Selective enrichment of TCF4 in GABAergic neurons during postnatal primate development

The transcription factor 4 ( TCF4 ) gene is essential for brain development, and its disruption causes Pitt-Hopkins syndrome. Common genetic variants in TCF4 also confer risk for schizophrenia and related psychiatric disorders. While the developmental roles of TCF4 are well established, its postnatal functions remain poorly defined, particularly during the prolonged maturation of the primate neocortex, when many neuropsychiatric symptoms first emerge. Here, we mapped cell-type-specific TCF4 expression across postnatal development in rhesus macaque neocortex using immunofluorescence, and in situ hybridization. We also re-analyzed public bulk-tissue and single-nucleus transcriptomic datasets spanning 16 brain regions in both humans and macaques. TCF4 was predominantly neuronal, with minimal expression in microglia, oligodendrocytes, and astrocytes. In late gestation, TCF4 was broadly expressed in excitatory and inhibitory neurons, but expression declined with maturation and became selectively enriched in inhibitory populations. By young adulthood, nuclear TCF4 levels were highest in somatostatin-positive and vasoactive intestinal peptide-positive interneurons, intermediate in parvalbumin interneurons (~6-fold above non-GABAergic cells), and lowest in cholecystokinin interneurons (~4-fold above non-GABAergic cells). This interneuron-subtype hierarchy was independently reproduced by single-nucleus RNA sequencing across all neocortical regions. Together, these findings position mature GABAergic interneurons as the principal site of TCF4 function in the primate neocortex, providing a cellular framework for linking TCF4 dysfunction to cortical circuit imbalance in Pitt-Hopkins syndrome and psychiatric disease.

Transcription factor 4 (TCF4, OMIM 602272 also known as ITF2, SEF2, E2-2) is a key regulator of neurodevelopment, controlling gene expression essential for neurogenesis (Fischer et al., 2014; Li et al., 2019), neuronal differentiation, and the maturation of neurons and oligodendrocytes (Phan et al., 2020; Wedel et al., 2020). Through these mechanisms, TCF4 shapes brain architecture by influencing cerebral cortical layering (Chen et al., 2016; Li et al., 2019), hippocampal development (Wang et al., 2020), and the formation of major fiber tracts (Mesman et al., 2020). The importance of TCF4 during brain development is evident in specific human diseases associated with mutations or dysregulation of this gene. Pitt-Hopkins syndrome (PTHS OMIM #610954) is the most well-defined condition directly caused by TCF4 dysfunction (Amiel et al., 2007; Brockschmidt et al., 2007; Zweier et al., 2007), arising from heterozygous mutations or deletions at chromosome 18q21.2. These genetic alterations result in TCF4 haploinsufficiency or production of defective proteins, often affecting the critical bHLH domain essential for dimerization and DNA binding. PTHS manifests with significant developmental delay, intellectual disability, limited speech, distinctive facial features, and breathing abnormalities (Pitt and Hopkins, 1978; Van Balkom et al., 2012; Teixeira et al., 2021).

Beyond its established role during development, a growing body of evidence suggests that TCF4 is also important for the mature brain’s function. In the adult brain, TCF4 is prominently expressed in the neocortex, hippocampus, cerebellum, and hypothalamic and amygdaloid nuclei, a pattern that is similar between humans and rodents (Rannals et al., 2016; Jung et al., 2018; Wang et al., 2020). The function of TCF4 in the mature brain remains poorly understood. Evidence suggests that it contributes to adult hippocampal neurogenesis and influences neuronal function and plasticity by regulating dendritic spine morphology, synaptic plasticity mechanisms such as long-term potentiation, neuronal excitability through ion channel gene regulation, and presynaptic function (Braun et al., 2020; Shariq et al., 2021; Crux et al., 2018). Through these functions, TCF4 may influence learning, memory, and sleep regulation.

The clinical importance of TCF4 in adulthood is highlighted by large-scale genetic studies that have identified variants within the TCF4 locus as major risk factors for schizophrenia, bipolar disorder, major depression, and post-traumatic stress disorder (Del-Favero et al., 2002; Cross-Disorder Group of the Psychiatric Genomics, 2013; Wray et al., 2018; Gelernter et al., 2019; Teixeira et al., 2021). To understand the role of TCF4 in the mature brain and the cellular basis of these risks, it is essential to identify the specific cells that express TCF4. Here, we investigate the cell-type-specific expression of TCF4 in the neocortex, a primary site of pathology in these psychiatric disorders. We conducted this study in the macaque brain, which serves as a closer model to the human brain than rodent models.

Rhesus macaque tissue

No animals were sacrificed specifically for this study. Brain tissue from 8 rhesus macaques (Macaca mulatta) was obtained from the Amaral laboratory repository across the following developmental stages: gestation day (GD) 151 (n = 1, male), 14 days (n = 2, one female, one male), 1 month (n = 2, males), 3 months (n = 2, males), and 5 years (n = 1, male). All tissue was cryopreserved at −80 °C using established protocols (Rosene et al., 1986) that ensure long-term stability for immunohistochemical studies.

