Brain developmental and cortical connectivity changes in transgenic monkeys carrying the human-specific duplicated gene SRGAP2C.

Human-specific duplicated genes contributed to phenotypic innovations during the origin of our own species, such as an enlarged brain and highly developed cognitive abilities. While prior studies on transgenic mice carrying the human-specific SRGAP2C gene have shown enhanced brain connectivity, the relevance to humans remains unclear due to the significant evolutionary gap between humans and rodents. In this study, to investigate the phenotypic outcome and underlying genetic mechanism of SRGAP2C, we generated transgenic cynomolgus macaques (Macaca fascicularis) carrying the human-specific SRGAP2C gene. Longitudinal MRI imaging revealed delayed brain development with region-specific volume changes, accompanied by altered myelination levels in the temporal and occipital regions. On a cellular level, the transgenic monkeys exhibited increased deep-layer neurons during fetal neurogenesis and delayed synaptic maturation in adolescence. Moreover, transcriptome analysis detected neotenic expression in molecular pathways related to neuron ensheathment, synaptic connections, extracellular matrix and energy metabolism. Cognitively, the transgenic monkeys demonstrated improved motor planning and execution skills. Together, our findings provide new insights into the mechanisms by which the newly evolved gene shapes the unique development and circuitry of the human brain.

ModeRNN: Harnessing Spatiotemporal Mode Collapse in Unsupervised Predictive Learning.

Learning predictive models for unlabeled spatiotemporal data is challenging in part because visual dynamics can be highly entangled, especially in real scenes. In this paper, we refer to the multi-modal output distribution of predictive learning as spatiotemporal modes. We find an experimental phenomenon named spatiotemporal mode collapse (STMC) on most existing video prediction models, that is, features collapse into invalid representation subspaces due to the ambiguous understanding of mixed physical processes. We propose to quantify STMC and explore its solution for the first time in the context of unsupervised predictive learning. To this end, we present ModeRNN, a decoupling-aggregation framework that has a strong inductive bias of discovering the compositional structures of spatiotemporal modes between recurrent states. We first leverage a set of dynamic slots with independent parameters to extract individual building components of spatiotemporal modes. We then perform a weighted fusion of slot features to adaptively aggregate them into a unified hidden representation for recurrent updates. Through a series of experiments, we show high correlation between STMC and the fuzzy prediction results of future video frames. Besides, ModeRNN is shown to better mitigate STMC and achieve the state of the art on five video prediction datasets.

Structural characteristics and diversity of the rhizosphere bacterial communities of wild Fritillaria przewalskii Maxim. in the northeastern Tibetan Plateau.

Introduction: Fritillaria przewalskii Maxim. is a Chinese endemic species with high medicinal value distributed in the northeastern part of the Tibetan Plateau. F. przewalskii root-associated rhizosphere bacterial communities shaped by soil properties may maintain the stability of soil structure and regulate F. przewalskii growth, but the rhizosphere bacterial community structure of wild F. przewalskii from natural populations is not clear. Methods: In the current study, soil samples from 12 sites within the natural range of wild F. przewalskii were collected to investigate the compositions of bacterial communities via high-throughput sequencing of 16S rRNA genes and multivariate statistical analysis combined with soil properties and plant phenotypic characteristics. Results: Bacterial communities varied between rhizosphere and bulk soil, and also between sites. Co-occurrence networks were more complex in rhizosphere soil (1,169 edges) than in bulk soil (676 edges). There were differences in bacterial communities between regions, including diversity and composition. Proteobacteria (26.47-37.61%), Bacteroidetes (10.53-25.22%), and Acidobacteria (10.45-23.54%) were the dominant bacteria, and all are associated with nutrient cycling. In multivariate statistical analysis, both soil properties and plant phenotypic characteristics were significantly associated with the bacterial community (p < 0.05). Soil physicochemical properties accounted for most community differences, and pH was a key factor (p < 0.01). Interestingly, when the rhizosphere soil environment remained alkaline, the C and N contents were lowest, as was the biomass of the medicinal part bulb. This might relate to the specific distribution of genera, such as Pseudonocardia, Ohtaekwangia, Flavobacterium (relative abundance >0.01), which all have significantly correlated with the biomass of F. przewalskii (p < 0.05). Discussion: F. przewalskii is evidently averse to alkaline soil with high potassium contents, but this requires future verification. The results of the present study may provide theoretical guidance and new insights for the cultivation and domestication of F. przewalskii.

PredRNN: A Recurrent Neural Network for Spatiotemporal Predictive Learning.

The predictive learning of spatiotemporal sequences aims to generate future images by learning from the historical context, where the visual dynamics are believed to have modular structures that can be learned with compositional subsystems. This paper models these structures by presenting PredRNN, a new recurrent network, in which a pair of memory cells are explicitly decoupled, operate in nearly independent transition manners, and finally form unified representations of the complex environment. Concretely, besides the original memory cell of LSTM, this network is featured by a zigzag memory flow that propagates in both bottom-up and top-down directions across all layers, enabling the learned visual dynamics at different levels of RNNs to communicate. It also leverages a memory decoupling loss to keep the memory cells from learning redundant features. We further propose a new curriculum learning strategy to force PredRNN to learn long-term dynamics from context frames, which can be generalized to most sequence-to-sequence models. We provide detailed ablation studies to verify the effectiveness of each component. Our approach is shown to obtain highly competitive results on five datasets for both action-free and action-conditioned predictive learning scenarios.

3D Genome of macaque fetal brain reveals evolutionary innovations during primate corticogenesis.

Elucidating the regulatory mechanisms of human brain evolution is essential to understanding human cognition and mental disorders. We generated multi-omics profiles and constructed a high-resolution map of 3D genome architecture of rhesus macaque during corticogenesis. By comparing the 3D genomes of human, macaque, and mouse brains, we identified many human-specific chromatin structure changes, including 499 topologically associating domains (TADs) and 1,266 chromatin loops. The human-specific loops are significantly enriched in enhancer-enhancer interactions, and the regulated genes show human-specific expression changes in the subplate, a transient zone of the developing brain critical for neural circuit formation and plasticity. Notably, many human-specific sequence changes are located in the human-specific TAD boundaries and loop anchors, which may generate new transcription factor binding sites and chromatin structures in human. Collectively, the presented data highlight the value of comparative 3D genome analyses in dissecting the regulatory mechanisms of brain development and evolution.