• <tr id="yyy80"></tr>
  • <sup id="yyy80"></sup>
  • <tfoot id="yyy80"><noscript id="yyy80"></noscript></tfoot>
  • 99热精品在线国产_美女午夜性视频免费_国产精品国产高清国产av_av欧美777_自拍偷自拍亚洲精品老妇_亚洲熟女精品中文字幕_www日本黄色视频网_国产精品野战在线观看 ?

    Computational Screening of Phase-separating Proteins

    2021-12-03 09:03:24BoyanShenZhaomingChenChunyuYuTaoyuChenMingleiShiTingtingLi
    Genomics,Proteomics & Bioinformatics 2021年1期

    Boyan Shen, Zhaoming Chen, Chunyu Yu, Taoyu Chen, Minglei Shi Tingting Li,*

    1 Department of Biomedical Informatics, School of Basic Medical Sciences, Peking University Health Science Center, Beijing 100191, China

    2 Institute of Systems Biomedicine, School of Basic Medical Sciences, Peking University Health Science Center, Beijing 100191, China

    3 MOE Key Laboratory of Bioinformatics, Bioinformatics Division and Center for Synthetic & Systems Biology, BNRist, School of Medicine, Tsinghua University, Beijing 100084, China

    KEYWORDS

    Abstract Phase separation is an important mechanism that mediates the compartmentalization of proteins in cells. Proteins that can undergo phase separation in cells share certain typical sequence features, like intrinsically disordered regions (IDRs) and multiple modular domains. Sequencebased analysis tools are commonly used in the screening of these proteins. However,current phase separation predictors are mostly designed for IDR-containing proteins,thus inevitably overlook the phase-separating proteins with relatively low IDR content. Features other than amino acid sequence could provide crucial information for identifying possible phase-separating proteins: protein–protein interaction(PPI)networks show multivalent interactions that underlie phase separation process; post-translational modifications (PTMs) are crucial in the regulation of phase separation behavior; spherical structures revealed in immunofluorescence (IF) images indicate condensed droplets formed by phase-separating proteins, distinguishing these proteins from non-phaseseparating proteins. Here, we summarize the sequence-based tools for predicting phaseseparating proteins and highlight the importance of incorporating PPIs,PTMs,and IF images into phase separation prediction in future studies.

    Introduction

    Cellular organelles can be categorized into two classes,membrane-bound organelles and membraneless organelles.Membrane-bound organelles include classic organelles such as the Golgi apparatus, mitochondrion, and lysosome. These cellular compartments enclosed by lipid bilayers have been well studied in the last century. However, in living cells, many biochemical reactions take place in membraneless organelles[1,2].The formation mechanisms and functions of membraneless organelles had remained perplexing until ten years ago. In 2009, Brangwynne et al. reported liquid-like behaviors of P granules, which are protein-rich membraneless organelles in the cytoplasm of cells from Caenorhabditis elegans [3]. These granules can flow,deform,and undergo fission freely,just like liquid droplets. Proteins within P granules are also highly mobile and can exchange rapidly with the surrounding cytoplasm. These findings suggest that liquid–liquid phase separation(LLPS,also called liquid–liquid demixing)could be one of the mechanisms underlying membraneless organelle formation.

    In a phase separation process,a set of macromolecules such as proteins and nucleic acids are separated from their surrounding environments and form an independent phase. The separated phase shares a similar molecular composition with the surrounding environment, yet at different concentrations[4]. In cells, proteins or nucleic acids form separated phases via intra- or intermolecular interactions [1,4,5], thereby allowing the formation of phase-separated compartments,which are also named membraneless organelles or biomolecular condensates.Besides P granules[3],the nucleoli[1,2],centrosomes[6],stress granules [7,8],and processing bodies (P-bodies) [8,9] are also membraneless organelles formed through phase separation. In addition, phase separation underlies many biological processes such as translation regulation [10,11], mRNA deadenylation [10,12], heterochromatin formation [13,14], and the control of signal transduction [15–17].

    Phase separation is a complex biophysical process.Changes to any property of the system, e.g., molecular composition,temperature, electrostatic property, and viscoelasticity of the solution, may affect the phase separation process [1,5,18].Being able to undergo phase separation under specific conditions may be a universal property of proteins. However, only a few proteins with specific sequence-dependent features have the potential to undergo phase separation in living cells [19].In this review, we name these proteins phase-separating proteins. Scientists have found that certain sequence features may correlate with phase separation behaviors, which brings out a range of useful bioinformatics tools to predict phaseseparating proteins. Herein, we summarize the sequencebased predicting tools for phase-separating proteins and integrate the available phase-separating protein data to evaluate their performances. Furthermore, we propose protein–protein interaction (PPI) networks, post-translational modifications(PTMs),and immunofluorescence(IF)images as three promising features to be incorporated into phase separation prediction in future studies.