Details of animal preparation are provided in Burette et al. (2024) (see also Gonzalez Ramirez et al., 2024). Briefly, animals were deeply anesthetized with sodium pentobarbital (50 mg/kg i.v.) and transcardially perfused with 1% followed by 4% paraformaldehyde in 0.1 M phosphate buffer (PB, pH 7.4). Brains were minimally postfixed for 6 h, cryoprotected in graded glycerol/DMSO solutions, and frozen in isopentane. Perfusion parameters were adjusted for fetal cases. Brains were sectioned coronally at 30 μm in a 1:8 series using a freezing microtome, and the cryopreserved sections were stored at −80 °C.

Hybridization chain reaction fluorescent in situ hybridization

Hybridization chain reaction fluorescent in situ hybridization (HCR-FISH) labeling kits were purchased from Molecular Instruments (Pasadena, California), and experiments were performed according to the manufacturer’s instructions. Probe sets used in this study include Tcf4 (B3 initiator), Sox9 (B2), Vglut1 (B2), and Vgat (B4) and were detected using appropriate amplifier hairpins labeled with Alexa Fluor 488, 546, or 647. If needed, sections were then processed for immunofluorescence staining (see next section). Sections were examined with Leica STELLARIS 8 FALCON microscopes.

Immunohistochemistry

Free-floating sections were rinsed twice in phosphate buffered saline (PBS, 5 min each) and then in PBS containing 0.1% Triton X-100 (PBS-T). Sections were blocked for 30 min at room temperature in PBS-T with 10% fetal bovine serum. For experiments using primary antibodies from different host species, sections were incubated overnight with all primary antibodies (Table 1). After PBS-T rinses, sections were incubated with fluorescence-conjugated secondary antibodies and counterstained with DAPI and/or NeuroTrace (Invitrogen ThermoFisher Scientific, Waltham, MA). Tyramide signal amplification (TSA) was used to label both TCF4 and proCCK antibodies, which were raised in rabbit (Wang et al., 1999). Sections were first incubated with proCCK antibody at a dilution below the detection threshold for conventional secondary antibodies (1:20,000), but sufficient for TSA detection. Following overnight incubation, sections were processed with the TSA Kit (Biotium, Fremont, CA) according to the manufacturer’s instructions, after which conventional labeling procedures were performed. Mounted sections were air-dried and coverslipped with Vectashield Plus (Vector Laboratories, H-1900). Images were acquired on a Leica STELLARIS 8 FALCON microscope.

Quantification

Cell and nuclear segmentation were performed using InstanSeg, a deep learning framework optimized for multiplexed fluorescence microscopy, implemented within QuPath (v0.5, RRID: SCR_018257). The pre-trained fluorescence model generates per-pixel embeddings that cluster pixels by individual cell identity while maintaining separation between neighboring cells, producing reliable masks. Only cells with both nucleus and cytoplasm fully detected were included in the analysis. Three sections per animal were analyzed. For each experimental set, confocal acquisition settings were optimized and maintained consistently across all sections and ages within that experiment.

Raw nuclear and cytoplasmic intensities for all labeling were computed for each cell and exported to OriginPro 2024b (OriginLab, Northampton, MA) for further analysis. Cell-type identity was assigned using marker-specific intensity thresholds: nuclear intensity for SOX9, combined nuclear and cytoplasmic intensity for NeuN, GAD, somatostatin, calretinin, and parvalbumin, and cytoplasmic intensity for CCK. Thresholds were set manually, guided by the bimodal distribution of staining intensities and confirmed by visual inspection. Because the markers used in this study exhibit a clear on/off expression pattern, threshold selection was straightforward. Thresholds were set conservatively, retaining only cells that were unambiguously marker-positive. Quantification could not be performed blind to developmental age, as animal age was readily apparent from tissue morphology.

Raw nuclear TCF4 intensities were normalized to a 0–1 scale using min-max normalization across all sections included in each analysis — either multiple sections from the same animal, or across animals of different ages the case of Figure 9. To facilitate descriptive comparison across cell types and developmental stages, normalized intensities were referenced to the section-level mean (x̄) and standard deviation (σ), and cells were assigned to one of four categories: negative-to-weak (< x̄), average (x̄ to x̄+σ), intense (x̄+σ to x̄+2σ), or very intense (> x̄+2σ).

Re-analysis of public transcriptomic datasets

Bulk-tissue developmental atlases

Two published developmental brain transcriptomic atlases were used to characterize TCF4 expression across different brain regions, developmental periods, and species. The Kang et al. (2011) dataset was obtained from the NCBI Gene Expression Omnibus (accession GSE25219). This dataset provides measurements from the Affymetrix GeneChip Human Exon 1.0 ST array for 1,340 brain tissue samples from 16 brain regions of 57 postmortem human donors, covering 15 developmental periods (P1-P15), spanning early embryonic development through middle adulthood. The transcript-level expression matrix (platform GPL5175; 17,565 transcript clusters; RMA-normalized log2 intensities) was downloaded using the GEOquery Bioconductor package (v2.78). TCF4 was identified by Affymetrix transcript cluster ID 3808854, mapped through the huex10sttranscriptcluster.db annotation package (v8.8). Median TCF4 log2 expression across all samples was 11.05 (range 8.26–12.74), significantly above the array background.

