Part 3:Distinct Genetic Signatures Of Cortical And Subcortical Regions Associated With Human Memory
Mar 21, 2022
Contact: Audrey Hu Whatsapp/hp: 0086 13880143964 Email: audrey.hu@wecistanche.com
Performance assessment of the framework
If our unsupervised approach is valid, for the memory analysis we expect that memory genes should have a higher correlation value from the memory analysis compared with the motor analysis (i.e., sanity check; Figs. 1G, 6). Furthermore, in the top-10 memory genes, we expect a greater number of memory genes in the memory analysis than expected by chance (i.e., statistical significance; Tables 5,6), and that we find more memory genes than motor function genes (i.e., method precision; Fig. 7).
Figure 6. Bootstrapped correlation value differences for all cortical and subcortical candidate genes of memory and motor analysis. For a given memory gene, we calculated the difference between memory and motor analysis r values by subtracting motor r from memory r. If the memory r was negative, we took the negative of the difference (to get a positive value). Vice versa for the motor genes. For each cognitive function, we subsampled the number of genes used to the lowest number for calculating the bootstrapped mean difference (231 memory genes and 146 motor genes, respectively, 10,000 iterations). If the 95th percentile did not overlap with the baseline of zero, the bootstrapped difference is considered significant (p 0.05). Note that for the motor cortical analysis, no negatively correlated genes survived the threshold and thus no motor cortical (–) gene list is shown here. See Extended Data Figure 6-1 for the complete list of correlation value differences for genes used in the bootstrap analysis. denotes p 0.05.
Using the candidate gene correlation values, we show that the memory genes displayed a significant positive difference between memory analysis r value and motor analysis r-value, as the 95th percentile (whiskers) did not overlap with zero (Fig. 6; all gene correlation values used in the bootstrap analysis in Extended Data Fig. 6-1). As such, our approach performs as expected.
We found that the method was highly effective. For all memory cortical and subcortical gene lists, the probability of obtaining the number of memory genes observed was significantly above chance (Table 5; full list of memory-related and motor function-related genes that constitute the chance probability in Extended DataTable 5-1). Like-wise, for all motor cortical and subcortical gene lists, the probability of deriving the number of motor genes observed was highly significant as well.
Using the putative gene functions inferred from the literature review, we also found that the method had high precision, as the difference in top-10 candidate gene list precision scores are non-negative [except for motor sub- cortex (–), Fig. 7; calculation of precision values in Extended Data Fig. 7-1]. These results suggest that the method is valid and specific in identifying genes associated with memory and motor function.
Discussion
Taken together, our results show that cortical and subcortical regions involved in human memory possess distinct genetic signatures. These genetic signatures are in agreement with prior research in animal models of memory and were dissociable from the control of motor function. Thus, we show that the strong similarities between the spatial patterns of human brain transcriptome and the functional neuroimaging map of memory can be exploited to highlight candidate biological processes and genes associated with human memory for future experimental investigations. This may contribute to our knowledge of the functional differences of cortical and subcortical regions in healthy human memory function and memory disorders.
Presently, human memory evidence is generally derived from popular non-invasive methods such as GWAS (Well- come Trust Case Control Consortium, 2007), which identifies links between gene variants and cognition (Heck et al., 2014). However, GWAS ignores the spatially dis- tributed gene expression in the brain by solely analyzing gene variants about the brain or behavioral measures (Hawrylycz et al., 2012; Mahfouz et al., 2017). Our approach relies on the spatial pattern of gene expression and identifies genetic profiles related to human memory. Crucially, our unsupervised approach is versatile as it can reveal unprecedented insights into any human cognitive function of interest, e.g., decision making. This insight may be especially useful in the case of clinically relevant functions but with a genetic basis that is less understood, e.g., attention (ADHD) and language (dyslexia).
To identify general human memory genes that function across the brain, we compared the differences and the overlap between cortical and subcortical memory genes (Fig. 1). Particularly, this overlap comparison is supported by the existence of genes underlying memory function as a whole, as in the case of neuronally-expressed immediate- early genes (IEGs) involved in memory function (Gallo et al., 2018). IEGs are a broad class of genes that are expressed in a rapid, transient manner in response to a plethora of cellular stimuli. Of the neuron-specific IEGs, c-fos, Egr1, and arc are broadly associated with various facets of memory across both cortical and subcortical areas. For example, the blockade of hippocampal c-Fos negatively impacted spatial long-term memory (Kemp et al., 2013), and its blockade in either the hippocampus or retrosplenial cortex induced deficits in the consolidation of fear memory (Katche et al., 2010; Katche and Medina, 2017). Such genes are relevant for different subtypes of memory across both cortical and subcortical areas, which we term whole-brain general memory genes.
If there are such general memory genes whose function in memory spans the whole brain, both cortical and sub-cortical analyses should show overlapping genes. We found that the cortical and subcortical areas possess largely dis- tinct genetic profiles, as identified by gene-functional spatial correlation (Fig. 5). There was no overlap in the top-10 cortical and subcortical memory genes, with some overlap for memory genes (9.6% out of 1397 genes) and biological process gene sets (2.5% out of 118 gene sets).
