Nature: correlation between oscillatory signals and gene expression supporting human episodic memory coding

1  introduction

Genome wide association study (genome-wide association studies, GWAS) And human brain gene expression profiles reveal the ability to study the genetic basis of complex brain phenomena . These datasets are mainly used for non-invasive imaging research , Especially with structure MRI Or resting state fMRI The relevance of . Existing methods rely on published postmortem brain gene expression data sets , This means that neurophysiological and behavioral data do not come from the same person who provides gene expression data . This limits such approaches to determining how genes support key cognitive processes ( Such as episodic memory ) The potential impact of , It also highlights the need to develop new data sets in which individuals contribute both neurophysiological and gene expression data . Another issue that has influenced previous research is , Neurophysiological measurements , Such as resting state fMRI, Not directly related to cognitive phenomena . therefore , We have previously tried to correlate gene expression levels with oscillatory features encoded by successful memory , Because the basic role of these oscillations in supporting memory behavior has been well established in rodents and humans . These oscillation characteristics are a measure of the degree to which the memory of successful coding modulates the oscillation power in a given frequency band . They used intracranial electrodes implanted with epilepsy mapping to quantify , They were recorded when the subjects performed episodic memory tasks . We used a more than 10 Intracranial electroencephalogram in (iEEG) A large database of records , Piece together the distribution of these oscillatory signals in various regions of the brain . We identified the genes associated with these oscillatory signals , Including those genes previously involved in memory formation in rodent research , And autism spectrum disorders (autism spectrum disorder, ASD) And other genes related to cognitive impairment , And new genes as the main target of further research . However , Like other studies , This data set cannot benefit from both neurophysiological and gene expression information from the same individual .

To elucidate the link between gene expression and brain oscillations , And to determine the beneficial target of neuromodulation in the treatment of memory disorders , ad locum , We compiled an unprecedented data set , come from 16 First accepted iEEG Of human subjects , in the meantime , We use a perfect signal processing pipeline to measure the oscillation characteristics of episodic memory coding . These subjects then underwent temporal lobectomy , in the meantime , Total resection of the lateral temporal lobe allows for high-quality tissue samples , These samples are located in a common brain region from a previously obtained living record (Brodmann Area 38(BA38)) Immediately after resection . This approach allows us to compare gene expression information with that obtained from the same individuals iEEG Data association to identify genes associated with memory oscillation characteristics . Use the following different steps to prioritize ： Multivariate analysis (MVAs), Then use correlation to decompose by brain oscillation ; Connectivity of gene regulatory networks and expression and function of specific cell types / Or epigenome status ; And immunofluorescence staining . This powerful analytical approach combines iEEG Human electrophysiological data and genomic data from the same subjects , It highlights the genes that may be related to episodic memory mechanism .

We decided a priori to focus on... In this analysis BA38, Here's why ：(1) A number of studies have shown that , This region shows strong memory related oscillation characteristics ;(2) In the whole temporal lobectomy , Resection of this area is standardized , This allows the blood supply to the area to be maintained ;(3) In this particular group , Before temporal lobectomy iEEG Inspection always includes sampling from the area .

An inherent feature of our data set is that the subjects had intractable epilepsy , This raises an important warning for interpreting the results . However , Recent experiments show that , There was no significant difference in the pattern of blood oxygen level dependence caused by successful coding in epileptic patients participating in cognitive research compared with healthy controls . Besides , Since we examined the association of these individuals with the characteristics of gene oscillations , Instead of comparing with another set of data , We can establish control methods to partially explain this concern .