    Driving force of phase separation

    Phase separation is a conditional process. Proteins indispensable to the formation of one condensate can be alternative in another or do not participate in phase separation under some conditions [19]. Multivalent interactions between condensate components are the driving force of phase separation[17,20,21].

    Proteins able to form multivalent interactions that promote phase separation can be classified into two types: one characterized by multiple modular domains and the other characterized by intrinsically disordered regions (IDRs). The first type of proteins often carry several folded interaction domains.The driving force of phase separation of these proteins is the multivalent interactions between their interaction domains[4]. An example is the interaction between small ubiquitinlike modifier (SUMO) and SUMO-interacting motif (SIM),which plays an essential role in the overall architecture of promyelocytic leukemia (PML) body. The PML protein contains a SUMO-interacting motif and multiple SUMOylation sites [22]. As the scaffold protein of the PML body, PML can not only self-assemble via interactions between its tripartite motifs(TRIMs)but also interact with itself and other proteins via SUMO–SIM interactions [23,24] (Figure 1A).Another example is the multivalent interactions of the nephrin–non-catalytic region of tyrosine kinase (NCK)–neuronal Wiskott–Aldrich syndrome protein (N-WASP) system in the actin regulatory signaling pathway. Nephrin is a transmembrane protein,the cytoplasmic tail of which harbors three tyrosine phosphorylation(pTyr)sites.Each of these three pTyr sites can bind to a Src homology 2(SH2)domain on NCK.NWASP contains six proline-rich motifs(PRMs),which can also bind to three of the SH3 domains on NCK.These interactions further stimulate actin assembly and phase separation [17](Figure 1B).

    The other type of phase-separating proteins are characterized by the presence of IDRs[25,26].Instead of having a fixed three-dimensional structure, IDRs can interconvert between a range of slightly different low-energy states with discrepant conformations [27–29]. Proteins with high proportions of IDRs are hot targets in phase separation studies,e.g.,the neurodegenerative disorder-related protein fused in sarcoma(FUS)[30–33](Figure 1C),the BUB3-interacting and GLEBS motif-containing protein BuGZ [34,35] (Figure 1D), and the microtubule-associated protein Tau [36–38]. However, not all IDRs facilitate phase separation.Phase-separating IDRs often share some sequence features. One type of IDRs that often undergo phase separation is the IDRs encompassing lowcomplexity regions (LCRs), which are regions with biased amino acid composition, often containing repeated segments.Moreover, phase-separating IDRs are often enriched with specific amino acid residues, e.g., arginine and glycine for the RGG/RG domains [39,40] or polar amino acid residues like serine, tyrosine, glutamine, and asparagine for the prion-like domains [41]. These amino acids do not appear randomly,but are often found as functional sites like short/eukaryotic linear motifs and alternating charge blocks [25,42].

    IDR content analysis in the prediction of phaseseparating proteins

    Figure 1 Schematic view of multivalent interactions that promote phase separationA. PML protein can not only self-assemble via interactions between its TRIMs but also interact via its SIM with SUMOs of itself and other proteins,such as DAXX and SP100.B.Schematic view of nephrin–NCK–N-WASP system.The cytoplasmic tail of nephrin contains three pTyr sites,each of which can bind to an SH2 domain on NCK.The three SH3 domains on NCK can also bind to PRMs within NWASP. C.Residue-wise plot of scaffold protein FUS. Information of PTM sites was collected from the PhosphoSitePlus database;IDR scores were predicted by IUPred for long disorder; the prion-like domain was identified with PLAAC score greater than zero; the picontact was identified with PScore greater than four, as indicated by dashed line. D. Residue-wise plot of scaffold protein BuGZ. PML,promyelocytic leukemia; TRIM, tripartite motif; SIM, SUMO-interacting motif; SUMO, small ubiquitin-like modifier; DAXX, death domain-associated protein; SP100, nuclear autoantigen Sp-100; NCK, non-catalytic region of tyrosine kinase; N-WASP, neuronal Wiskott–Aldrich syndrome protein;pTyr,tyrosine phosphorylation;SH,Src homology;PRM,proline-rich motif;FUS,fused in sarcoma;PTM, post-translational modification; IDR, intrinsically disordered region; BuGZ, BUB3-interacting and GLEBS motif-containing protein ZNF207; Phos, phosphorylation; Ub, ubiquitination; Ac, acetylation; Me, methylation; AA, amino acid.

