Biology subjects

Zhou, C.

Publications and source records attributed to Zhou, C..

At least 163 records · Page 9Linked to original sources

bioRxiv2022

The effect of subconjunctival TNF-α inhibitors on retinal ganglion cell apoptosis and optic nerve degeneration after corneal surgery or trauma--of possible prophylactic value against secondary glaucoma.

AbstractO_ST_ABSBackground and PurposeC_ST_ABSLate secondary glaucoma is an often-severe complication after acute events like anterior segment surgery, trauma, infection, etc. TNF- is a major mediator that is rapidly upregulated, diffusing also to the retina and causes apoptosis of the ganglion cells and degeneration of their optic nerve axons (mediating steps to glaucomatous damage). Anti-TNF- antibodies are in animals very effective in protecting the retinal cells and the optic nerve--and might therefore be useful prophylactically against secondary glaucoma in future such patients.Here we evaluate 1) toxicity and 2) efficacy of two TNF- inhibitors (adalimumab and infliximab), in rabbits by subconjunctival administration. MethodsFor drug toxicity, animals with normal, unburned corneas were injected with adalimumab (0.4, 4, or 40 mg), or infliximab (1, 10, or 100 mg). For drug efficacy, other animals were subjected to alkali burn before such injection, or steroids (for control). The rabbits were evaluated clinically with slit lamp and photography, electroretinography, optical coherence tomography, and intraocular pressure manometry. A sub-set of eyes were stained ex vivo after 3 days for retinal cell apoptosis (TUNEL). In other experiments the optic nerves were evaluated by paraphenylenediamine staining after 50 or 90 days. Loss of retinal cells and optic nerve degeneration were quantified. ResultsSubconjunctival administration of 0.4 mg or 4.0 mg adalimumab were well tolerated, whereas 40.0 mg was toxic to the retina. 1, 10, or 100 mg infliximab were also well tolerated.Analysis of the optic nerve axons after 50 days confirmed the safety of 4.0 mg adalimumab and of 100 mg infliximab. For efficacy, 4.0 mg adalimumab subconjunctivally in 0.08 mL provided practically full protection against retinal cell apoptosis 3 days following alkali burn, and infliximab 100 mg only slightly less. At 90 days following burn injury, control optic nerves showed about 50% axon loss as compared to 8% in the adalimumab treatment group. ConclusionsSubconjunctival injection of 4.0 mg adalimumab in rabbits shows no eye toxicity and provides excellent neuroprotection, both short (3 days) and long-term (90 days). Ourtotal accumulated data from several of our studies, combined with the present paper, suggest that corneal injuries, including surgery, might benefit from routine administration of anti-TNF- biologics to reduce inflammation and future secondary glaucoma.

pathology↗

bioRxiv2022

Reconstruct a eukaryotic chromosome arm by de novo design and synthesis

The genome of an organism is inherited from its ancestor and keeps evolving over time, however, how much the current version could be altered remains unknown. Here, we use the left arm of chromosome XII (chrXIIL) as an example to probe the genome plasticity in Saccharomyces cerevisiae. A neochromosome was designed to harbor originally dispersed genes. The essentiality of sequences in chrXIIL was dissected by targeted DNA removal, chromosome truncation and random deletion. Notably, 12 genes were sufficient for survival, while 25 genes are required to retain robust fitness. Next, we demonstrated these genes could be reconstructed using synthetic regulatory sequences and recoded open-reading frames with "one-amino-acid-one-codon" strategy. Finally, we built a neochromsome, which could substitute for chrXIIL for cell viability, with these reconstructed genes. Our work not only highlights the high plasticity of yeast genome, but also illustrates the possibility of making functional chromosomes with completely artificial sequences. HIGHLIGHTSO_LIA neochromosome was designed to facilitate the assembly of exogenous DNA for stable expression in yeast C_LIO_LIThe left arm of chrXII could be minimized to just 12 genes to maintain viability, but additional genes were required to retain robust fitness C_LIO_LIComprehensive recoding and transcriptional refactoring using artificial regulatory sequences produced a functional chromosome arm C_LIO_LIA completely reconstructed neochromosome could replace the chrXIIL to maintain comparable fitness C_LI

synthetic biology↗

bioRxiv2022

Creating an atlas of the bone microenvironment during oral inflammatory-related bone disease using single-cell profiling