The Zhu et al. (2018) dataset was obtained from the Human Brain Evolution portal.1 This dataset comprises 460 human samples from 36 donors and 366 rhesus macaque samples from 26 donors, covering 16 brain regions across prenatal and postnatal development. TCF4 was identified via Ensembl ID ENSG00000196628.14. RPKM values were transformed to log2 [max (RPKM, 0.1)] for downstream visualization.

To ensure consistency across the two bulk-tissue datasets, we restricted our analysis to 16 anatomically matched brain regions: 11 neocortical areas (orbital prefrontal cortex, dorsolateral prefrontal cortex, ventrolateral prefrontal cortex, medial prefrontal cortex, primary motor cortex, primary somatosensory cortex, posterior inferior parietal cortex, primary auditory cortex, superior temporal cortex, inferolateral temporal cortex, and primary visual cortex), 4 subcortical regions (hippocampus, amygdala, striatum, and mediodorsal nucleus of the thalamus), and the cerebellar cortex. The analysis cohort comprised 1,281 Kang et al. (2011) human samples from 56 donors, 439 Zhu et al. (2018) human samples from 36 donors, and 366 Zhu et al. (2018) macaque samples from 26 donors. Sample developmental ages were mapped onto the 15 developmental periods defined for human development by Kang et al. (2011): P1, embryonic (4–8 postconceptional weeks [pcw]); P2, early fetal (8–10 pcw); P3, early fetal (10–13 pcw); P4, early mid-fetal (13–16 pcw); P5, early mid-fetal (16–19 pcw); P6, late mid-fetal (19–24 pcw); P7, late fetal (24–38 pcw); P8, neonatal (birth–6 months); P9, early infancy (6–12 months); P10, late infancy (1–3 years); P11, early childhood (3–6 years); P12, middle–late childhood (6–12 years); P13, adolescence (12–20 years); P14, young adulthood (20–40 years); and P15, middle adulthood (40–60 years). For the Zhu et al. (2018) dataset, the authors’ precomputed predicted period harmonization was used, which maps macaque developmental ages onto the Kang et al. (2011) human period scale. All bulk-atlas analyses were performed in R 4.5.3 using GEOquery, Biobase, AnnotationDbi, data.table, ggplot2, and patchwork.

Single-nucleus RNA-seq data mining

To evaluate TCF4 expression at a cellular resolution, we analyzed single-nucleus RNA-sequencing (snRNA-seq) data from the adult rhesus macaque brain (Chiou et al., 2023), accessed via the CELLxGENE Discover portal. Data were derived from three-level single-nucleus combinatorial-indexing RNA-seq (sci-RNA-seq3) profiling of four 10-year-old male rhesus macaques. We utilized two complementary subsets: a brain-wide atlas (1.5 million nuclei across 30 regions and 17 cell classes) and a focused GABAergic-neuron dataset (371,548 nuclei). TCF4 transcripts were identified using the Macaca mulatta Ensembl ID ENSMMUG00000012001.

For the brain-wide analysis, per-nucleus mRNA abundance was extracted from the 1.5 million-cell matrix. These data represent library-size-normalized, log-transformed counts, with each value reflecting TCF4 sequencing reads from an individual nucleus, corrected for total sequencing depth. Each nucleus was cross-referenced with its anatomical region and cell-class annotation from the original metadata. To identify enrichment patterns, we calculated the average per-nucleus value for each of the 30 regions × 17 cell classes. This included all nuclei, even those with zero detected TCF4 transcripts. The resulting 30 × 17 mean-expression matrix was visualized as a heatmap to reveal cell-type-specific and regional expression patterns.

For the comparison of cortical GABAergic versus glutamatergic cells, only those derived from the three cortical regions (frontal, parietal, temporal) were included and labeled as either GABAergic or glutamatergic neurons.

For the interneuron-subtype analysis, data from a 371,548-cell GABAergic subset was used. Since the deposited cell subcluster labels are numbered generically (“GABAergic neurons 1” through “GABAergic neurons 20”) without specific subtype names, each subcluster was identified anew by its marker-gene signature. Per-cell expression of seven canonical markers was extracted: the medial ganglionic eminence (MGE)/caudal ganglionic eminence (CGE) lineage transcription factors LIM homeobox 6 (LHX6), and the subtype markers parvalbumin (PV), somatostatin (SST), calbindin 2 (i.e., calretinin, CR), vasoactive intestinal peptide (VIP), cholecystokinin (CCK), and lysosomal-associated membrane protein 5 (LAMP5). The mean expression of each marker within each subcluster was calculated, and the six candidate subtype markers were column-z-scored across subclusters to identify the most enriched marker in each. Each subcluster was then assigned to a subtype based on its top z-scored marker, provided the top z-score was >0.3. Subclusters that did not meet this threshold were labeled “Mixed / LHX6.” The analysis was limited to cells from the four neocortical regions, resulting in 273,253 neocortical interneuron nuclei. All single-nucleus analyses were conducted in Python 3.12 using anndata (v0.12.10) and scanpy (v1.12).