    Table 1 IDR content analysis tools

    IDR-containing proteins account for a large proportion of phase-separating proteins, although our understanding of how IDRs are involved in the phase separation process and what features these IDRs share is still limited [43]. Therefore,IDR content analysis is often utilized in bioinformatics screening of potential phase-separating proteins(Table 1).DisProt is one of the earliest and most influential disorder databases[44,45]. Two long-tested IDR predictors, ESpritz [46] and IUPred[47,48],are frequently used in the IDR content analysis of potential phase-separating proteins. In recent years, meta-IDR predictors like D2P2 [49] and MobiDB [50,51], which integrate predictions from several predictors currently available and provide comprehensive results, are widely used as well.For example,in a previous study on how intrinsically disordered linkers influence the interplay between phase separation and gelation, researchers used the meta-IDR predictor D2P2 to identify proteins containing disordered regions from the human proteome [52].

    Phase separation predictors

    Although multivalent, transient interactions that drive phase separation are often facilitated by IDRs [21], only proteins with certain types of IDRs facilitate phase separation. Non-IDR interacting elements like the coiled-coil domain can also promote these interactions [6]. Bioinformaticians thus aim to develop prediction tools that are based explicitly on phase separation-specific sequence features.

    By studying the sequences of several well-known phaseseparating proteins, e.g., FUS, DEAD box protein 4(DDX4), TATA-binding protein-associated factor 15(TAF15), Ewing sarcoma protein (EWS), TAR DNAbinding protein 43 (TDP43), and heterogeneous nuclear ribonucleoprotein A1 (HNRNPA1), two common types of LCRs, prion-like domains (PLDs) and RGG domains, were found to promote weak interactions and thereby promote phase separation [33,39,53,54]. Furthermore, low-complexity aromatic-rich kinked segments (LARKSs) and steric zipper motifs have been found to promote the transition between different physical properties of biomolecular condensates[55,56].Such features have inspired bioinformaticians to design algorithms to predict the phase separation propensities of proteins[57]. These include PScore [58], prion-like amino acid composition (PLAAC) [59], PSPer [60], catGRANULE [61], R + Y[41], LARKS [55], and ZipperDB [56] (Table 2).

    PScore

    Pi–pi interactions can occur between protein sequences enriched in pi-orbital-containing residues. In contrast to the conventional view, pi–pi interaction involves not only amino acids with an aromatic ring (tyrosine, phenylalanine, tryptophan, and histidine) [62], but also non-aromatic amino acids with pi bonds on their side chains(glutamine,asparagine,glutamic acid, aspartic acid, and arginine) and small amino acids with exposed backbone peptide bonds (glycine, serine, threonine, and proline). Face-to-face interactions formed by these pi-containing groups are called planar pi–pi contacts.In 2018,the Forman-Kay group reported that planar pi–pi contact represents a predominant interaction type and is highly relevant to self-association and phase separation of proteins [58]. Then a planar pi–pi contact predictor named PScore was developed for screening potential phase-separating proteins.

    PLAAC

    PLAAC is an application initially designed to screen PLDs[59], which utilizes a hidden Markov model (HMM) for thediscrimination of PLDs and non-PLDs based on amino acid composition. PLAAC was originally trained on yeast proteome but later extended to screen human proteins [63,64]. It supports both single-protein and proteome scanning.

    Table 2 Phase separation analysis tools

    PSPer

    PSPer is a rule-based model developed for screening prion-like RNA-binding phase-separating proteins. Expected properties of the FUS-like phase-separating regions are used to build an HMM-like model [41], which identifies PLDs, RNArecognition motifs, and disordered, arginine-rich regions within a protein [60].

    catGRANULE

    catGRANULE is a phase separation prediction algorithm with good performance for predicting dosage-sensitive proteins[61].This tool is based on the discovery that cellular toxicity mechanisms of some dosage-sensitive proteins in yeast are well-explained by LLPS theory, as these proteins take part in the formation of cytoplasmic foci in a concentration-related manner. Further studies reveal that these proteins have an increased nucleic acid binding propensity. catGRANULE was therefore developed to screen these proteins by combining nucleic acid binding propensities,structural disorder,sequence length, and content of arginine, glycine, and phenylalanine.Although initially trained against the yeast proteome, cat-GRANULE has been successfully applied to mammalian and even human proteomes [37].