2  Methods and results

2.1  Characteristics of intra individual memory oscillations and generation of gene expression data sets

To determine the relationship between memory related brain oscillations and gene expression , We analyzed the subjects' iEEG, And from the same 16 Individual gene expression data . By comparing successful and unsuccessful memory patterns during encoding , From recorded iEEG The signal calculation succeeds in memorizing the oscillation characteristics of the coding ( Subsequent memory effect (subsequent memory effects, SMEs)). We use free recall tasks , This is a standard test of episodic memory that oscillatory patterns can be well described , And use our perfect signal processing pipeline ( chart 1) The oscillation characteristics are calculated . On average, , The subjects remembered 24.3% Memory items , List intrusion ( False memory ) The ratio is 5.4%. These characteristics reveal the expected pattern of free recall , Including priority and recent effects , And the time adjacency of real-time delay .

chart 1 In person study design and quality control

By first decomposing the wavelet iEEG The signal is normalized , And in 2 to 120 Hz Of 56 Statistically compare the oscillation values between successful and unsuccessful coding events on log interval frequencies , Extracted from electrodes located in the temporal pole SMES. This is done using the permutation program , That is to say, the test label shall be tested in each recording electrode 1000 Second shuffle . We decided a priori to pass the expert neuroradiology examination , To be located in the anterior temporal pole (BA38) The oscillation data of each individual on all electrodes are averaged , Because this seems to be the most common method . result , Between subjects in all frequency bands SME The variance of the value is greater than the variance within the subject , This supports the effectiveness of the method . The data averaged... Per subject 3.6 An electrode . Before inputting these data into our model to estimate gene correlation values , The resulting oscillation characteristics are averaged to six predefined frequency bands ( chart 1d, f). Between successful and unsuccessful coding , The proportion of electrodes showing a significant difference in oscillation power indicates that there is a significant memory related oscillation mode , Significant effects at the individual level are also shown ( chart 1d). Besides , Observed SME The correlation between values reveals low and high frequencies SMEs The expected relationship between . We noticed that , The observed difference may be due to narrowband oscillation or broadband power drift ( Or a mixture of the two ) Due to the functional change of . this 16 Study subjects subsequently underwent temporal lobectomy . The techniques standardized in these subjects were used from BA38 Get organization ( chart 1e). Subjects with temporal pole onset were not included in our data .

We from 16 individual BA38 The complete transcriptome is produced in the sample RNA Sequence (RNA-seq) data . except 16 A human with matched oscillatory signature measurements and gene expression data , We also started from another 11 Temporal lobectomy produced... In people who did not obtain oscillatory measurements BA38 RNA-seq data , As well as from 12 A healthy person and 8 Autopsy tissues from patients with epilepsy , To use permutations / Guide to verify our predictions ( chart 1F And methods ). Principal component analysis showed that , The gene expression of each sample is consistent , No outliers . Before further analysis , Analyze and eliminate problems caused by technology 、 Differences in interpretation of biological and sequencing covariates . These adjusted gene expression values are used to calculate the correlation of gene oscillation characteristics between individuals in each frequency band and co expression network .

2.2  Memory oscillations are associated with gene expression

To determine the relationship between memory oscillation characteristics and gene expression , We did MVA, And then use Spearman Rank correlation decomposes brain oscillations , This includes the above arrangement / guide . The correlation between gene expression and brain oscillations was carried out among subjects , Each subject contributed one gene expression value and one in each frequency band SME value .MVA A total of... Were detected 753 individual SME Rate of gene expression related wigs (FDR) correction P value <0.05( chart 2a) Genes . We identified important genes F- The statistics are robust , And compared with the non significant genes F- The statistics are large . then , We use correlation analysis to analyze the memory related oscillation signal in 6 Expression in frequency bands . stay MVA Detected 753 Among genes , Yes 300 Genes are associated with memory effects in specific frequency bands , A high proportion of them 2-4 Hz Of δ Frequency band oscillation ( chart 2a). Most of the identified genes target only one frequency band , Only a few genes are mainly caused by δ Share with another band ( chart 2b).Spearman Of ρ The value is robust , And greater than random expectations . These results further confirmed the importance of the identified genes .