Oral inflammatory diseases such as apical periodontitis are common bacterial infectious diseases that may affect the periapical alveolar bone tissues. A protective healing process occurs simultaneously with the inflammatory tissue destruction, in which mesenchymal stem cells (MSCs) play a primary role. However, a systematic and precise description of the cellular and molecular composition of the microenvironment of bone affected by inflammation is lacking. In this study, we created a single cell atlas of cell populations that compose alveolar bone in healthy and inflammatory disease states. We investigated changes in expression frequency and patterns related to apical periodontitis, as well as the interactions between MSCs and immunocytes. Our results highlight an enhanced self-supporting network and osteogenic potential within MSCs during apical periodontitis-associated inflammation. MSCs not only differentiated towards osteoblast lineage cells, but also expressed higher levels of osteogenic related markers, including SPARC and Col11. This was confirmed by lineage tracing in transgenic mouse models and human samples from oral inflammatory-related alveolar bone lesions. In summary, the current study provides an in-depth description of the microenvironment of MSCs and immunocytes in both healthy and disease states. We also identified key apical periodontitis-associated MSC subclusters and their biomarkers, which could further our understanding of the healing process and the underlying mechanisms of oral inflammatory-related bone disease. Taken together, these results enhance our understanding of heterogeneity and cellular interactions of alveolar bone cells under pathogenic and inflammatory conditions. We provide these data as a tool for investigators not only to better appreciate the repertoire of progenitors that are stress responsive but importantly to help design new therapeutic targets to restore bone lesions caused by apical periodontitis and other inflammatory-related bone diseases.

cell biology↗

bioRxiv2022

MOAHIT: A Web Tool for visualizing tumormulti-omics data with human anatomy heatmaps

Multi-omics data plays an important role in cancer research, helping clinicians to better explore drug targets and biomarkers. At present, there exist several databases including TCGA (the Cancer Genome Atlas) and GDSC (Genomics of Drug Sensitivity in Cancer), which contain multi omics data on multiple cancer species, as well as various web tools for analyzing the multi-omics data, which are widely used in oncology research. Tumor heterogeneity is a widespread phenomenon, reflected by differences in multi-omics data of cancers originating from different tissues and organs. However, there is a lack of convenient analysis and visualization tools to explore tumor heterogeneity. As a result, we developed a web tool called Multi-Omics Anatomy Heatmap in Tumors (MOAHIT) based on shiny programming. This tool enables users to analyze and visualize the heterogeneity of human tissues. In addition, MOAHIT enables users to produce a beautiful human anatomy heatmap to visualize the heterogeneity between distinct cancers in the multi-omics data.

bioinformatics↗

bioRxiv2022

Astrocytic TDP-43 dysregulation impairs memory by modulating antiviral pathways and interferon-inducible chemokines

TDP-43 pathology is prevalent in dementia but the cell type-specific effects of TDP-43 are not clear and therapeutic strategies to alleviate TDP-43-linked cognitive decline are lacking. We found that patients with Alzheimers disease (AD) or frontotemporal dementia (FTD) have aberrant TDP-43 accumulation in hippocampal astrocytes. In mouse models, induction of widespread or hippocampus-targeted accumulation in astrocytic TDP-43 caused progressive memory loss and localized changes in antiviral gene expression. These changes were cell-autonomous and correlated with impaired astrocytic defense against infectious viruses. Among the changes, astrocytes had elevated levels of interferon-inducible chemokines and neurons had elevated levels of the corresponding chemokine receptor CXCR3 in presynaptic terminals. CXCR3 stimulation altered presynaptic function and promoted neuronal hyperexcitability, akin to the effects of astrocytic TDP-43, and blockade of CXCR3 reduced this activity. Ablation of CXCR3 also prevented TDP-43-linked memory loss. Thus, astrocytic TDP-43 dysfunction contributes to cognitive impairment through aberrant chemokine-mediated astrocytic-neuronal interactions. SummaryIn dementia, protein buildup in glia enhances chemokine signaling to synapses and impairs specific aspects of neurocognitive function.

neuroscience↗

bioRxiv2022

Synaptic degeneration in neuronal circuits hinders memory recall, memory rescue, and learning

Neurodegeneration is a characteristic of Alzheimers disease (AD), but its effect on neural activity dynamics underlying memory deficits is unclear. Here, we studied the effects of synaptic degeneration on neural activities associated with memory recall, memory rescue by slow-gamma stimulation, and learning a new memory, in an integrate-and-fire neuronal network. Our results showed that reducing connectivity decreases the neuronal synchronisation of memory neurons at slow-gamma frequencies and impairs memory recall performance. Although slow-gamma stimulation rescued memory recall and slow-gamma oscillations, the rescue with reduced connectivity caused a side effect of activating mixed memories. During the learning of a new memory, reducing connectivity caused an impairment in storing the new memory, but did not affect previously stored memories. Our results reveal potential computational mechanisms underlying the memory deficits caused by synaptic degeneration in AD.

