization method, then integrated with the harmony method (21) in order to facilitate accurate comparative analysis across both datasets. The 25 first principal components were used to calculate a neighbor graph (FindNeighbors), based on which Louvain clustering was performed with a resolution of 0.3 identifying 21 biologically relevant cell clusters (FindClusters). The clusters annotation was performed manually, taking advantage of the HypoMap atlas (22).

The top 25 principal components were embedded onto 2 dimensions based on the UMAP algorithm using the RunUMAP function with 20 neighbors. The differentially expressed features were identified between both condition for each cluster (FindMarkers).

Quantitative Reverse Transcriptase Polymerase Chain Reaction

Total RNA was converted to cDNA using MMLV reverse transcriptase (Promega, Wisconsin, USA). Quantitative reverse transcriptase polymerase chain reaction (RT-qPCR) was performed using SYBRGreen mix (BioRad iQ supermix). The results were analyzed according to the ΔΔCT method (23). The housekeeping gene Hprt was used as the reference.

Results

Single Nuclei RNA-seq

We prepared two 10 × Genomics droplet-based libraries for single nuclei RNA (snRNA)-seq, LibTHlow and LibTHhigh, to address the response of mediobasal hypothalamus cells to TH over 24 hours (see “Methods”). Previous experiments indicate that this setting maximizes the chance to detect a robust hormonal response, while the relatively short duration of TH treatment minimizes its indirect effects, secondary to its systemic influence.

Deep sequencing (310 × 106 and 260 × 106 reads respectively for LibTHlow and LibTHhigh) provided sufficient data after filtering to explore the transcriptome of 18 658 cells (9 504 and 9154 cells; 28 717 and 27 868 total unique genes, 2250 and 1872 median genes per cell for LibTHlow and LibTHhigh, respectively). In particular, low mitochondrial level (<1%) and doublets were observed in datasets confirming the good quality of sample preparation.

Cell Clusters Identification

We used the Seurat package to cluster the integrated datasets and to visualize two-dimensional uniform manifold approximations and projections (UMAPs). Clustering can be operated with arbitrary resolution, as the definition of cell types relies on hierarchical classification. However, high resolution reduces the number of cells per cluster and reduces the statistical power of differential gene expression analysis. A preliminary analysis indicated that setting the clustering resolution to 0.3 maximizes the chance to discover differentially expressed genes (DEGs) in each cell cluster. At this resolution, 21 cell clusters were defined and ranked according to the number of cells that they contain (Fig. 1A). Examples of expression patterns allowing for cluster discrimination are given in Supplementary Fig. S1 (24). The contribution of LibTHlow and LibTHhigh was balanced except for cluster 15, for which more nuclei were found in LibTHhigh, and for cluster 18, where the situation was opposite (Table 1). After a global search for markers for each cluster (Fig. 1B and 1C), we used HypoMap (22) and the Allen Brain Cell Atlas (25) as references to select the most relevant markers and identify the cell types corresponding to each cluster (Fig. 1D). Except for cluster 21, which contains a very small number of cells, this allowed for an unambiguous identification of cell types (Table 1). The analysis of genes expressed in specific hypothalamic nuclei indicates that some types of neuronal cells might be present in a single location (Fig. 1E). For example, the Sim1 gene, which is mainly expressed in the paraventricular nucleus, is detected in clusters 4 and 19. It is also expressed in cluster 8 but the concomitant expression of Onecut2 and Myo5b rather suggests a main contribution of the ventro-medial nucleus. Cluster 9 gathers GABAergic neurons expressing Vipr2 which are only found in the suprachiasmatic nucleus. Ventromedial hypothalamic nucleus neurons which express the ppp1r17 gene are in cluster 13. Therefore, the snRNA-seq dataset provides spatial information, and might be used to predict the response of specific hypothalamic nuclei to TH stimulation. To favor this neuroanatomical interpretation of the data, we presented the data as violin plots for marker genes commonly used in histological analyses (Supplementary Fig. S2) (24). We also plotted the expression level of the main genes of the TH signaling pathway across the cell clusters (Fig. 1F). As expected, Thra and Thrb and the genes encoding their main coactivators and corepressors were broadly expressed, with different patterns. Only one gene encoding a TH transporter, Slc16a2 encoding MCT8, was expressed at a significant level, and its expression was restricted to a few cell clusters.