TCF4 expression in the young adult macaque neocortex is predominantly neuronal

We examined TCF4 expression in non-neuronal cells within the neocortex of a five-year-old macaque. Consistent with previous reports, TCF4 immunostaining was exclusively nuclear, with staining intensity varying from weak to intense (Figure 1) (Burette et al., 2024). Intensely labeled nuclei were distributed across all cortical layers but showed notable concentration in layers 2 and 4 (Figure 1A). This characteristic bilayer distribution pattern was most prominent in the temporal cortex and absent in the motor cortex, which has a much-reduced layer 4. To avoid unnecessary repetition, we focus our narrative on the temporal cortex from this point forward, as we observed no notable differences in TCF4 expression patterns across cell types in other neocortical regions (e.g., somatosensory, motor, and visual cortices).

To examine TCF4 expression in microglia, we used TMEM119, a type I transmembrane protein that selectively labels resident microglia in their homeostatic state but does not mark peripheral macrophages (Bennett et al., 2016; Vankriekelsvenne et al., 2022). TMEM119-labeled cells were distributed uniformly throughout all cortical layers (Figure 1A). They exhibited the characteristic ramified morphology typical of microglia in their resting or surveillance state (Vidal-Itriago et al., 2022) (Figure 1A2). We did not detect TCF4 immunoreactivity in TMEM119-positive cells, suggesting that non-activated microglia have negligible or absent TCF4 protein.

Next, we investigated TCF4 expression in oligodendrocytes. We performed triple immunolabeling for TCF4, OLIG2 (a transcription factor expressed in oligodendrocytes and their precursors), and DAPI (nuclear label). OLIG2-positive nuclei were uniformly distributed across all cortical layers and showed enrichment in the white matter (Figure 2A). Most OLIG2-positive nuclei exhibited weak TCF4 immunoreactivity (Figure 2B). Quantitative analysis showed that OLIG2-positive cells displayed either negative-to-weak TCF4 intensity (defined as below average, < x̄) or average TCF4 intensity (defined between the mean and one standard deviation above the mean, x̄ to x̄+ σ) (Figure 2C). Cells exhibiting intense TCF4 labeling (defined as between one and two standard deviations above average, x̄+σ to x̄+2σ) or very intense TCF4 labeling (defined as more than two standard deviations above average, > x̄+2σ) were not positive for OLIG2.

We next examined TCF4 expression in astrocytes using in situ hybridization combined with immunofluorescence staining for SOX9, a pan-astrocyte marker. SOX9-positive astrocytes showed weaker TCF4 mRNA expression compared to neurons, which were identified by their larger size and distinctive morphology with NeuroTrace staining (Figure 3). To evaluate TCF4 protein levels in astrocytes, we conducted quintuple immunolabeling for TCF4, SOX9, GFAP, NeuN, and DAPI (Figure 4). SOX9-positive/GFAP-negative astrocytes were distributed uniformly across all cortical layers (Figure 4A). By contrast, SOX9-positive/GFAP-positive astrocytes—which provide structural support and maintain blood–brain barrier integrity—were primarily localized to superficial (layers 1 and 2) and deep (layers 5 and 6) cortical layers, the subpial region, the vicinity of major blood vessels, and adjacent white matter. Qualitative assessment showed consistently low TCF4 protein expression in SOX9-positive nuclei across all examined ages, regardless of GFAP co-expression status (Figures 4B–I). Only a small subpopulation of SOX9-positive/GFAP-negative astrocytes exhibited high TCF4 levels (see Figure 4E). Quantitative analysis confirmed the visual inspection. The results showed SOX9-positive/NeuN-negative cells have a lower TCF4 signal than the general cell population, whereas NeuN-positive/SOX9-negative cells have the highest signal (Figure 5).

A postnatal transition establishes high TCF4 expression in GABAergic neurons

To investigate the population of neurons expressing high TCF4 levels, we began by using multiplex in situ hybridization in the young adult neocortex. We detected TCF4 mRNA in both excitatory (VGLUT1-positive) and inhibitory (VGAT-positive) neurons (Figure 6). The signal intensity varied significantly from cell to cell, with inhibitory neurons generally exhibiting more TCF4 labeling than excitatory neurons.

We next tracked TCF4 expression in excitatory and inhibitory neurons across development by analyzing neocortex from late gestation (GD 151), neonatal (2 and 4 weeks old), infant (3 months), and young adult (5.5 years) macaques. We performed quadruple immunolabeling for TCF4, NeuN, GAD, and DAPI (Figures 7, 8). Initial examination of TCF4 and NeuN labeling confirmed highly variable TCF4 expression in neurons (Figures 7, 8). Across all examined ages, TCF4 was concentrated in layers 2 and 4, with neurons displaying a wide range of labeling intensities. From GD 151 to 5 years of age, overall TCF4 levels decreased.