neuroscience↗

bioRxiv2022

Capturing the Hierarchically Assorted Modules of Protein Interaction in the Organized Nucleome

Nuclear proteins are major constituents and key regulators of the topological organization of nucleome. To elucidate the global connectivity of nucleomic proteins and to decipher the hierarchically organized modules of protein interaction that are involved in nucleomic organization and nuclear events, both formaldehyde and CBDPS crosslinkers were applied sequentially on the in vivo prefixed nuclei to perform a double chemical crosslinking with mass spectrometry (XL-MS) analysis. The integration of dimethyl-labelling with XL-MS generated a quantitative XL-MS workflow (qXL-MS) that consequently identified 5,340 cross-linked peptides (crosslinks) from nucleome. These crosslinks were construed into 1,297 nuclear protein-protein interactions (PPIs), from which discovered were 250 and 26 novel interactors of histones and nucleolar box C/D snoRNP complex, respectively. MONET-based modulomic analysis of their Arabidopsis orthoglous PPIs constructed 27 and 24 master nuclear protein interaction modules (NPIMs) that contain the condensate-forming protein(s) and the intrinsically disordered region (IDR)-containing proteins, respectively. These NPIMs successfully captured the previously reported nuclear protein complexes and nuclear bodies in nucleome. Surprisingly, modulomic analysis showed that these NPIMs were hierarchically assorted into four communities of NPIMs in nucleome including Genome Community and Nucleolus Community. The qXL-MS-based quantitative interactomics finally revealed 17 Hormone-specific module variants participating in a broad range of nuclear events. Thus, this integrated pipeline of qXL-MS and MONET modulomics, named as CHAMPION, is capable of capturing both nuclear protein complexes and nuclear bodies, constructing the topological architecture of protein interaction modules and module variants in nucleome and probably of mapping the protein compositions of condensates. HighlightsO_LIThe formaldehyde and CBDPS crosslinkers coupled qXL-MS discovered 5,340 crosslinked peptides. These crosslinks were construed into 1,297 nuclear protein-protein interactions (PPIs), protein components of which contained 250 and 26 novel interactors of histone octamer and nucleolar box C/D snoRNP complex, respectively, in the intricately organized nucleome. C_LIO_LIThe MONET-based modulomic analysis of these crosslinks captured 95 nuclear protein interaction modules (NPIMs), a portion of which contain both the condensate-forming and the intrinsically disordered region (IDR)-containing proteins. Especially, some NPIMs captured 6 previously reported nuclear protein complexes. C_LIO_LIA number of Hormone-specific module variants were identified by modulomics upon hormone treatment using the hormone significantly up-regulated crosslinks from qXL-MS. Several PPIs and NPIMs have been substantiated with alternative biological experiments. C_LIO_LIThis CHAMPION pipeline has partitioned these NPIMs into four hierarchically and topologically organized communities in nucleome. The molecular functions of those proteins partitioned into C1 and C2 community are specialized in genome organization and nucleolar functions, respectively. C_LI

bioinformatics↗

bioRxiv2022

Flnc: Machine Learning Improves the Identification of Novel Full-length Long Noncoding RNAs from RNA Sequencing Data Without Transcriptional Initiation Profiles

Long noncoding RNAs (lncRNAs) play critical regulatory roles in human development and disease. However, many lncRNAs have yet to be annotated. The conventional approach to identifying novel lncRNAs from RNA sequencing (RNA-seq) data is to find transcripts without coding potential. This approach has a false discovery rate of 30-75%. The majority of these misidentified lncRNAs are RNA fragments or transcriptional noise and lack defined transcription start sites, which are marked by H3K4me3 histone modifications. Therefore, the accuracy of lncRNA identification can be improved by incorporating H3K4me3 chromatin immunoprecipitation sequencing (ChIP-seq) data. However, because of cost, time, and limited sample availability, most RNA-seq data lacks such data. This paucity of H3K4me3 data greatly hinders the efforts to accurately identify novel lncRNAs. To address this problem, we have developed software, Flnc, to identify both novel and annotated full-length lncRNAs from RNA-seq data without H3K4me3 profiles. Flnc integrates machine learning models built incorporating four types of features: transcript length, promoter signature, multiple exons, and genomic location. Flnc achieves state-of-the-art prediction power with an AUROC score over 0.92. Flnc significantly improves the prediction accuracy from less than 50% using the conventional approach to over 85%. Flnc is available via https://github.com/CZhouLab/Flnc.