Figure 1.

Figure 1.

Open in a new tab

Single nuclei RNA-seq analysis of the TH response of hypothalamus from hypothyroid mice. A, UMAP representing the data collected from 18 658 nuclei, embedding LibTHlow and LibTHhigh and colored by cluster. Clusters are numbered by decreasing nuclei numbers and associated with main markers, as defined by differential expression analysis (see Table 1 and Supplementary Fig. S2) (24). B, Heatmap of expression for genes differentially expressed between clusters, for clusters 1-9. High expression in yellow, low expression in pink. C, Heatmap for clusters 10-21. D, Dot plot for a selection of genes marking each cluster. Genes were selected on the heatmap and to favor comparison with previously published data. E, Coincidence of neuronal population with specific hypothalamic nuclei. Sim1 is only expressed in neurons of the paraventricular nucleus (PVN) and is found in cluster 4, 8, and 19. The neurons in cluster 9 that express Vipr2 mainly belong to the suprachiasmatic nucleus (SCN) while Ppp1r17-expressing neurons mainly belong to the ventromedial nucleus (VMN). F, Violin plot summarizing the distribution in the different clusters of cells expressing the main genes of the TH signaling pathway. Thra and Thrb, respectively encode TRα1, TRβ1/2, the nuclear receptors of T3. Ncoa1 and Ncoa2 encode histone acetyltransferases that act as TR coactivators. Ncor1 and Ncor2 encode the histone-deacetylase corepressors which interact with the unliganded TR. Dio2 encodes type 2 deiodinase, which converts T4 to T3. Slc16a2 encodes the Mct8 transporter, which favors crossing of the brain-blood barrier by TH. Note that only a small fraction of oligodendrocytes (cluster 2) expresses the Thra, Thrb, Ncoa1, and Ncoa2 genes at detectable levels.

Table 1.

Characteristics of single nuclei RNA-seq clusters

Seurat cluster Number of cells Mean gene count Categorya Main markers Main locationb No. of DEG Fold-change >0.66
LibTHlow LibTHhigh Up Down
1 1842 1632 3019 GABA.n. Gad2 225 7 17
2 1306 1144 581 Oligod. Mag 57 13 8
3 1010 1068 1430 Astrocytes Slc1a3 167 34 30
4 949 886 1654 Glut.n. Il1rapl2 PVN 119 3 16
5 802 836 2968 Glut.n. Fezf1 DMN 126 10 26
6 668 658 4087 GABA.n. Lef1 39 2 8
7 579 442 3909 Glut.n. Lmx1a 38 3 12
8 408 567 3036 Glut.n. Onecut2,  Myo5b VMN 23 2 9
9 320 302 1868 GABA.n. Vipr2, Syt10 SCN 0 0 0
10 259 323 3064 GABA.n. Sst ARC 10 0 5
11 192 225 3145 Glut.n. Tbx3, Pomc ARC 7 1 4
12 221 179 2837 OPCs Pdgfra 3 0 2
13 154 155 3651 Glut.n. Tafa4 Ppp1r17 AHN VMN 0 0 0
14 150 154 967 Microglia Aif1 1 0 1
15 88 197 3569 Tanycytes Rax, Col23a1 9 1 8
16 121 160 4531 GABA.n. Npy ARC 2 0 1
17 107 114 1116 Pars tub. Tshb Pituitary 0 0 0
18 210 10 3443 Glut.n. Slc17a7 8 8 0
19 83 59 2777 GABA.n. Avp PVN 3 2 1
20 18 41 3348 Glut.n. Tac2 ARC 0 0 0
21 17 2 2576 ND 0 0 0
Total 9504 9154 653 68 134
Open in a new tab

The main markers derive from Fig. 1B-1D. Heading abbreviations: DEG, differentially expressed gene; Lib, library.

a Category abbreviations: GABA.n., GABAergic neurons; Glut.n., glutamatergic neurons; OPCs, oligodendrocytes precursor cells.