The temporal–spatial development pattern of the neocortex allowed for the visualization of the maturational process within individual sections, with the superficial layers (Figure 7) being less mature than the deeper layers (Figure 8). At GD 151, we observed minimal GAD-positive neurons in superficial layers; consequently, most intensely TCF4-labeled neurons were NeuN-positive and GAD-negative. From 2 weeks to 3 months, the number of GAD-positive cells increased, with many exhibiting high TCF4 levels. While many NeuN-positive/GAD-negative cells also showed high TCF4 levels, most displayed low to moderate expression. By 5 years, cells with high TCF4 expression were predominantly GAD-positive.

In layer 4 at GD 151, GAD-positive neurons were common. Still, they expressed only low to moderate TCF4 levels, with intensely labeled neurons being GAD-negative (Figure 8). From 2 weeks to 3 months, the number of TCF4-labeled GAD-positive cells increased. By 5 years, the most strongly labeled cells were also GAD-positive.

Quantification of nuclear TCF4 labeling showed that TCF4 expression decreased dramatically from GD 151 to 2 weeks. At GD 151, neurons, regardless of GAD expression, showed high TCF4 levels exceeding the general cell population (Figure 9). At 2 weeks, neurons still exhibited higher TCF4 expression than the general cell population, with GAD-positive neurons expressing more TCF4 than GAD-negative neurons. From 2 weeks to 5 years, TCF4 expression in GAD-negative neurons approached levels similar to the general cell population, while GAD-positive neurons maintained higher expression than GAD-negative neurons.

Additionally, at 2 weeks, GAD-positive neurons displayed wide variability in TCF4 expression, which decreased during neocortical maturation. At 5 years, though reduced, substantial variability persisted, suggesting potential differential TCF4 expression across GABAergic neuron subtypes.

TCF4 expression varies significantly across GABAergic neuron subtypes

Our next step was to examine TCF4 expression within subcategories of GABAergic cells. We performed sextuple immunofluorescence labeling for the simultaneous visualization of TCF4 and the nuclear stain DAPI, along with established markers for key GABAergic cell types: parvalbumin (PV), somatostatin (SST), calretinin (CR), and pro-cholecystokinin (proCCK) (Figure 10). This combination of markers provided broad coverage, allowing us to assess TCF4 expression within a majority of GABAergic neurons and across their principal classifications. Quantitative analysis in the 5-year-old neocortex shows that GABAergic neurons express higher TCF4 levels than other cell types (Figure 11). SST-positive and CR-positive neurons exhibit the highest TCF4 levels, approximately 8-fold higher than cells negative for all markers. PV-positive neurons show moderate TCF4 enrichment (6.6-fold), while proCCK-positive neurons have the lowest expression among marker-positive populations, yet still 3.9-fold above baseline.

The distribution analysis revealed additional differences. Most cells for each GABAergic cell type, except proCCK-positive neurons, have an average (x̄ to x̄+ σ) or intense (x̄+σ to x̄+2σ) TCF4 labeling. CR-positive and PV-positive neurons show similar distributions, with both crossing into intense TCF4 labeling around the 40-45th percentile. SST-positive neurons show a unique distribution pattern with a sharp transition around the 50th percentile: with approximately 45% showing negative-to-weak TCF4 labeling, yet with approximately 47% displaying very intense TCF4 labeling (more than two standard deviations above average, > x̄+2σ).

Public transcriptomic datasets corroborate TCF4 developmental dynamics across the primate brain

To provide broader context for the TCF4 expression patterns observed through in situ hybridization and immunofluorescence, we examined developmental dynamics and cell-type specificity using complementary publicly available bulk-tissue and single-nucleus transcriptomic datasets.

In the human brain (Kang microarray; 1,281 samples from 56 donors across 16 brain regions), TCF4 mRNA levels peaked during mid-to-late fetal periods (P5-P6, approximately 13–24 postconceptional weeks) in both the cerebral cortex and subcortical structures. They then declined by about twofold through neonatal and early infancy periods (P8-P10) to a stable postnatal plateau maintained into adulthood (Figure 12A). The cerebellar cortex followed a distinct pattern: TCF4 expression did not decline sharply after birth. Instead, it peaked briefly between developmental stages P5 and P7 before stabilizing at levels higher than those in the cortex into adulthood. The same prenatal-peak/postnatal-decline pattern was independently observed in both human and macaque samples from the Zhu et al. (2018) RNA-seq dataset (Figures 12B,C). The human trajectories from the Kang microarray (Figure 12A) and the Zhu RNA-seq (Figure 12C), measured on two independent cohorts and different platforms, showed strong agreement, confirming the robustness of the observed developmental pattern. Direct comparison of human and macaque TCF4 trajectories within the Zhu et al. (2018) dataset showed strong cross-species conservation across matched brain regions (Figure 12D). Both species showed the prenatal peak and a postnatal decline, with the macaque trajectory exhibiting the expected faster developmental timing compared to humans. Per-region trajectory analysis confirmed that the developmental decline was consistent across neocortical, subcortical, and cerebellar regions in both species (Figures 12E,F). One exception was the mediodorsal thalamus, where macaque TCF4 expression seemed to increase after birth, unlike in humans, where it decreased. However, this apparent species difference should be interpreted with caution. Prenatal macaque sampling for MD was sparse, with only 5 samples distributed across periods 4–6, and one of these was a clear low-expression outlier (RPKM 1.5 versus 3.3 for the other sample at the same period), which likely contributes to the apparent prenatal-to-postnatal increase. Whether this reflects a genuine interspecies difference in thalamic TCF4 regulation or a sampling artifact remains unclear. Overall, these bulk transcriptomic data provide independent, multi-region, cross-species support for the developmental dynamics of TCF4 demonstrated by our immunohistochemical analysis.