chart 2 And SMEs The genes involved are different

From here 16 Personal data also includes a paradigm for controlling behavior , In this paradigm , Individuals complete simple mathematical problems , This allows us to observe the oscillatory features associated with this independent cognitive domain . We performed the same analysis as above , To test whether gene oscillatory association has specificity for memory processing . Our dataset also includes each individual BA38 Estimation of cortical thickness , These estimates are from our FreeSurfer Extracted from the handler , Enables us to perform additional comparative analysis , To find the genes associated with this measurement . We did not observe any overlap with these alternative data , All genes highlighted below using co expression network analysis are memory specific ( in other words , Gene oscillation correlation is the specificity of memory related oscillation effect ). Last , We look for genes associated with memory performance ( namely , Behavior data observed without considering any oscillation characteristics ). Only one gene associated with oscillatory characteristics overlaps with those identified in these control analyses , This strengthens the ability to examine genes - Unique memory related information obtained by oscillation feature correlation ( chart 2c).

2.3  The network perfects the molecular pathways associated with memory

We try to understand the functional properties of genes that have been identified as associated with oscillatory features encoded by successful memory . We performed consensus weighted gene co expression network analysis using gene expression and postmortem gene expression data sets from excised temporal lobe tissues (WGCNA). We put memory genes in a system level context , To identify coexpressive networks associated with brain oscillations , So as to further determine the priority of genes . We require that the identified module be robust in these multiple expression datasets , A total of 26 A module , Among them is 6 Genes associated with oscillation characteristics were significantly correlated ( chart 3a).

chart 3 Gene coexpression networks highlight the cellular processes involved in memory coding

The two modules are significantly related to the characteristics of incremental oscillation , A module has δ And low γ Oscillate , Three modules and β Oscillation characteristics are significantly correlated ( chart 3a). It is worth noting that , We did not detect a modular Association of genes associated with cortical thickness or recall scores , Genes associated with oscillations in mathematical tasks are associated with two independent modules , This further confirms that the genes related to memory coding and its network are different ( chart 3b). Besides , In three of these modules (WM4、WM12 and WM21) in , We found that SME Genes are significantly enriched . Compared with other genes , Of the three modules SME Genes show higher connectivity , This indicates that the oscillation signal related genes are BA38 Plays a central role in the transcriptome . We also observed the convergence of genes and modules related to genes related to oscillation characteristics ( chart 3c). Use different patient groups and methods , The convergence of these findings makes us more confident in the inference of the link between these genes and memory processes . And δ The oscillation signal is positively correlated to the two modules (WM4 and WM12) Rich in genes related to ion channel activity ( chart 3d). It is worth noting that ,WM4 It contains previously discovered genes related to memory oscillation signals , and WM12 It enriches the previously found synapse related modules related to memory oscillation signals ( chart 3c).

because SHANK2 yes WM12 One of the central genes , Encoding synaptic scaffold protein .SHANK2 And autism 、 Mental disability is related to schizophrenia , It is also associated with learning and memory deficits , This further confirms WM12 The key role of central genes in memory coding . It is important to , Modules related to different oscillation bands show different functional characteristics . And δ The related modules are the opposite , And β Oscillation signal (WM11 and WM22) Linked modules were significantly associated with alternative splicing and chromatin remodeling ( chart 3d). According to the previous results , We observed that both modules are enriched SME15 Genes in , The module is related to a gene that contains splicing related genes β Oscillation signals are linked ( chart 3c). These data may support alternative splicing regulation as a mechanism for oscillatory characteristic changes observed between individuals .