b Main location abbreviations: AHN, anterior hypothalamus nucleus; ARC, arcuate nucleus; DMN, dorsomedial nucleus; PVN, paraventricular nucleus; SCN, suprachiasmatic nucleus; VMN, ventromedial nucleus.

snRNA-seq Differential Gene Expression Analysis

Table 1 summarizes the result of an analysis that identified a total of 653 differentially expressed genes (adjusted P value <.05 DEGs) between LibTHlow and LibTHhigh in the different clusters. Complete results are in Supplementary Table S1 (24) (note that for cluster 18, the unbalance between LibTHlow and LibTHhigh contributions might bias the results). This number drops to 202 if an additional threshold is applied for the fold-change (log2 fold-change >0.66 or < −0.66; Fig. 2 and Fig. 3 for up- and downregulated genes respectively). Few genes were regulated in more than one cluster and a recent compilation indicates that only some of these (B930025P03Rik, Daam2, Fancc, Klf9, Osbpl6, Ptgds, Pdcd7) have been repeatedly identified as TH-responsive genes in other tissues or cell types (26). However, a large fraction of these DEGs (82%) was found in only one cell cluster, indicating that different cell types display different responses to TH.

Figure 2.

Figure 2.

Open in a new tab

Genes upregulated after TH treatment. Differential expression analysis identified 68 genes for which the log2 fold-change >0.66 and an adjusted P value <.05 in at least one cell cluster (dark orange). Light orange corresponds to adjusted P value <.05 and a lower log2 fold-change. Note the higher representation of cluster 3 (astrocytes).

Figure 3.

Figure 3.

Open in a new tab

Genes downregulated after TH treatment. Differential expression analysis identified 136 genes for which the log2 fold-change < −0.66 and an adjusted P value <.05 in at least one cell cluster (dark green). Light green corresponds to adjusted P value <.05 and a lower log2 fold-change.

Subclustering

The chosen resolution produces 7 large cell clusters (>103 cells) which might hide a heterogeneity in TH response. To address this possibility, we performed an additional step of clustering for these 7 clusters, optimizing again to favor the discovery of DEGs. At low resolution (0.1), this subclustering step allowed us to identify 75 additional DEGs, of which 63 were not identified in another cluster (+10%; Supplementary Table S2) (24). Interestingly, 26 of these were found in astrocytes (cluster 3), suggesting that cells within this population differ in their response to TH.

Oligodendrocytes Are Less Sensitive to TH

The capacity to detect DEGs within a cluster is expected to decrease with the number of cells included in the cluster that is considered, as the statistical power of the differential analysis decreases. To illustrate this trend, we performed subclustering at various resolution to subdivide cluster 3 in cell groups of different size. This showed that a linear relationship exists between the number of cells in a cluster, or subcluster, and the number of DEGs, and that differential expression can no longer be detected when the number of cells in a subcluster is below 200 (Supplementary Fig. S3A) (24). The same relation was observed when all clusters were plotted (Supplementary Fig. S3B) (24). However, some deviations to this trend were evident, indicating that the sensitivity to TH varies across cell types. In particular, oligodendrocytes (cluster 2) stood out as less responsive to TH stimulation, compared to GABAergic neurons (cluster 1) and astrocytes (cluster 3) (Table 1). This low sensitivity of oligodendrocytes was correlated with a reduced expression of Thra, Thrb, Ncoa1, and Ncoa2 encoding the main elements of the TH signaling pathway: the TRα1 and TRβ1/2 nuclear receptors, and their coactivators with histone acetyl transferase activity (Fig. 1F).

Comparisons With Previous Single Nuclei Analysis.

We compared our snRNA-seq data with 2 datasets, which were recently published for related experimental settings (Supplementary Table S3) (24). The first is a snRNA-seq analysis of the hypothalamus from heterozygous mice carrying the dominant-negative ThraR384C mutation, which inhibits the response to TH from early developmental stages in many tissues (27). We found very limited overlap between the lists of DEGs resulting from our experiment and this one. Notably, the analysis of mutant mice mainly pinpointed a large set of deregulated genes in oligodendrocytes that we did not find to be TH responsive in hypothyroid mice. Thus, the deregulations found in the oligodendrocytes of mutant mice are not direct consequences of altered TH signaling. They appear only after the prolonged and ubiquitous expression of the TRα1R384C receptor and might correspond to a developmental defect.