bioinformatics↗

bioRxiv2022

Multi-omics integrated analysis reveals a specific phenotype of CD8+ T cell may contribute to immunothromosis via Th17 response in severe and critical COVID-19

T lymphocyte reduction and immunosenescence frequently occur in severe and critical coronavirus disease 2019 (COVID-19) patients, which may cause immunothrombosis and numerous sequelae. This study integrated analyzed multi-omics data from healthy donors, pneumonia, COVID-19 patients (mild & moderate, severe, and critical), and convalescences, including clinical, laboratory test, PBMC bulk RNA-seq, PBMC scRNA-seq and TCR-seq, BAL scRNA-seq, and lung proteome. We revealed that there are certain associations among T lymphocyte reduction, CD8+ T cell senescence, Th17 immune activation, and immunothrombosis. A specific phenotype (S. P.) CD8+ T cells were identified in severe and critical COVID-19 patients in both PBMC and BAL scRNA-seq, which showed highly TCR homology with terminal effector CD8+ T cells and senescent CD8+ T cells. Pseudotime analysis showed that the S. P. CD8+ T cells were located in the transition trajectory from mild to severe disease. Which may be activated by terminal effector CD8+ T cells or senescent CD8+ T cells, thereby promoting Th17 cell differentiation. This phenomenon was absent in healthy donors, mild and moderate COVID-19 patients, or convalescences. Our findings are an important reference for avoiding the conversion of patients with mild to severe diseases and provide insight into the future prevention and control of COVID-19 and its variants.

bioinformatics↗

bioRxiv2022

Cognitive aging and categorical representations in visual working memory

A traditional view on cognitive aging is that visual working memory (VWM) capacity declines in older adults. Recent work has shown that visual information can be stored in VWM in different forms of representations. Specifically, VWM becomes more reliant on categorical representations (e.g., a prototypical red) as compared to continuous representations (e.g., a light reddish color) as memory load increases. Here, we replicated these findings and tested whether this holds for older adults. Participants memorized one to four colors; after a delay, an arrow pointed at the location of the color that needed to be reported. We used an extended mixture model (Zhou et al., 2022) to examine the extent to which memory responses were biased in the direction of the category prototypes. Our results showed that for both younger and older adults, VWM became more biased towards category prototypes with increasing memory load. Importantly, we found no difference in the extent to which VWM was biased towards category prototypes between younger and older adults. However, older adults showed an overall lower precision as compared to younger adults. Taken together, our results demonstrated that both younger and older adults became more reliant on categorical representations with increasing memory load; importantly, the extent to which categorical representations were involved in maintaining VWM information was insensitive to age.

neuroscience↗

bioRxiv2022

Structural characterization of human RPA70N association with DNA damage response proteins

The heterotrimeric Replication protein A (RPA) is the ubiquitous eukaryotic single-stranded DNA (ssDNA) binding protein and participates in nearly all aspects of DNA metabolism, especially DNA damage response. The N-terminal OB domain of the RPA70 subunit (RPA70N) is a major protein-protein interaction element for RPA. Previous crystallography studies of RPA70N with p53, DNA2 and PrimPol fragments revealed that RPA70N binds to amphipathic peptides that mimics ssDNA. NMR chemical-shift studies also provided valuable information of RPA70N residues interacting with target sequences. However, it is still not clear how RPA70N recognizes and distinguishes such a diverse group of proteins. Here we present high resolution crystal structures of RPA70N in complex with peptides from HelB, ATRIP, RMI1, WRN and BLM. The structures showed that in addition to the ssDNA mimicry mode of interaction, RPA70N employs multiple ways to bind its partners, some of which may serve to increase the avidity of RPA70N binding.

biophysics↗

bioRxiv2022

Selective Regulation of Tuft Cell-Like Small Cell Lung Cancer by Novel Transcriptional Co-activators C11orf53 and COLCA2