2.4  Memory oscillation signal module is related to neuropsychiatric disorder

Next , We studied SME Association between modules and genomic data from brain diseases . Using comprehensive transcriptional and genetic data from a variety of diseases , We use chain disequilibrium (linkage disequilibrium, LD) Fractional regression was used to evaluate the enrichment and expression of dysregulated genes in neuropsychiatric disorders GWAS Enrichment of . And δ Oscillation signal related WM4 Modules are significantly enriched in down regulated genes in autism and autism related variants ( chart 4a, b).WM12 Display and attention deficit / hyperactive disorder 、 Bipolar disorder 、 Severe depressive disorder 、 Schizophrenia and variants related to education and intelligence GWAS Showing rich characteristics ( chart 4b). most important of all , We are not memorizing related modules ( chart 4b) Any significant enrichment of epilepsy related sites was detected in , Moreover, the enrichment of non brain related features and disease-related variants is very low . We also SFARI In the genetic database ASD Related genes WM4 and WM12 Enrichment of ( chart 4c).

chart 4 SME Specific modules capture genes that regulate disorders in neuropsychiatric disorders

Next , We compared the relevant modules with those found in the meta-analysis of transcription data of neuropsychiatric disorders .WM4 and WM12 Are enriched by a serious ASD Affected modules RBFOX1. Interestingly ,RBFOX1 It's also WM12( chart 3d) A hub in , This provides further support for the role of this gene in neuropsychiatric disorders and memory .β modular WM21 Rich in schizophrenic variants ( chart 4a, b), and β modular WM22 Splicing modules rich in the effects of schizophrenia . Overall speaking ,δ and β The association between oscillatory signal related modules and memory impaired neuropsychiatric disorders provides further support for the role of these genes and pathways in episodic memory .

2.5  Modules of memory oscillation characteristics are associated with specific cell types

To establish a cell type specific association of identified related genes , We performed mononuclear tests on the tissues of six subjects RNA-seq(snRNA-seq) analysis , Four of the subjects provided oscillation data . We are right. 17632 Nuclear transcripts were sequenced , detected 11498 Unique molecular identifier (unique molecular identifiers, UMIs) and 4069 The overall median of genes . Before dimensionality reduction , We consider technical and biological covariates . We initially identified 24 Star clusters . Next , We used a publicly available from the middle temporal gyrus snRNA-seq Data sets , Our initial cluster is further defined by cell type and layer specificity . After comparison based on marker enrichment , We focus on a strong set of 20 A transcription defined cluster ( chart 5a). We defined 9 Inhibitory neurons ,8 Excitatory neurons and 3 There are three main non neuronal clusters . These clusters show high expression of known major markers of their respective cell types ( chart 5b).

chart 5 SME Specific modules enrich excitatory and inhibitory neurons

We found that , And δ Related modules WM4 and WM12 It has a strong enrichment effect on excitatory and inhibitory neurons ( chart 5c). To be specific ,WM4 and WM12 Yes Rorb+THEMIS+Fezf2+ The combination of deep excitatory neurons is highly enriched . These deep neurons are connected to the hippocampus GABA Can input the memory coding circuit . Besides ,δ Rhythmicity may come from deeper intrinsic neurons that project to other subcortical regions . therefore , These results further emphasize the importance of these deep excitatory neurons in episodic memory coding . Besides , Two module pairs SST+VIP+PVALB+ The combination of inhibitory neurons showed enrichment . Interestingly , Contains fast spikes of small proteins (PVALB) The basket cells of decisively control excitatory output , They regulate neocortex - Necessary for memory consolidation in the hippocampal circuit . meanwhile , Express somatostatin (SST) The neurons target the distal dendrites of pyramidal cells , They play a role in the synchronization of memory circuits and cerebral cortex oscillations . When SST+PVALB+ Intermediate neurons specifically inhibit pyramidal neurons ,VIP+ Neurons neither inhibit nor inhibit pyramidal neurons , And it may be related to the working memory circuit .

Besides , And δ Oscillation signal negative correlation module WM21 It is abundant for glial cells , Oligodendrocyte related genes dominate ( chart 5c), This supports the possible role of oligodendrocytes in memory circuits and neuronal synchronization . Besides , Use from ASD Or the brain tissue of Alzheimer's disease patients snRNA-seq data , We found that WM4 Significantly enriched ASD pass the civil examinations 2-4 Layer excitatory neurons and SST+ Suppresses dysregulated genes in neurons , and WM21 Significantly enriched oligodendrocyte markers affected by Alzheimer's disease . These results confirm that δ The role of oscillatory signal related modules in cognitive impairment at the cell type level .