The overlap was also limited between our data set and the results of a snRNA-seq analysis performed on the neocortex of euthyroid mice that received short TH treatment at adult stage as in the present study (28). There were, however, few similarities in the TH response of glial cells (Supplementary Table S3) (24). Furthermore, when we compare this gene list to the lists present in a database gathering all recent transcriptome analysis of TH response in various tissues and cell types (26), only a few similarities were observed. Overall, we found very limited overlap between our and these datasets (97/653 genes). Therefore, most of the T3-responsive genes appear to be novel and different from those reported in other situations or cell types.

Bulk RNA-seq of Sorted Cells: Hypothalamus Astrocytes

The main drawback of snRNA-seq is the low sequencing depth/nucleus, which limits the sensitivity of the analysis in a given cell type and the capacity to identify differential expression. To overcome this limitation, we treated hypothyroid mice with TH, as above, and performed bulk RNA-seq from sorted astrocytes and GABAergic neurons. In each case, we identified the expression of more than 60% of the annotated genes, while this ratio was only 9% in our snRNA-seq, confirming that bulk analysis of sorted cells provides a much deeper analysis of the transcriptome. For astrocytes, we used magnetic sorting to isolate Glast + cells from the hypothalamus and analyzed whole cell RNA. Enrichment was evaluated a posteriori by ranking the genes according to their expression level and comparing with a previously generated whole hypothalamus RNA-seq (not shown). This highlighted a strong enrichment for Slc1a2/Glast and Gfap astrocytes markers and a depletion of neuronal and oligodendroglial markers (Gad1, Gad2, Mag). No obvious enrichment was observed for tanycytes (Rax, Col23a1) and microglia (Runx, Dock8) which also express the Slc1a2/Glast gene. In summary, this bulk RNA-seq essentially reflects TH regulation of gene expression in astrocytes. We also extended the TH treatment to 48 hours for some mice, expecting a more robust response. Although the protocols differ at several levels, bulk RNA-seq analysis confirmed most of the previous observations in astrocytes and significantly extended the list of DEGs (Supplementary Table S4) (24). Extending TH treatment to 48 hours did not lead to a stronger response (Fig. 4).

Figure 4.

Figure 4.

Open in a new tab

Bulk RNA-seq analysis of hypothalamic astrocyte response to TH stimulation. The heatmap (red for high expression, blue for low expression) presents the 50 genes with the highest fold-change among 850 DEG detected after the RNA-seq analysis of sorted hypothalamic astrocytes. Note that the expression of DEGs does not change significantly between 24 hours and 48 hours of TH treatment.

Bulk RNA-seq of Sorted Nuclei: Hypothalamus GABAergic Neuron Nuclei

In order to isolate nuclear RNA from the GABAergic neurons, we used transgenic mice expressing a nuclear fluorescent protein and ran nuclei suspensions prepared from hypothalamus in a Fluorescence-Activated Cell Sorter before RNA extraction. By contrast, this bulk RNA-seq identified a very small number of DEGs (Supplementary Table S4) (24). Few changes in gene expression were visible at whole hypothalamus level by RT-qPCR. Interestingly, these included the downregulation of the Htr3b gene, which is only expressed in a very small number of GABAergic neurons within Cluster 1 and was only identified as DEG by bulk RNA-seq (Supplementary Fig. S3) (24).

Discussion

Combining single nuclei and bulk RNA-seq analysis, we list genes whose expression is up- or downregulated when TH levels are restored in hypothyroid mice. The present atlas (Supplementary Table S1) aims to support future research on the very complex influence of T3 on hypothalamic functions. One main conclusion is that the repertoire of genes varies widely across cell types, even within subcategories of neurons. Our unsuccessful attempt to find overlaps between the TH response of cortical and hypothalamic cell types further reinforces the view that different cell types display a very different response to TH stimulation. This opens a broad avenue for later investigations that should help to understand both the molecular basis of the diverse cellular responses and their physiological consequences.