Single-nucleus RNA-seq confirms brain-wide GABAergic enrichment and replicates the interneuron-subtype hierarchy

To evaluate whether the GABAergic enrichment of TCF4 extends beyond the temporal cortex examined by immunofluorescence, we analyzed publicly available snRNA-seq data from the adult rhesus macaque brain (Chiou et al., 2023). We first looked at TCF4 expression across all 30 sampled brain regions and all 17 annotated cell classes using the 1.5 million-cell subset from the Chiou et al. atlas (Figure 13A). Among the 17 cell classes, GABAergic neurons showed the highest overall mean TCF4 expression across the brain (mean log-normalized TCF4 = 0.99; 31.2% of cells TCF4-positive), followed by ependymal cells (0.97), oligodendrocyte precursor cells (0.87), and oligodendrocytes (0.79). Glutamatergic neurons ranked seventh (mean 0.72), and microglia showed low expression (mean 0.30; 8.4% of cells positive), consistent with the absence of detectable TCF4 immunoreactivity in TMEM119-positive microglia. Interestingly, medium spiny neurons, which are GABAergic projection neurons in the striatum, showed the lowest TCF4 levels of any neuronal class (mean 0.13; 4.6% positive), suggesting that the GABAergic enrichment of TCF4 is specific to telencephalic interneurons rather than a feature of all GABAergic populations.

Focusing the analysis on neurons labeled as GABAergic or glutamatergic within the three cortical regions showed consistently higher TCF4 expression in GABAergic populations across all examined regions aligning with the cell-type-specific pattern observed in our primary immunohistochemical analysis (Figure 13B).

To test whether the interneuron-subtype TCF4 hierarchy observed by immunofluorescence is also seen at the transcriptomic level, we analyzed 273,253 neocortical GABAergic nuclei from the Chiou et al. (2023) GABAergic subset. Interneuron subtypes were identified de novo using marker-gene signature analysis. The TCF4 ranking across subtypes showed VIP-positive (mean 1.47) and SST-positive (1.45) as the highest, followed by PV-positive (1.33), CCK-positive (1.17), LAMP5-positive (1.09), Mixed/LHX6 (0.68), and CR-positive (0.40) (Figure 13C). The ranking SST > PV > CCK matches the quantitative immunofluorescence data in Figure 11, independently confirmed across all neocortical regions—not just temporal cortex—in a separate cohort of four macaque donors using a different method. VIP-positive interneurons, which were not directly assessed by immunohistochemistry in this study, ranked highest in the transcriptomic ranking, consistent with rodent transcriptomic data showing TCF4 enrichment in the VIP-positive cluster (Tasic et al., 2016). The CR-positive subtype showed lower mean TCF4 than the CR-positive population in Figure 11; this likely reflects the fact that anti-calretinin immunohistochemistry labels both VIP-positive/CR-positive interneuron-specific cells (which have high TCF4 and are assigned to the VIP-positive transcriptomic subtype) and VIP-negative/CR-positive populations, whereas transcriptomic subclustering separates these two populations.

In this study, we provide a detailed analysis of TCF4 expression across cell types in the postnatal nonhuman primate neocortex. Our observations show that TCF4 is predominantly neuronal in the postnatal macaque neocortex, with little to no expression in microglia, oligodendrocytes, and most astrocytes. In late gestation, TCF4 is broadly expressed in neurons. As the neocortex matures, however, expression shifts toward GABAergic neurons, such that by young adulthood, TCF4 is most enriched in specific GABAergic interneuron subtypes. These findings provide a neuroanatomical framework for understanding TCF4’s role in cortical development and its contributions to disorders.

TCF4 expression is predominantly neuronal and low in glia during postnatal life

We observed consistently low levels of TCF4 in glial populations, including astrocytes, oligodendrocytes, and microglia. Microglia lacked detectable TCF4, consistent with findings from rodent TCF4-GFP reporter studies (Kim et al., 2020). Mature oligodendrocytes and astrocytes displayed only weak expression, suggesting that although TCF4 is required for oligodendrocyte and astrocyte precursor differentiation (Phan et al., 2020; Wedel et al., 2020; Zhang et al., 2021; Zhang et al., 2024), its importance diminishes as these cells mature. The pronounced myelin abnormalities observed in PTHS (Phan et al., 2020) likely stem from oligodendrocyte vulnerability during prenatal stages, when TCF4 expression is higher. A small subset of SOX9-positive astrocytes showed TCF4 levels comparable to GABAergic neurons, raising the possibility of specialized astrocyte populations with TCF4-dependent functions. In contrast, neurons showed robust, but heterogeneous expression, suggesting that TCF4 plays a predominantly neuronal role in the postnatal primate neocortex, while its glial functions are most critical during development.