The DepMap project has generated a huge resource for investigating selectively essential genes, which represent potential cancer therapeutic targets. However, manually sorting out which of the hundreds of selectively essential genes is understudied and warrants investigations is time-consuming and probably not practical. To efficiently identify uncharacterized, selectively essential genes, we collected and ranked the 347 selectively essential genes from the DepMap dataset by their PubMed publication counts, based on the assumption that genes with low publication counts are un-studied or under-studied. We successfully validated two of the top candidates in our ranking system, C11orf53 and COLCA2, as new vulnerabilities that are selectively essential in the class II POU domain transcription factor POU2F3-dependent tuft cell-like small cell lung cancer (SCLC) cell lines. Importantly, we found that the sequence motif, which mediates physical interactions of the transcriptional co-activator POU2AF1 with two of the class II POU domain-containing family of transcription factors (POU2F1 and POU2F2), is also present in the N-terminal regions of C11orf53 and COLCA2. We further confirmed that 1) COLCA2 physically interacts with POU2F3 through this conserved sequence motif; 2) this interaction is important for COLCA2 to regulate tuft cell-like SCLC cell growth, and 3) both C11orf53 and COLCA2 contain transcriptional co-activator domains. Consistently, we find similar transcriptomic changes in response to the loss of COLCA2 or POU2F3 in SCLC cells. In summary, our analysis pipeline enables identification and prioritization of understudied but important, selectively essential genes, leading to the identification of two new transcriptional co-activators for the class II POU domain transcription factors. Disruption of this important physical interaction is predicted to be a potential therapeutic strategy to selectively inhibit tuft cell-like SCLCs.

molecular biology↗

bioRxiv2022

Kynurenic acid promotes A1 to A2 astrocyte conversion against HIV-associated neurocognitive disorder

Although extensive astrocyte activation is present in the CNS of patients suffering from HIV-associated neurocognitive disorder (HAND), little is known about the contribution of reactive astrocytes to HAND neuropathology. Here, we report that the imbalance between neurotoxic (A1 astrocytes) and protective astrocytes (A2 astrocytes) occurring in HIV-1 gp120 transgenic mice impedes the resolution of inflammation in the CNS, perpetuating neuron damage and cognitive deficits. Kynurenic acid (KYNA), a tryptophan metabolite with 7 nicotinic acetylcholine receptors (7nAChR) inhibitory property, blunts gp120-induced A1 astrocyte activation while promoting A2 astrocyte formation through blockade of 7nAChR/JAK2/STAT3 signaling activation. Meanwhile, we provide evidence that 7nAChR deletion and tryptophan administration convert astrocyte phenotypes, ultimately facilitating neuronal and cognitive improvement in the gp120tg mice. These initial and determinant findings mark a turning point in our understanding of the role of 7nAChR in gp120-mediated astrocyte activation, which opens new opportunities to control reactive astrocyte generation through KYNA and tryptophan administration as a pharmacological approach for treating HAND.

microbiology↗

bioRxiv2022

Single-Cell RNA-Seq Uncovers Dynamic Processes Orchestrated by RNA-Binding Protein DDX43 in Chromatin Remodeling during Spermiogenesis

Advances in single-cell RNA sequencing (scRNA-seq) have allowed for elucidating biological mechanisms at cell state level. Mammalian spermatogenic process showcases dynamic switches of gene expression pattern with delicate morphological and functional alterations of germ cells, but it is unclear how such dynamics is genetically controlled. Here we demonstrate that mouse testis-enriched RNA helicase DDX43, as well as its ATP hydrolysis site, is required for spermiogenesis. Genetic mutation of Ddx43 renders spermatids heterogeneously defective in multiple steps of chromatin remodeling, resulting in incomplete substitution of transition protein by protamine and less condensed sperm nucleus. Through scRNA-seq analyses of testicular cells derived from adult wild-type and Ddx43 mutant testes in mice, we reveal that the DDX43 deficiency-elicited perturbation in the dynamic RNA regulatory processes underlies the differentiation deficiency of spermatids. Further, focused analyses on early-stage spermatids combined with enhanced CLIP sequencing (eCLIP-seq) identify Elfn2 as DDX43-targeted hub gene, whose in vivo knockdown shows similar phenotypic defects as Ddx43 mutant. Our study illustrates an essential role for DDX43 in post-meiotic chromatin remodeling and highlights the single cell-based strategy for a refined dissection of stage-specific regulation of germline differentiation.

developmental biology↗

bioRxiv2022

YaHS: yet another Hi-C scaffolding tool

We present YaHS, a user-friendly command-line tool for construction of chromosome-scale scaffolds from Hi-C data. It can be run with a single-line command, requires minimal input from users (an assembly file and an alignment file) which is compatible with similar tools, and provides assembly results in multiple formats, thereby enabling rapid, robust and scalable construction of high-quality genome assemblies with high accuracy and contiguity. YaHS is implemented in C and licensed under the MIT License. The source code, documentation and tutorial are available at https://github.com/c-zhou/yahs.