Although this is better documented in other areas of the brain, T3 has an important function during the development of the hypothalamus (29). This early influence of T3 is probably the main explanation for the divergence between our data, in which neurodevelopment is not affected, and the data previously obtained from mice in which the dominant-negative TRα1R384C receptor was present since embryonic life (27). The main defect of the hypothalamus of these mice is a defect in oligodendrocytes differentiation, which is consistent with that well-documented observation that differentiation of oligodendrocyte progenitors, which takes several days, is dependent on TH (30, 31). On the contrary, according to our results, differentiated oligodendrocytes display limited sensitivity to TH stimulation after terminal maturation and express the Thra and Thrb genes at a low level. However, the action of TH might influence the function of these cells. In particular, the TH-responsive gene Hcn2, which encodes the cyclic nucleotide-gated 2 ion channel, is known to influence myelin sheath length and conduction along myelinated axons (32).

Beyond this example, it is difficult to infer from our results the many consequences that the unveiled TH regulations might have on hypothalamic neuronal networks. This is most notably because the nuclear fraction of RNA that we analyzed contains many noncoding RNA and unannotated transcripts of unknown functions. For example, one of the main downregulated gene is BC1, which produces a noncoding RNA with elusive function (33, 34). Attempts to use gene ontology analysis to highlight pathways and cellular functions in which TH could exert a coordinated influence were for the most part unsuccessful. The exception was astrocytes, for which both single nuclei and bulk RNA-seq highlighted a peculiar sensitivity to TH stimulation. In these cells, we identified a large set of genes upregulated by TH that encode calcium or potassium transporters (GO:0034765; GO:0015079). In particular, bulk RNA-seq identified 15 genes encoding potassium channels (GO:0015079) out of 336 upregulated genes, which represents a 6.5-fold enrichment (P value: 6 × 10−9) over the expected ratio. This raises the interesting hypothesis that T3 regulates the capacity of astrocytes to rectify potassium concentration near synapses and plays a role in maintaining normal neuronal excitability. This upregulation of genes encoding potassium channels is concomitant with a negative response of genes encoding extracellular matrix components (GO:0062023; enrichment 3.2-fold; P value 10−6). This implies that TH might also regulate the capacity of astrocytes to participate in the breakdown, synthesis, and remodeling of the extracellular matrix around neurons (35). It has recently been shown that a specific type of extracellular matrix, called the perineuronal net, plays a major role in stabilizing the synapses of some hypothalamic neurons and can modulate the activity of the neuronal network (36). Thus, TH might interfere with these regulations by acting in astrocytes. These are only a few speculations suggested by these new data, and we expect that this atlas will help other investigators to elaborate additional hypotheses on the mechanism of TH action in the hypothalamus.

Limitations

The ultimate aim of this project is to unravel the molecular mechanisms by which TH regulate the functions of the hypothalamus. A prerequisite is to identify the TR target genes in all types of hypothalamic cells, that is, the genes which transcription is directly regulated by the T3-bound TRα1 and TRβ1/2 receptors. As we found an equivalent number of genes upregulated or downregulated by TH, while TRs are mainly transcription activators, many of the TH-responsive genes that we describe might not be under direct transcriptional control of TRs. An important step would be to address chromatin occupancy by TRs on a genome-wide scale and single-cell level, which is not currently possible. Our understanding of the hypothalamic response to TH would also benefit from a deeper analysis of the single nuclei. Scaling up the analysis to a much larger number of cells would increase its statistical power and identify additional differentially expressed genes in rare cell types. A time course analysis might also establish a hierarchy between early and late response, and help to better pinpoint the initial events stimulated by TH. Notably, our single nuclei analysis did not capture the negative regulation of the Trh gene expression by TH in the hypophysiotrophic neurons of the paraventricular nucleus (37), which reaches a maximum after several days of TH treatment (38). Finally, as we pooled samples from males and females, our data cannot be used to analyze the interaction between sex and TH response.

