RNA trafficking and subcellular localization-a review of mechanisms, experimental and predictive methodologies

Abstract

RNA localization is essential for regulating spatial translation, where RNAs are trafficked to their target locations via various biological mechanisms. In this review, we discuss RNA localization in the context of molecular mechanisms, experimental techniques and machine learning-based prediction tools. Three main types of molecular mechanisms that control the localization of RNA to distinct cellular compartments are reviewed, including directed transport, protection from mRNA degradation, as well as diffusion and local entrapment. Advances in experimental methods, both image and sequence based, provide substantial data resources, which allow for the design of powerful machine learning models to predict RNA localizations. We review the publicly available predictive tools to serve as a guide for users and inspire developers to build more effective prediction models. Finally, we provide an overview of multimodal learning, which may provide a new avenue for the prediction of RNA localization.

Keywords: RNA; localization; machine learning; multimodality; subcellular.

Figure 1

Spatial distribution of RNAs among various cell compartments. The cellular plot is divided into two sets for different perspectives. The top set represents the x–y axis, while the lower set represents the x–z axis. Squiggles and dots denote different RNAs, and distinct colors indicate different compartments.

Figure 2

The introduction of the molecular mechanism of localization through three mainly investigated manners. (A) In fibroblasts, β-actin mRNAs are localized to the leading edge of cell with the RBP KH3 domain binding to the 3′ end of the sequence and KH4 domain binding to the 5′ end of sequence. Interestingly, these two domains can identify specific sequence motif. For example, KH4 recognizes CGGAC and KH3 for C/A-CA-C/U, where the first base can be substitute as C or A, and the fourth base can be C or U. (B) The transport of β-actin mRNAs to the leading edge of fibroblasts can be achieved by binding the mRNPs with multiple motors. For example, β-actin mRNAs have two binding domains, which mediate the binding of ZBP1, and in turn, the binding of myosin. Afterward, this mRNP-motor complex can be transported via actin filaments. (C) In Drosophila melanogaster embryo, CCR4–NOT complex was recruited to decay the mRNA when RBP Smung bind to the Nos mRNAs. However, if the Nos mRNAs are localized to the posterior pole of the embryo, Smung was detached from Nos mRNA and was replaced by Oskar proteins, results in the protection from degradation and improves the concentration of Nos mRNA at the posterior site. (D) In Escherichia coli, nascent mRNAs can localize to the ribosome-rich poles or to the membrane from the nucleoid by random diffusion with the speed of 0.05 μm²/s. (E) During the oogenesis of D. melanogaster, hundreds of mRNA, including Nos mRNA, were squeezed from the nurse cells into the oocyte, where they are entrapped in the germ plasm followed by the cytoplasmic streaming. mRNP, mRNA-protein.

Figure 3

The basic principles of MERFISH (multiplexed error-robust RNA fluorescence in situ hybridization) and seqFISH+ are outlined. (A) MERFISH uses a two-step hybridization process. Initially, multiple nonfluorescent encoding probes are hybridized to RNA molecules, providing several readout hybridization sequences for rapid binding of fluorescently labeled readout probes. The fluorescent dyes attached to the readout probes can be chemically deactivated, enabling multiple cycles of hybridization with different readout probes, fluorescence imaging and signal quenching. This results in a sophisticated error-robust binary barcode readout for each target transcript. (B) In seqFISH+, mRNAs are hybridized with 24 primary probes, which can be imaged after 20 cycles of readout hybridization. The in situ mRNAs are then imaged using fluorescence, with each localized mRNA assigned a pseudocolor. After four rounds of imaging hybridization, genes can be encoded using different imaging channels, typically 640 nm, 561 nm and 488 nm.

Figure 4

The design of a bimodal fusion incorporates data from multiple sources. There are three primary strategies for fusing data from different modalities. Early fusion: This strategy involves encoding each modality and projecting them into a high-dimensional representation, which is then used as a single input to construct the prediction model. Intermediate fusion: Instead of fusing data through voting or average-based methods, encoded unimodal representations are integrated through a separate model to form multimodal representations. These representations are then used as inputs to binary or multi-class classifiers. Late fusion: In this approach, each modality is encoded into its own unimodal representation, which is then processed using simple calculations such as average, weighted vote or majority vote.