Developmental transition to GABAergic neuron enrichment

TCF4 expression in the primate temporal cortex undergoes a striking reorganization across development. During late gestation, expression is broad, spanning nearly all neuronal lineages. Postnatally, this widespread pattern shifts: TCF4 gradually declines in excitatory neurons while remaining high and becomes selectively enriched in GABAergic interneurons. By five years of age, the “very intense” (>x̄+2σ) TCF4-positive population is almost exclusively inhibitory. This enrichment varies among subtypes. Using immunofluorescence, SST-positive and CR-positive neurons showed the highest nuclear TCF4 intensities, while PV-positive neurons showed moderate levels, and proCCK-positive neurons had the lowest. It is worth noting that anti-CR immunohistochemistry labels a heterogeneous population that includes both VIP-positive/CR-positive interneuron-selective cells, which have high TCF4 levels, and VIP-negative/CR-positive populations; transcriptomic clustering separates these two groups. The TCF4 enrichment observed in CR-positive cells by immunohistochemistry, therefore, likely reflects predominant labeling of the VIP-positive/interneuron-selective subset.

The hierarchy raises the possibility that TCF4 may have different regulatory roles across inhibitory lineages, possibly related to the varying physiological needs of somatodendritic- versus perisomatic-targeting cells. Overall, this developmental pathway reflects the extended maturation process of primate inhibitory circuits (Hashimoto et al., 2009; Hoftman and Lewis, 2011; Chung et al., 2017; Hosseini Fin et al., 2025).

The broad gestational expression likely reflects TCF4’s involvement in early neurodevelopmental processes — progenitor proliferation, neuronal differentiation, and laminar positioning (Chen et al., 2016; Li et al., 2019). In excitatory neurons, TCF4 regulates early maturation programs, including voltage-gated channel scaling and dendritic spine morphology (Rannals et al., 2016; Crux et al., 2018). As these neurons complete structural maturation, ongoing TCF4 activity appears dispensable, consistent with the postnatal decline we observed in NeuN-positive/GAD-negative cells. The postnatal interneuron enrichment may instead reflect roles in critical-period processes such as dendritic arborization and synaptic integration, potentially through TCF4’s function as a heterodimeric partner of ASCL1 and NeuroD family members (Fischer et al., 2014), and its regulation of downstream targets including voltage-gated sodium channel subunits (Rannals et al., 2016).

Transcriptomic support for TCF4 cellular distribution

The transcriptomic re-analyses agree with the immunohistochemical and in situ hybridization results on all key points while broadening their scope. The prenatal peak and postnatal decrease of TCF4 protein, as shown by immunofluorescence in macaque temporal cortex, are reflected at the mRNA level across 16 brain regions and both species in the Kang et al. (2011) and Zhu et al. (2018) bulk atlases, confirming that this pattern is a conserved, widespread feature of primate brain development rather than a property of a single cortical area examined histologically. Postnatal GABAergic enrichment of TCF4 is also evident in snRNA-seq data, where GABAergic neurons express the highest average TCF4 levels across all 30 sampled regions. The interneuron-subtype hierarchy identified by immunofluorescence (SST > PV > CCK) is mirrored at the transcriptomic level (Figure 13C). As previously noted, CR-positive interneurons detected via immunohistochemistry represent a mixed population that transcriptomic clustering separates into VIP-positive (high TCF4) and CR-positive (lower TCF4) subtypes, explaining the apparent discrepancy in their rankings. Importantly, VIP-positive interneurons, not directly evaluated by immunohistochemistry in this study, ranked highest in the snRNA-seq analysis (mean log-normalized TCF4 = 1.47), which is consistent with rodent transcriptomic data (Tasic et al., 2016). The low TCF4 signal in microglia seen by immunohistochemistry is likewise confirmed, with microglia displaying the lowest average TCF4 among major cell classes in the snRNA-seq atlas.

Interestingly, the medium spiny neuron population of the striatum, which, although GABAergic, has the lowest TCF4 levels among all neuronal classes in the snRNA-seq atlas (mean 0.13; 4.6% positive). This dissociation between neurotransmitter identity and TCF4 enrichment suggests that TCF4’s high expression in cortical interneurons may reflect their shared developmental origin from the ganglionic eminence rather than GABAergic neurotransmission per se.

For the cerebellum, different techniques produce less consistent results. Both bulk atlases show persistent TCF4 mRNA into adulthood, matching our previous immunohistochemical evidence of strong TCF4 protein in granule cells (Burette et al., 2024). However, the Chiou et al. (2023) snRNA-seq data show unexpectedly low cerebellar expression. This likely reflects the low per-cell transcript capture of the sci-RNA-seq3 platform, which systematically underestimates transcript counts in small-nuclei populations such as cerebellar granule cells (Cao et al., 2019; Ding et al., 2020), rather than genuinely low cerebellar TCF4.