Acknowledgments

We acknowledge the contribution of SFR Biosciences (Université Claude Bernard Lyon 1, CNRS UAR3444, Inserm US8, ENS de Lyon): Nadine Aguilera and the Plateau de Biologie Expérimentale de la Souris (ANIRA-PBES) for mouse breeding; ANIRA-CYTOMETRIE and particularly Sébastien Dussurgey for nuclei sorting; and ANIRA-AGC for mouse genotyping. This work was supported by the EquipEx + Spatial-Cell-ID under the “Investissements d'avenir” program (ANR-21-ESRE-00016). We acknowledge the China Scholarship Council for a scholarship to Shijia Wu to do her PhD project in ENS de Lyon. We acknowledge funding from Agence Nationale de la Recherche (Hypothyro program; ANR-22-CE14-0026-01).

Abbreviations

DEG

differentially expressed gene

GFP

green fluorescent protein

NSB

Nuclei Suspension Buffer

RT-qPCR

quantitative reverse transcriptase polymerase chain reaction

snRNA

single nuclei RNA

T3

triiodothyronine

T4

thyroxine

TH

thyroid hormone (T3 and T4)

TR

thyroid hormone nuclear receptor

TRH

thyrotropin-releasing hormone

UMAP

uniform manifold approximations and projections

Contributor Information

Shijia Wu, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France; Shanghai Key Laboratory of Regulatory Biology, Institute of Biomedical Sciences and School of Life Sciences, East China Normal University, Dongchuan Road 500, Shanghai 200241, China.

Julien Dellinger, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Suzy Markossian, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Yves Dusabyinema, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Romain Guyot, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Sandrine Hughes, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Denise Aubert, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Marie Fackeure, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Karine Gauthier, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Benjamin Gillet, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Wenzheng Jiang, Shanghai Key Laboratory of Regulatory Biology, Institute of Biomedical Sciences and School of Life Sciences, East China Normal University, Dongchuan Road 500, Shanghai 200241, China.

Frédéric Flamant, Institut de Génomique Fonctionnelle de Lyon, UMR5242, École Normale Supérieure de Lyon, INRAE USC1370, CNRS, Lyon 69007, France.

Disclosures

The authors have nothing to disclose.

Data Availability

The raw data are available online In BioStudies (ArrayExpress), referenced E-MTAB-14810 for single nuclei RNAseq (fastq and processed files resulting of the Seurat V5 analyses), E-MTAB-15049 for bulk RNA-seq from hypothalamus astrocytes, and E-MTAB-15050 for bulk RNA-seq from hypothalamus GABAergic neurons. The scripts used to perform the snRNA-seq analyses are available at https://gitbio.ens-lyon.fr/igfl/psi/snRNAseq_hypothyro_short_reads. Supplementary figures and tables are available at figshare:  https://doi.org/10.6084/m9.figshare.28804142.v1.

