We have demonstrated differential expression of protein coding or noncoding alternative transcripts and genes across expression levels spanning over six orders of magnitude for individual layers of the mouse neocortex with what is, to our knowledge, the first deep sequencing of transcriptomes from separate mammalian neocortical layers. An interactive interface to explore the data, and links to download them, are available from Belgard et al. (2011)
. This resource should assist future studies that seek a detailed molecular and functional taxonomy of cortical layers and neuronal cell types. Our results increase by 3 to 4-fold the number of known (Lein et al., 2007
) layer-specific marker genes (Figure S3
) and furthermore introduce 66 lincRNA loci as new markers. These markers can assist studies of cortical cell types, neurodevelopment, and comparative neuroanatomy. Our data even augment known marker genes by providing a more objective grounding for their laminar classifications on the basis of quantitative expression level. They also reveal novel observations on each layer's neurological functions that lead to new lines of enquiry, for example regarding the roles of Alzheimer's disease genes or MHC genes in layers 2/3 or of mitochondrial biology in layer 5. Our findings in mouse are expected to be highly relevant to human biology owing to these species' strong similarities in brain transcriptomes (Strand et al., 2007
) and to the similarities of layer markers between mouse and ferret (Rowell et al., 2010
), which is an evolutionary outgroup to rodents and primates. Even unexpected findings, such as the significant and replicated association between coronary artery disease and layer 5 expression, may reflect genetic underpinnings of previously described clinical associations between vascular and neurological disease (Beeri et al., 2006; Santos et al., 2009
Our application of a machine learning classifier to carefully annotated high-throughput in situ hybridizations (Lein et al., 2007
) yields expression levels and predictions of laminar patterning that are based on transcripts, as well as on genes, and on noncoding loci, as well as on known genes. The expression levels assessed by RNA-seq are more sensitive to smaller differences, and these can be explored on Belgard et al. (2011)
. Results reflect the genome as a whole (except for repetitive sequence within which mapping of reads is problematic), rather than for the limited sequence targeted by microarray probe sets. This revealed numerous lincRNA transcripts, mostly novel, which were evolutionarily constrained, sometimes imprinted (Gregg et al., 2010
), and at least one that was most strongly expressed outside of cortex, opening new avenues for research into their extracortical functions. Additionally, we found transcripts from the same gene exhibiting expression divergence across neocortical layers, which should be investigated for potential physiological consequences. None of this would have been possible with currently available microarray-based methods.
Nevertheless, our approach will be limited by imperfections in dissection, and by contributions to one layer of transcripts emanating from radial processes of cells whose soma lie in another. These limitations will degrade the classifier's performance and hence will contribute to a large number of genes (56%) whose maximum predicted probabilities lie below 0.5. Nevertheless, the approach still provides at least a 10-fold difference in the relative probability of enrichment in different layers for over 10,857 (95%) classifiable genes and thus is effective at inferring transcriptional levels among mixed populations of cells in their milieu
, rather than for cells that have been sorted, purified, or microdissected (Markram et al., 2004; Molyneaux et al., 2007; Nelson et al., 2006
). Indeed, there is a recent demand for integration of neuronal, glial, and vascular interactions on a molecular and cellular level within the same neuronal structures (Neuwelt et al., 2011
). Our findings make possible future comparisons of whole transcriptomes across both isolated cell-types and cell layers that should yield further insights into the molecular components of the neuronal circuitry underlying higher brain functions. Finally, the data set shall enable us to begin to compare various species (including sauropsids and primates) in which the dorsal cortex has a less or more complex layering pattern with different levels of cellular diversity and complexity.