MainMultimodal measurements, including the simultaneous measurements of gene expression, chromatin accessibility1,2,3 and antibody binding in single-cell4 and spatial genomics5,6 experiments, are of high importance in neurobiological investigations and modern-day genomics. We have previously devised methods to sequence full-length transcripts, alternative exons and exon combinations in single-cell and single-nuclei studies7,8,9. Here, we introduced chromatin accessibility as an additional modality to observe splicing and chromatin accessibility (assay for transposase accessible chromatin (ATAC)) simultaneously. Moreover, gene expression and ATAC have been used to define a gene’s ‘cell state’, defined as states where transcription (induction and repression) and chromatin (opening and closing) are coupled or decoupled10. However, whether such cell states can result in distinct splicing regulation remains unexplored. A recent study showed that genes can exist in distinct states based on transcriptional activity and chromatin accessibility, defined as priming, coupled-on, decoupled and coupled-off (corresponding to cell states 0, 1, 2 and 3). Li et al. defined these states as follows. Priming marks chromatin opening before transcription begins, coupled-on reflects active transcription coupled with open chromatin, decoupling marks the end of transcription, when chromatin closing and transcriptional repression are out of sync, and coupled-off indicates inactive transcription and closed chromatin10. We applied this ‘cell state’ framework to identify cell-type-specific splicing changes by cell state.Both splicing and chromatin organization distinguish cell types within a brain region and across brain regions7,9,11,12,13,14. Moreover, multiple modalities have undergone evolutionary changes and are affected in complex diseases including Alzheimer’s disease (AD)15,16,17. A key question is whether chromatin and splicing alterations reflect the same underlying processes.The brain is divided into interconnected regions that are disproportionately affected by distinct neurological diseases. The prefrontal cortex (PFC) is involved in executive and cognitive function, whereas the visual cortex involves visual inputs18,19. The PFC is known to be affected in frontotemporal dementia, AD and psychiatric disorders, whereas the visual cortex is affected in cerebral visual impairment20,21,22,23,24,25,26. These differences highlight the need to understand brain-region-specific molecular features. Macaques (Macaca mulatta), although widely used as models due to their evolutionary proximity to humans27, may not fully replicate human cell-type-specific molecular patterns. Therefore, detailed analyses of species-specific splicing and chromatin alterations across cell types is essential to assess the reliability of model organisms compared to humans. Last, both splicing and chromatin changes have been linked to AD. Although splicing data mostly come from bulk tissue16, single-cell chromatin alterations have been studied. However, it remains unclear whether cell types are equally affected in AD-specific splicing and if the most-affected cell types change between modalities.Therefore, we devised single-cell isoform RNA sequencing coupled with ATAC (ScISOr–ATAC), which measures gene expression, splicing and chromatin accessibility in the same individual cells. We used this method to show that distinct cell states (chromatin–transcriptome coupling/decoupling states) can reveal distinct splicing patterns. We then applied ScISOr–ATAC to the macaque PFC and visual cortex, macaque and human PFC and AD diseased and control PFC (Fig. 1a). To circumvent differences in statistical power between cell types, we developed downsampling software that compares statistically equal changes between cell types or conditions (Methods and Code availability).Fig. 1: ScISOr-ATAC pipeline and data overview.a, Outline of ScISOr–ATAC experimental and analysis pipeline; GEM, Gel Bead-In Emulsion; snRNA-seq, single-nucleus RNA sequencing; snATAC-seq, single-nucleus ATAC with sequencing; TSO, template switch oligo; poly(dT)VN, poly-dT primer sequence. b, Uniform manifold approximation and projection (UMAP) of macaque PFC and visual cortex (VIS) samples; ASC, astrocytes; INN, inhibitory neurons; VLMC, vascular and leptomeningeal cells; MG, microglia; OLIG, oligodendrocytes; OPCs, oligodendrocyte precursor cells; ENDC, endothelial cell. Excitatory neurons are indicated by L, IT or ET and gene markers. c, UMAP of human AD and control PFC samples. d, UMAP of human nuclei from integrated control human PFC and macaque samples. e, UMAP of macaque nuclei from integrated control human PFC and macaque samples.Full size imageWe consider multiple cell subtypes, especially subtypes of excitatory neurons. We denote excitatory neurons by cortical layer (L), intratelencephalic (IT)/extratelencephalic (ET), corticothalamic (CT) and near-projecting (NP) categories and gene markers. In macaques, we identified three main excitatory subtypes based on layer-specific marker expression of CUX2, RORB and both, together with other cortical neuron markers (Methods and Supplementary Fig. 1), termed L2–L3 IT_CUX2, L3–L5/L6 IT_RORB or L2–L4 IT_CUX2.RORB. Neuronal subtypes are transcriptionally distinct with unique synaptic properties28,29,30,31,32. In mice, CUX2 marks upper-layer neurons and regulates synaptic functions33,34, whereas RORB is highly expressed in L4 neurons and is essential for synaptic and chromatin organization35.In brain region