bioinformatics↗

bioRxiv2022

SARS-CoV-2 S protein antagonizes type I interferon downstream signal pathway through interacting and attenuating phosphorylation of STAT1/STAT2

Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) may keep patients in a clinically asymptomatic state by blocking cellular innate antiviral immunity, but the molecular mechanism remains unclear. Here, we screened the viral proteins of SARS-CoV-2 and found that the spike (S) protein inhibits the activation of interferon-stimulated genes (ISGs) and even reduces the expression of these genes to below background values. Mechanistically, the S protein interacted with STAT1, STAT2, and IRF9 and impedes the phosphorylation of STAT1/STAT2, thus preventing the formation of the interferon-stimulating gene factor 3 (ISGF3) complex and inhibiting the downstream production of Interferon-stimulated genes (ISGs). Remarkably, we also have found that the inhibitory mechanism of the S protein was conservative among SARS-CoV-2 variants and other human coronaviruses, including SARS-CoV, MERS-CoV, HCoV-229E, HCoV-NL63, and HCoV-HKU1. Truncation studies indicated that the most conserved S2 domain played a major inhibitory role. Altogether, our findings unveil a new mechanism by which SARS-CoV-2 S protein attenuated the hosts antiviral immune response and provide new insights into the pathogenic mechanism of coronavirus.

microbiology↗

bioRxiv2022

A kinesin mediates VEGFR2 recycling and regulates VE-cadherin phosphorylation, essential for vascular permeability

Excessive vascular endothelial growth factor-A (VEGF-A) signaling induces vascular leakage and angiogenesis in diseases. VEGFR2 trafficking to the cell surface, mediated by kinesin-3 family protein KIF13B, is essential to respond to VEGF-A in inducing angiogenesis. However, the precise mechanism of how KIF13B regulates VEGF-induced signaling and endothelial permeability is unknown. Here we show that KIF13B-mediated recycling of internalized VEGFR2 through Rab11-positive recycling vesicle regulates VE-cadherin phosphorylation and endothelial permeability. Phosphorylated VEGFR2 at the cell-cell junction was internalized and associated with KIF13B in Rab5-positive early endosomes. KIF13B mediated VEGFR2 recycling through Rab11-positive recycling vesicle, and inhibition of this recycling attenuated phosphorylation of VEGFR2 at Y951, Src, and VE-cadherin at Y685, which are necessary for endothelial permeability. Failure of VEGFR2 trafficking to the cell surface induced accumulation and degradation of VEGFR2 in lysosomes. Furthermore, in the animal model of wet age-related macular degeneration (AMD), inhibition of KIF13B-mediated VEGFR2 trafficking also mitigated vascular leakage. Thus, the present results identify the fundamental role of VEGFR2 recycling to the cell surface in mediating vascular permeability, suggesting a promising strategy for mitigating vascular leakage associated with inflammatory diseases.

cell biology↗

bioRxiv2022

A Unified Dynamic and Computational Mechanism for Persistent and Transient Neural Activity Patterns During Delayed-Response Tasks

Understanding how the brain preserves information despite intrinsic noise is a fundamental question in working memory. Typical working memory tasks consist of delay phase for maintaining information, and decoding phase for retrieving information. While previous works have focused on the delay neural dynamics, it is poorly understood whether and how the neural process during decoding phase reduces memory error. We studied this question by training recurrent neural networks (RNNs) on a color delayed-response task. We found that the trained RNNs reduce the memory error of high-probability-occurring colors (common colors) by decoding/attributing a broader range of neural states to them during decoding phase. This decoding strategy can be further explained by a continuing converging neural dynamics following delay phase and a non-dynamic biased readout process. Our findings highlight the role of the decoding phase in working memory, suggesting that neural systems deploy multiple strategies across different phases to reduce memory errors. SignificancePreserving information under noise is crucial in working memory. A typical working memory task consists of a delay phase for maintaining information, and a decoding phase for decoding the maintained into an output action. While the delay neural dynamics have been intensively studied, the impact of the decoding phase on memory error reduction remains unexplored. We trained recurrent neural networks (RNNs) on a color delayed-response task and found that RNNs reduce memory error of a color by decoding a larger portion of the neural state to that color. This strategy is supported both by a converging neural dynamic, and a non-dynamic readout process. Our results suggest that neural networks can utilize diverse strategies, beyond delay neural dynamics, to reduce memory errors.

neuroscience↗