A Six-Million-Cell Map of Gene Activity in the Human Prefrontal Cortex: How the PsychAD Cohort and Dreamlet Push Brain-Disease Research to Population Scale

The work builds the largest map to date of gene activity in the human dorsolateral prefrontal cortex, covering 1,494 donors and more than 6.3 million individual nuclei, including neurons, immune cells and vascular cells, with donors ranging from infants to an individual aged 108, diverse ancestries, neurotypical controls and people diagnosed with one of eight brain disorders. Single-cell studies of the human brain have traditionally been limited to relatively small numbers of individuals; this work moves the analysis into a population-scale setting so that interindividual variation and disease-related variation can be compared within one framework. Both the external story and the paper abstract give explicit scale figures: 1,494 donors, more than 6.3 million nuclei, eight diagnoses; the paper describes multiplexed samples, genotype-based demultiplexing, a unified cellular taxonomy and technical replicates.

Cross-disorder comparison reveals universal signatures enriched in basic cellular functions such as mRNA processing and protein localization, and after discounting these shared signatures, stronger genetic and transcriptomic concordance appears among Alzheimer's disease, diffuse Lewy body disease, vascular dementia and Parkinson's disease. Placing multiple neurodegenerative and neuropsychiatric diseases in one cohort under one cellular taxonomy makes the separation of shared and distinct processes less dependent on stitching together separate datasets. The abstract explicitly reports the universal signatures and the post-discounting concordance; the cohort is organized into cross-disorder, AD-phenotype and neuropsychiatric-symptom tiers, with stated evaluation of power and effect-size variability.

For Alzheimer's disease, the study characterizes transcriptomic variation among AD phenotypes distinct from healthy ageing, reports reduced neuronal abundance together with increased immune and vascular cell populations in more severe AD, and finds increased abundance of deep-layer excitatory neurons associated with a broad range of neuropsychiatric symptoms. By combining pathological and cognitive measures (CERAD plaque density, Braak staging, cognitive impairment) with cell composition and transcriptome trajectories, the analysis separates disease progression from normal ageing. The abstract and main text state that the AD phenotype analysis uses a subset of 696 individuals and the neuropsychiatric symptom survey uses 234 individuals with AD; conclusions rest on compositional analysis and disease trajectory modelling.

The companion paper introduces and evaluates Dreamlet, an open-source R package that performs cross-subject differential expression analysis using a pseudobulk approach based on precision-weighted linear mixed models, reporting computational and statistical performance on real data. Existing methods were designed for small to moderate datasets and struggle to jointly handle repeated measures, high-dimensional batch effects from sample multiplexing and the compute cost of large-scale data; Dreamlet is designed around these constraints. The paper reports pseudobulk for 1,000 donors across eight cell types and 4.1 million cells in 31 minutes using 26 Gb of memory, and analysis of 12 cell types across 326 subjects in 45 CPU minutes; false positive rates were assessed with a simulation pipeline from an independent group and with permuted disease labels in real postmortem brain data.

The atlas focuses on one brain region, the dorsolateral prefrontal cortex, and the authors state that additional regions will be needed to understand the full picture; the findings apply to the single-nucleus transcriptome level in postmortem donated brain tissue and are meant to generate mechanistic hypotheses and target leads rather than to guide clinical diagnosis or treatment decisions. Dreamlet targets cross-subject differences in mean expression and is suited to large cohorts with repeated measures and multiplexing-related batch effects.

The paper itself notes that technical batch effects can be substantial for some genes and that measurement precision varies widely across cell clusters, so a finding that a gene is differentially expressed in only one cell cluster may reflect cell-type-specific disease biology but could also reflect lower power in other clusters; the authors recommend caution in interpreting cell-type-specific findings. In addition, the molecular role of PTPRG in Alzheimer's disease is unclear, and genome-wide association studies do not identify risk variants in the region of the gene, suggesting that its up-regulation is reactive and may vary with disease stage and progression; pseudobulk approaches do not retain within-sample expression distributions, so transcriptional bursting and continuous expression gradients require single-cell-level analyses. This evidence bundle reflects abstract-level reading without figures or supplementary material, so specific effect sizes, full subtype-level results and statistical details still need to be checked against the original articles.