comparisons, L3–L5/L6 IT_RORB neurons show the strongest splicing specificity, whereas L2–L4 IT_CUX2.RORB cells show the highest chromatin specificity. Between macaque and human PFC, chromatin and splicing often affect different cell types. In AD, glial cells show stronger dysregulation than neurons across both modalities. Moreover, exon inclusion varies with the chromatin–transcription cell state, which suggests that these states should be considered as a hidden variable in the analyses. In summary, chromatin and splicing show distinct contributions to within-species brain region specificity, species divergence and AD dysregulation, among distinct cell types, subtypes and chromatin–transcription cell states; however, in specific comparisons, both modalities can agree.ResultsDefinition of cell typesFrom two adult male rhesus macaques (Methods), we collected PFC and visual cortex samples guided by the Allen Brain Atlas36. Using a 10x Genomics Multiome kit, we prepared single-nucleus RNA and ATAC libraries and sequenced 293 million–385 million paired-end reads for RNA and 350 million–381 million reads for ATAC (Supplementary Fig. 1a). After downsampling reads to similar read numbers per cell and analyzing the RNA data using published tools37,38,39, we identified 36 cell types and subtypes, including astrocytes, oligodendrocytes, oligodendrocyte precursor cells, microglia, endothelial cells and various subtypes of excitatory and inhibitory neurons (Methods, Supplementary Fig. 1b and Supplementary Table 1). Overall, we found 6,858–13,710 cells per sample after filtering, with excitatory neurons being the most abundant (Supplementary Fig. 1c). Within the excitatory neurons, three subtypes stood out: (1) L3–L5/L6 IT_RORB neurons, mainly characterized by RORB expression along with IL1RAPL and MKX; (2) L2–L3 IT_CUX2 neurons, marked by CUX2, HPCAL1 and CBLN2; and (3) L2–L4 IT_CUX2.RORB neurons, which coexpress both RORB and CUX2 (Fig. 1b and Supplementary Fig. 2a). In primates, RORB excitatory neurons reside in layers L3–L5/L6, CUX2.RORB excitatory neurons reside in layers L2–L4 and CUX2 excitatory neurons reside in layers L2–L3 (refs. 40,41,42,43,44). Average numbers of RNA and ATAC unique molecular identifiers (UMIs) per cell type between PFC and visual cortex samples (Supplementary Figs. 1f,g and 2b,c) were correlated (Supplementary Figs. 1d,e and 2d,e). Analysis of ten healthy and nine AD-affected human PFC samples (Supplementary Table 2 and Methods) revealed expected brain cell types (Fig. 1c) and largely matched those in macaques. However, two excitatory neuron clusters coexpressing CUX2 and RORB (L2–L4 IT_CUX2.RORB and L2–L4 IT_CUX2.RORB.ACAP3) were rare in human samples (Fig. 1d,e), potentially due to species differences or sampling bias45.Overall, excitatory neurons were highly abundant across brain regions and species (Fig. 1d,e). To gain insight into disease and synaptic processes, we custom designed an Agilent enrichment array covering all annotated splice junctions in 3,224 macaque and 3,630 human genes (Methods). These consist of genes linked to synaptic function46, AD16, TDP43 knockdown47, autism spectrum disorder (ASD)48,49,50, schizophrenia51 and amyotrophic lateral sclerosis (ALS)52 and genes with cell-type-specific splicing patterns in our human PFC8 data (Supplementary Fig. 3a,b). We applied this enrichment array to the 10x cDNA for Oxford Nanopore Technologies (ONT) long-read sequencing (Supplementary Fig. 4). We achieved 79% to 83% on-target capture using the enrichment panel, compared to ~2% for the unenriched Illumina reads after in silico extension to the average ONT read length (Supplementary Fig. 4a). This extension artificially expands the mapped Illumina reads to the average ONT read length, enabling fair comparisons of equal length. Conservative calling of barcodes yielded 20 million–33 million perfectly matching barcoded reads per sample (Supplementary Fig. 4b). Reads were mapped to the macaque genome using minimap2 (ref. 53) and assigned to genes using scisorseqr9. We filtered spliced reads from the mapped and barcoded reads (Supplementary Fig. 4c). Spliced ONT reads mapping to the same gene were considered distinct UMIs if their edit distance was ≥4 (Methods and Supplementary Fig. 4d). ONT read lengths showed similar distributions with a median of 713 bp (Supplementary Fig. 4e). The median of long-read UMI counts varied by cell type, where the lowest was observed in oligodendrocytes (Supplementary Fig. 4f,g), whereas the three main excitatory neuron subtypes (L2–L3 IT_CUX2, L2–L4 IT_CUX2.RORB and L3–L5/L6 IT_RORB) showed similar UMI distributions (Supplementary Fig. 4h,i). The exon junction targeting before long-read sequencing removes purely intronic reads as we have shown before8. Moreover, exon-overlapping short-read and long-read UMI counts showed correlations between 0.74 and 0.77 per dataset. This suggests that the targeting process is not drastically biased to certain exons (Supplementary Fig. 4j).Region-specific splicing patterns are distinct from chromatinDifferential gene expression analysis between PFC and visual cortex revealed stronger changes in RNA splicing-related genes in excitatory neurons than in inhibitory neurons (Methods and Supplementary Fig. 5). Given their cortical importance and abundance, we tested 4,818 exons for differential exon inclusion (Δpercent spliced in (ΔΨ)) in excitatory neurons using 2 × 2 exon tests8,9,54 and a Benjamini–Yekutieli (false discovery rate (FDR)) correction55. We identified 143 significant exons (FDR