WM4 and WM12 Are rich in δ Genes associated with oscillatory signals 、 Variants and multiple neuronal types associated with cognitive diseases . To test our method , To determine the future development goals of neuromodulation strategies for brain diseases and cell types , We're from one of them δ A central gene was selected in the module .IL1RAPL2 yes WM4 Module HUB gene , And similar IL1RAPL1 Together , Promote the formation of functional excitatory synapses and dendritic spines , And with ASD relevant . our snRNA-seq Data show that ,IL1RAPL2 stay RORB+ The expression was highest in deep excitatory neurons , But also SST+LAMP5+ It is expressed in the upper inhibitory neurons ( chart 5d). Fluorescence immunohistochemistry of independently obtained tissue sections (IHC) The analysis shows that ,IL1RAPL2 And excitatory neuron markers (CAMKII) The overlapping expression is the most , And inhibitory neuron markers (GAD67) overlap , And astrocytes (GFAP) Or oligodendrocyte markers (OLIG2) No overlap ( chart 5e, f). In addition to its role in the formation of excitatory synapses ,snRNA-seq The correlation with memory oscillation signals also shows that IL1RAPL2 It may play an important role in the regulation of human memory coding . All in all , These results highlight further research IL1RAPL2 The importance of its role in the etiology of memory and excitatory inhibitory synapses .

2.6 snATAC-seq It is revealed that transcription factors are key regulators of memory related modules

Next , We are trying to understand what transcription factors are (transcriptionfactors, TFs) Adjust the memory oscillation signal module . We performed mononuclear tests on tissues from three different subjects ATAC-SEQ(snATAC-SEQ) analysis . We assessed 22177 Chromatin state of nuclei , The median of the recognition peak is 7733 individual . We integrate snATAC-seq Data and snRNA-seq data ( chart 6a) Identify 17 Clusters marked . The distribution of nuclear proportions from the three subjects in the cluster was similar . We noticed that snRNA-seq and snATAC-seq Difference in the resolution of the data set in terms of the identified cell types ,snATAC-seq The proportion of non neural cells in the data set is very high . We speculate , This difference may be due to more of the nerve cell types RNA And express genes that cause snRNA-seq Deviation in data . in fact , The ratio of glial cells to neurons in human gray matter (GNR) stay 1.13 and 1.64 Change between .snATAC-seq Analytic GNR In line with this assumption , and snRNA-seq The figures underestimate GNR.

chart 6 snATAC-seq Emphasize adjustment SME Related modules TFs

Overall speaking , This multiomics approach enables us to detect cell type specific regulatory sites whose accessibility characteristics are consistent with cell type gene expression . Using motif analysis , We explored TF Enrichment in cell type specific regulatory sites associated with identified memory oscillation signal modules . In modules with cell type associations , Only in WM12 Motif enrichment detected in ( chart 6b). Interestingly , We found that WM12 It shows that SMAD3 Enrichment of motifs , This is a WM12 Central gene ( chart 6b). It is worth noting that , In other neuropsychiatric disorders and memory related WM12 Promoter regions of central genes were also observed SMAD3 Motif , Such as SHANK2( chart 6c). Besides ,WM12 Contains genes related to neuronal etiology , We found that SMAD3 It is mainly expressed in excitatory neurons ( chart 6d). Fluorescence of independently obtained tissue sections IHC The analysis further confirmed this result ( chart 6e,f). Overall speaking , These results highlight specific TFs Role in the regulation of chromatin status , These chromatin profiles are necessary for the expression of putative genes associated with memory oscillations , It also provides a new molecular entry point for understanding human memory .