Evolutionary considerations

By characterizing TCF4 expression in macaque neocortex, our study provides insight into the conservation and potential divergence of TCF4 function across species. In adult mouse cortex, TCF4 is enriched in GABAergic neurons, particularly parvalbumin-positive interneurons (Tasic et al., 2016; Ling et al., 2020), consistent with epigenomic data showing enrichment of TCF4 binding motifs within ATAC-seq peaks of parvalbumin neurons. In PTHS mouse models, Tcf4 mutations reduce the density of PV- and VIP-positive cells (Chen et al., 2023). Our primate data align with these rodent findings, suggesting evolutionary conservation of TCF4’s cell-type-specific roles within the inhibitory network, with VIP- and SST-positive interneurons consistently showing the highest TCF4 burden across both rodents and primates.

Implications for Pitt-Hopkins syndrome and psychiatric diseases

The developmental shift of TCF4 expression toward inhibitory neurons provides a cellular framework for understanding disease mechanisms. Many symptoms of Pitt-Hopkins syndrome likely originate from TCF4’s broad developmental roles during prenatal brain formation. Still, our findings suggest that its enrichment in GABAergic neurons may also contribute to postnatal phenotypes such as seizures, consistent with the established importance of inhibitory circuits in seizure regulation (Magloire et al., 2019). The preferential expression of TCF4 in mature interneurons is also relevant to psychiatric disorders linked to TCF4 variants, where cortical inhibitory dysfunction is a common pathological feature. In major depressive disorder, for example, reduced GABAergic signaling correlates with symptom severity, while normalization associates with remission (Cutler et al., 2023). Altered parvalbumin interneuron activity has been implicated in post-traumatic stress disorder (Huang et al., 2023; Su et al., 2025), and both schizophrenia and bipolar disorder are characterized by interneuron pathology, particularly affecting PV- and SST-positive populations (Taylor and Tso, 2015; Gigase et al., 2019; Krolewski et al., 2025; Morris et al., 2008; Jiang et al., 2013; Van Derveer et al., 2021; Dienel et al., 2023; Marin, 2024) Thus, the heightened expression of TCF4 in these interneuron subtypes aligns with known vulnerability points in psychiatric disease. Even subtle alterations in TCF4 dosage or function within mature interneurons may destabilize cortical excitatory–inhibitory balance, impair synaptic plasticity, and compromise cognition, thereby increasing susceptibility to neurodevelopmental and psychiatric conditions.

These observations have direct implications for Pitt-Hopkins syndrome therapeutics. Because mature interneurons — particularly VIP-, SST-, and PV-positive subtypes — represent the principal sites of TCF4 activity in the adult primate cortex, they constitute compelling cellular targets for gene reinstatement approaches. Precise dosage control will be critical, as both over- and under-expression of TCF4 can be detrimental to inhibitory circuits (Brzózka et al., 2010; Badowska et al., 2020). Strategies that harness endogenous TCF4 promoters and enhancers, or emerging gene-enhancement tools such as CRISPRa, may provide more physiologically regulated control than constitutive overexpression approaches. Recent demonstrations of successful CRISPR-based TCF4 activation (Papes et al., 2022) underscore the feasibility of this direction.

Methodological considerations and limitations

While our findings provide a detailed cellular map of TCF4 expression in the macaque neocortex, several limitations should be acknowledged. Immunofluorescence intensity may not perfectly reflect protein abundance, particularly across different developmental stages where tissue quality and antigen accessibility vary. Furthermore, although the antibody’s antigen target theoretically enables recognition of all TCF4 isoforms, differential binding affinities across isoforms are possible. Such variability could introduce systematic bias in observed staining patterns, potentially skewing our interpretation of cell-type-specific or developmental-stage-specific expression. The limited number of animals at each developmental stage reflects inherent constraints in nonhuman primate research and prevents fully-powered statistical comparisons across ages and cell subtypes. To facilitate descriptive analysis, we used categorical intensity classifications (“weak, average, intense, very intense”) as standardized descriptive tools, with thresholds defined relative to each section’s mean and variance. While these categories mirror qualitative impressions of staining intensity, they are descriptive rather than inferential and are not substitutes for statistical inference.

Future work with emerging spatial transcriptomic platforms like MERFISH, Xenium, or DART-FISH applied to macaque neocortex across development could enable the simultaneous quantification of TCF4 alongside dozens of cell-type markers at single-cell resolution, providing a more complete picture of the trajectory described here.

In summary, we show that TCF4 expression undergoes a developmental transition from broad neuronal distribution in early stages to selective enrichment in inhibitory interneurons during primate cortical maturation. Within interneurons, SST and CR populations exhibit the highest TCF4 expression, positioning these cells as key mediators of TCF4 function in the mature cortex. These results provide a neuroanatomical framework for understanding how TCF4 haploinsufficiency contributes to Pitt-Hopkins syndrome and how common variants increase psychiatric disease risk. By identifying interneuron subtypes as principal sites of TCF4 activity, this work highlights new avenues for targeted investigation of TCF4-related disease mechanisms.