References

  • 1. Pearson  CA, Placzek  M. Development of the medial hypothalamus: forming a functional hypothalamic-neurohypophyseal interface. Curr Top Dev Biol. 2013;106:49‐88. [DOI] [PubMed] [Google Scholar]
  • 2. Ortiga-Carvalho  TM, Chiamolera  MI, Pazos-Moura  CC, Wondisford  FE. Hypothalamus-pituitary-thyroid axis. Compr Physiol. 2016;6(3):1387‐1428. [DOI] [PubMed] [Google Scholar]
  • 3. Lopez  M, Alvarez  CV, Nogueiras  R, Dieguez  C. Energy balance regulation by thyroid hormones at central level. Trends Mol Med. 2013;19(7):418‐427. [DOI] [PubMed] [Google Scholar]
  • 4. Sentis  SC, Dore  R, Oelkrug  R, Kolms  B, Iwen  KA, Mittag  J. Hypothalamic thyroid hormone receptor alpha1 signaling controls body temperature. Thyroid. 2024;34(2):243‐251. [DOI] [PubMed] [Google Scholar]
  • 5. Wu  Z, Martinez  ME, DeMambro  V, Francois  M, Hernandez  A. Developmental thyroid hormone action on pro-opiomelanocortin-expressing cells programs hypothalamic BMPR1A depletion and brown fat activation. J Mol Cell Biol. 2023;14(9):mjac078. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6. Coppola  A, Liu  ZW, Andrews  ZB, et al.  A central thermogenic-like mechanism in feeding regulation: an interplay between arcuate nucleus T3 and UCP2. Cell Metab. 2007;5(1):21‐33. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7. Dardente  H, Hazlerigg  DG, Ebling  FJ. Thyroid hormone and seasonal rhythmicity. Front Endocrinol (Lausanne). 2014;5:19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. de Vries  EM, Nagel  S, Haenold  R, et al.  The role of hypothalamic NF-kappaB signaling in the response of the HPT-axis to acute inflammation in female mice. Endocrinology. 2016;157(7):2947‐2956. [DOI] [PubMed] [Google Scholar]
  • 9. Sinko  R, Mohacsik  P, Kovari  D, et al.  Different hypothalamic mechanisms control decreased circulating thyroid hormone levels in infection and fasting-induced non-thyroidal illness syndrome in male thyroid hormone action indicator mice. Thyroid. 2023;33(1):109‐118. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 10. Hernandez  A, Quignodon  L, Martinez  ME, Flamant  F, St Germain  DL. Type 3 deiodinase deficiency causes spatial and temporal alterations in brain T3 signaling that are dissociated from serum thyroid hormone levels. Endocrinology. 2010;151(11):5550‐5558. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Yen  PM, Ando  S, Feng  X, Liu  Y, Maruvada  P, Xia  X. Thyroid hormone action at the cellular, genomic and target gene levels. Mol Cell Endocrinol. 2006;246(1–2):121‐127. [DOI] [PubMed] [Google Scholar]
  • 12. Richard  S, Guyot  R, Rey-Millet  M, et al.  A pivotal genetic program controlled by thyroid hormone during the maturation of GABAergic neurons. iScience. 2020;23(3):100899. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Kriaucionis  S, Heintz  N. The nuclear DNA base 5-hydroxymethylcytosine is present in purkinje neurons and the brain. Science. 2009;324(5929):929‐930. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Galaxy  C. The Galaxy platform for accessible, reproducible, and collaborative data analyses: 2024 update. Nucleic Acids Res. 2024;52(W1):W83‐W94. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15. Langmead  B, Trapnell  C, Pop  M, Salzberg  SL. Ultrafast and memory-efficient alignment of short DNA sequences to the human genome. Genome Biol. 2009;10(3):R25. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16. Anders  S, Pyl  PT, Huber  W. HTSeq--a Python framework to work with high-throughput sequencing data. Bioinformatics. 2015;31(2):166‐169. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Love  MI, Huber  W, Anders  S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2. Genome Biol. 2014;15(12):550. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 18. Metsalu  T, Vilo  J. ClustVis: a web tool for visualizing clustering of multivariate data using Principal Component Analysis and heatmap. Nucleic Acids Res. 2015;43(W1):W566‐W570. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 19. Hao  Y, Stuart  T, Kowalski  MH, et al.  Dictionary learning for integrative, multimodal and scalable single-cell analysis. Nat Biotechnol. 2024;42(2):293‐304. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20. McGinnis  CS, Murrow  LM, Gartner  ZJ. DoubletFinder: doublet detection in single-cell RNA sequencing data using artificial nearest neighbors. Cell Syst. 2019;8(4):329‐337 e324. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 21. Korsunsky  I, Millard  N, Fan  J, et al.  Fast, sensitive and accurate integration of single-cell data with harmony. Nat Methods. 2019;16(12):1289‐1296. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 22. Steuernagel  L, Lam  BYH, Klemm  P, et al.  HypoMap-a unified single-cell gene expression atlas of the murine hypothalamus. Nat Metab. 2022;4(10):1402‐1419. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 23. Bookout  AL, Cummins  CL, Mangelsdorf  DJ, Pesola  JM, Kramer  MF. High-throughput real-time quantitative reverse transcription PCR. Curr Protoc Mol Biol. 2006;Chapter 15:Unit 15 18. [DOI] [PubMed] [Google Scholar]