3  Discuss

We set out to understand the genomic basis of the oscillatory patterns that support human episodic memory coding , The aim is to identify which genes are favorable targets for neuromodulation strategies to treat memory disorders . Use from 16 Data set of celebrity subjects , These include measurements of brain oscillations associated with successful episodic memory coding , And transcriptional data from the temporal pole of the same individual , We identified gene modules that link specific cell types and cell functions to memory related oscillatory features .

The oscillatory relationship of successful memory coding represents the relationship between gene regulation and memory behavior “ Intermediate steps ”. Oscillations are located in brain regions , Record them with intracranial depth electrodes , And decomposed into frequency bands with different characteristics . Link neurophysiological measurements to gene expression data , In the subsequent study of these identified genes, specific testable hypotheses are established . chart 3d The central genes described in may represent the most favorable targets for subsequent testing using animal models or other methods .

Our work reveals the molecular mechanism that leads to the oscillatory Association of successful memory coding . We observed that δ Oscillatory signals are associated with ion channel genes , And these genes tend to be expressed in oligodendrocytes , This leads to an interesting implication , That is, the generation of low-frequency oscillation patterns related to human memory processing depends at least in part on the regulation of oscillation by glial cells . This is based on our observations and mononuclear expression analysis of all subjects . This conclusion leads to the role of oligodendrocytes in learning and memory , It acts on the depolarization of membrane potential , Accelerated axonal conduction and ion channel activity , This is reflected in the positive correlation δ modular (WM4 and WM12) On . Besides , The genes expressed in these positive correlation modules δ The deep layers of excitatory neurons involved in rhythmic formation and in SST+VIP+PVALB+ Overexpression in expressed interneurons , These interneurons mediate the cortex during memory encoding - Hippocampal communication is crucial . These results further support the role of recognized genes in memory coding , In particular, the cell types that may be related to episodic memory .

We observed that δ Interesting properties of oscillation related genes , But no θ Oscillate , This is contrary to the rodent data , The latter generally indicates θ Frequency activity is associated with successful memory formation . However , In the human temporal lobe ,4-9 Oscillations outside the Hertz range usually exhibit memory related properties , Including cross frequency coupling ; therefore , Our findings are consistent with previous observations using oscillatory features encoded by human successful memory . In humans , These low-frequency oscillations represent the consistent characteristics of memory formation oscillation , Including the impact on unit activity time . In our analysis , And δ The significant expression of genes related to oscillatory characteristics may reflect the functional importance of these low-frequency components in humans .

Several features of our analysis , For example, there is a lack of enrichment of genetic variants associated with epilepsy , And data integration with epilepsy and health organizations , Let's believe , Our findings represent a more general link between gene expression and brain oscillations . Besides , In our analysis , Strict artifact rejection criteria are adopted , The electrodes located in the onset area of seizures were removed , Thus, the influence of abnormal activity on the observed oscillation signal is reduced . We also integrated several control steps into our analysis , Including adjusting gene expression values by incorporating epileptic duration . Last , We identified several key genes ( for example IL1RAPL2 and SMAD3) It has been independently demonstrated to be related to memory processing in data from non epileptic individuals and genetic rodent models . Although these correlation analyses do not imply causality , But these genes have been strictly related to statistical data 、 The high connectivity of modules related to memory oscillations and the specificity of cell type expression . Using this analytical method , We will IL1RAPL2 and SMAD3 Genomic markers defined as episodic memory , In order to further study the model system at the molecular level .

Overall speaking , In this paper, an experimental and analytical method is established , Used to deconstruct human behavioral and cognitive characteristics using integrated physiological and multigroup techniques . By integrating a single core 、 Transcriptional and epigenome data , We can determine the cell type specificity of memory related gene coexpression modules and the potential regulators of these modules . This molecular feature of human memory highlights key genes that can be further studied in model systems . We expect , This intra individual approach can be used in future research , Molecular pathways that highlight other complex human characteristics , The goal is to identify therapeutic targets , And link clinical and genomic data at the individual level . It is important to , Studies using animal and in vitro models will be necessary , To determine the memory related characteristics of the genes identified in our analysis .