At the biological process level, we found differences in cortical and subcortical memory. In the cortex, the identified gene sets included epigenetic regulation and im- immune signaling. The latter received recent interest as a central factor in the onset and progression of dementia (Litteljohn et al., 2014; Kim and Kaang, 2017; Hammond et al., 2019). In the subcortex, the identified genes are involved in neurogenesis and glial cell differentiation. Furthermore, we identified gene sets with a less understood link to memory as well. For instance, astrocytes and oligodendrocytes were recently discovered to be involved in linking glial-mediated potassium homeostasis and myelination to memory deficits (Hertz and Chen, 2016; Pepper et al., 2018). It is still unclear how myelinating oligodendrocytes may enable plasticity in memory (Pepper et al., 2018). Our work suggests that glial cell differentiation may play a complementary role in memory function, and should be further investigated for a comprehensive understanding of cellular contributions to memory. Overall, this may suggest inherent differences in the biological processes supporting cortical and subcortical memory regions. Future work may look into the interplay of these processes and clarify their differential contributions toward cortical and subcortical memory function.
At the gene level, enriched genes for cortical and sub-cortical memory were similarly distinct. Of the enriched genes that are associated with the biological processes above (in sets S and S-), a small proportion of genes (9.6%, or 135 genes) were shared between cortical and subcortical regions (Fig. 5). These genes are related to the Arp2/3 complex, GABA and AMPA ligand-gated ion channels, and SRP-dependent protein localization to the membrane. The Arp2/3 complex is necessary for the maturation of dendritic spines, hippocampal and extra- hippocampal AMPA receptors are involved in excitatory ion channels in memory, and GABA receptor subunits are part of inhibitory ion channels in memory function (Collinson et al., 2002; Freudenberg et al., 2016; Spence et al., 2016). As such, this recapitulates known literature and hints at basic requirements for general memory function. Overall, this may suggest differences due to gross cortico-subcortical differences in transcriptome profiles and function in healthy memory function and disease (Huber et al., 1986; Salmon and Filoteo, 2007). Future work may look into how the convergence and divergence between cortical and subcortical genetic profiles and how those enable cortico- subcortical-specific functions in memory.
Additionally, our approach also identified memory-associated genes with poorly understood relations to memory. For example, the MIS18BP1 gene was identified in the subcortical memory genes (Extended Data Table 2-1). This gene is required for the recruitment of centromere proteins to centromeres and allows normal chromosome segregation during mitosis (Moree et al., 2011). It is unclear whether such cell division genes play a role in memory across subcortical areas. However, the gene has been linked to hippocampal neurogenesis, which is critical for hippocampal function in memory (Shin et al., 2015; Gonçalves et al., 2016). Such lesser-known genes constitute a crucial contribution of our framework, as their im- mediate link to memory is yet to be established and should be examined in future research.
Our analyses of gene expression and neuroimaging maps are not without limitations. These include the limited sample size, the validity of a text-mining-like approach with GSEA and Gene Ontology library, and the spatial resolution of the AHBA. First, the limited donor sample size and reduced genome coverage after preprocessing may contribute to reduced power, but not statistical precision, of our approach. Although the future increase in sample size may identify more genes using this method, we found the current results to be robust as our results are significantly better than chance (i.e., statistical significance). Furthermore, the identified genes were specific to memory, as demonstrated by the precision of our framework. Second, GSEA utilizes the Gene Ontology library to identify enriched gene sets, and associates these enriched genes with the library’s ontological terms, e.g., synaptic plasticity. We concede that the Gene Ontology library is continually being extended with manual curation efforts, and thus is vulnerable to being outpaced by the deluge of recent experimental findings (Baumgartner et al., 2007; Dutkowski et al., 2013; Gaudet and Dessi- Moz, 2017). As such, it is possible that the database is incomplete and does not reflect all biological functions associated with each gene. This may lead to false negatives, where we miss genes that should be considered enriched. Nevertheless, our approach demonstrates high effectiveness (as seen in the top-10 memory and motor function genes) and the results are in concordance with known experimental literature independent of ontology libraries. Additionally, unsupervised methods of identifying- ing candidate genes always require manual curation and selection of these genes for further investigation. Third, this approach is also limited by the spatial resolution of the human brain transcriptome. Despite being the most appropriate human transcriptional atlas with its whole-genome and high-resolution whole-brain coverage, the AHBA map still has a lower resolution compared to functional imaging maps, especially in the cortex (Hawrylycz et al., 2011). As such, we expect the precision and statistical power of our approach to grow as the spatial resolution and sample size of the AHBA database increase. Furthermore, as the translation of gene mRNA into a functional product is subject to regulation, donor brain proteomes may be complementary in identifying genes linked to memory (Lubec et al., 2003; Park et al., 2006; Sjöstedt et al., 2015).
Conclusion
Here, using the Allen Institute brain transcriptional atlas and Neurosynth neuroimaging maps, we demonstrate that cortical and subcortical memory regions have distinct genetic signatures. These genetic signatures provide novel biological processes and molecular targets for the understanding of human memory function. Crucially, we hope that our unsupervised and spatially guided approach may help guide researchers toward productive gene and biological process candidates for understanding how complex cognitive functions such as memory may be enabled by the molecular components of the brain.