  • 24. Wu  S. Suppl material “An atlas of thyroid hormone responsive genes in adult mouse hypothalamus”. 10.6084/m9.figshare.28804142.v1. [DOI] [PMC free article] [PubMed]
  • 25. Yao  Z, van Velthoven  CTJ, Kunst  M, et al.  A high-resolution transcriptomic and spatial atlas of cell types in the whole mouse brain. Nature. 2023;624(7991):317‐332. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26. Zekri  Y, Guyot  R, Flamant  F. An atlas of thyroid hormone receptors’ target genes in mouse tissues. Int J Mol Sci. 2022;23(19):11444. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 27. Sreenivasan  VKA, Dore  R, Resch  J, et al.  Single-cell RNA-based phenotyping reveals a pivotal role of thyroid hormone receptor alpha for hypothalamic development. Development. 2023;150(3):dev201228. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 28. Hochbaum  DR, Hulshof  L, Urke  A, et al.  Thyroid hormone remodels cortex to coordinate body-wide metabolism and exploration. Cell. 2024;187(20):5679‐5697.e5623. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29. Alkemade  A. Thyroid hormone and the developing hypothalamus. Front Neuroanat. 2015;9:15. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 30. Billon  N, Jolicoeur  C, Tokumoto  Y, Vennstrom  B, Raff  M. Normal timing of oligodendrocyte development depends on thyroid hormone receptor alpha 1 (TRalpha1). Embo J. 2002;21(23):6452‐6460. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Picou  F, Fauquier  T, Chatonnet  F, Flamant  F. A bimodal influence of thyroid hormone on cerebellum oligodendrocyte differentiation. Mol Endocrinol. 2012;26(4):608‐618. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 32. Swire  M, Assinck  P, McNaughton  PA, Lyons  DA, Ffrench-Constant  C, Livesey  MR. Oligodendrocyte HCN2 channels regulate myelin sheath length. J Neurosci. 2021;41(38):7954‐7964. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 33. Lewejohann  L, Skryabin  BV, Sachser  N, et al.  Role of a neuronal small non-messenger RNA: behavioural alterations in BC1 RNA-deleted mice. Behav Brain Res. 2004;154(1):273‐289. [DOI] [PubMed] [Google Scholar]
  • 34. Skryabin  BV, Sukonina  V, Jordan  U, et al.  Neuronal untranslated BC1 RNA: targeted gene elimination in mice. Mol Cell Biol. 2003;23(18):6435‐6441. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35. Zhang  N, Song  B, Bai  P, et al.  Perineuronal nets’ role in metabolism. Am J Physiol Endocrinol Metab. 2024;327(4):E411‐E421. [DOI] [PubMed] [Google Scholar]
  • 36. Beddows  CA, Shi  F, Horton  AL, et al.  Pathogenic hypothalamic extracellular matrix promotes metabolic disease. Nature. 2024;633(8031):914‐922. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Fekete  C, Lechan  RM. Central regulation of hypothalamic-pituitary-thyroid axis under physiological and pathophysiological conditions. Endocr Rev. 2014;35(2):159‐194. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. Wondisford  FE. Evaluating the hypothalamic-pituitary-thyroid (HPT) axis in mice. Methods Mol Biol. 2018;1801:155‐161. [DOI] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Data Citations

  1. Wu  S. Suppl material “An atlas of thyroid hormone responsive genes in adult mouse hypothalamus”. 10.6084/m9.figshare.28804142.v1. [DOI] [PMC free article] [PubMed]

Data Availability Statement

The raw data are available online In BioStudies (ArrayExpress), referenced E-MTAB-14810 for single nuclei RNAseq (fastq and processed files resulting of the Seurat V5 analyses), E-MTAB-15049 for bulk RNA-seq from hypothalamus astrocytes, and E-MTAB-15050 for bulk RNA-seq from hypothalamus GABAergic neurons. The scripts used to perform the snRNA-seq analyses are available at https://gitbio.ens-lyon.fr/igfl/psi/snRNAseq_hypothyro_short_reads. Supplementary figures and tables are available at figshare:  https://doi.org/10.6084/m9.figshare.28804142.v1.

Articles from Endocrinology are provided here courtesy of The Endocrine Society

RESOURCES