    R + Y

    R + Y is a predictor built upon the analysis of the molecular grammar of FET family proteins, including FUS, EWS, and TAF15 [65]. In 2018, the Hyman, Alberti, and Pappu groups reported that phase separation behaviors of FET family proteins are determined by interactions between the tyrosine-rich PLDs and the arginine-rich RNA-binding domains[41].Further analyses indicate that the numbers of tyrosine and arginine residues are inversely correlated with the measured saturation concentrations for phase separation. Extrapolating these findings to the prediction of non-FET proteins, researchers developed the R + Y predictor and utilized it for a human proteome-wide analysis, which has remarkable prediction performance on DNA-and RNA-binding proteins(RBPs).

    ZipperDB

    ZipperDB is a database that includes predicted fibril-forming segments from more than 20,000 putative amyloid-forming protein sequences [56]. Fibrils are highly ordered aggregates characterized by a ‘‘steric zipper” structure, whose formation is an essential step in amyloidosis.Amyloid deposition is occasionally observed in cells. Recent evidence indicates that the formation of such deposits could be attributed to a liquid-tosolid phase transition or an atypical phase separation process that forms solid-state compartments. ZipperDB utilizes an algorithm named 3D profiling to analyze the probability of forming a steric zipper structure for every hexapeptide segment.

    LARKS

    LARKSs share similar structural characteristics with steric zippers yet have lower binding energy, as aromatic residues predominate the kinks and affect LARKS stability through weak interactions. LARKSs are characterized by the stacking kinked β sheet pairs,which promote the formation of amyloid fibrils and hydrogels during phase transition. Similar to ZipperDB,3D profiling was utilized to identify potential LARKSs from inquiry protein sequences. After querying the human proteome of 20,120 proteins, a list of 400 proteins with the most enriched LARKS was provided [55].

    In summary, all current phase separation prediction tools are developed based on sequence-dependent features. PScore calculates the pi-contact propensities of each residue in a given protein sequence.PLAAC and PSPer identify specific domains like PLDs based on HMM. catGRANULE calculates the granule propensity of each residue in a given sequence.R+Y calculates the number of tyrosine and arginine residues within disordered regions of a given sequence. LARKS and ZipperDB adopt an algorithm named 3D profiling to measure the probability of a given sequence to fold into a LARKS or a steric zipper.

    Performance evaluation of phase separation predictors

    Vernon et al. have recently reviewed several phase separation predictors and compared their prediction performance comprehensively [57]. They find that since each algorithm predicts different kinds of interactions and sequence features, very different protein categories are covered by these predictors. The only exceptions are RBPs, as high prediction confidence is obtained from all predictors for RBPs. However, due to the insufficient phase-separating protein data available, evaluations could only be made on a set of 30 human proteins [57].

    At least four LLPS protein databases were released until 2020, including LLPSDB [66], PhaSePro [67], PhaSepDB[68],and DrLLPS[69](Table 3).LLPSDB collects in vitro data on LLPS-related proteins. The current version of LLPSDB includes 295 independent proteins, which are integrated into 1192 entries with corresponding experimental phase separation conditions. PhaSePro provides the experimental data on 121 proteins driving phase separation in living cells, which is less focused on in vitro phase separation conditions and contains a broader array of information than LLPSDB. PhaSepDB contains less detailed information than either LLPSDB or Pha-SePro, but provides a larger set of data with 2914 nonredundant proteins localized in more than 30 different organelles, which comprise PubMed-reviewed data, UniProtreviewed data, and high-throughput data. DrLLPS contains 9285 curated proteins that are known to be associated with LLPS, including 150 scaffold proteins, 987 regulator proteins,and 8148 potential client proteins. All four databases extensively reference the original literature,allowing the user to verify information or conduct further scrutinization.

    With the availability of high-quality data provided by the four databases mentioned above, a more comprehensive comparison of phase separation predictors is possible. We constructed a non-redundant positive set (set P) of 278 human proteins involving 90,59,233,and 86 proteins from LLPSDB,PhaSePro, the PubMed-reviewed part of PhaSepDB, and the scaffold part of DrLLPS,respectively(Table S1).The remaining majority of human proteome was collected from UniProt and defined as the negative set (set N), which includes 20,227 proteins (Table S1).

    To compare the prediction performance of available phase separation predictors,we scored the proteins in both set P and set N using PScore, PLAAC, catGRANULE, and PSPer,which provide online/offline batch prediction (Table S2). For R + Y and LARKS without prediction tools, R + Y scores of 2657 human proteins and 400 human proteins enriched with LARKSs were downloaded from the resepctive websites(Table S2). ZipperDB provides neither batch prediction tools nor bulk download.As a result,we collected prediction scores from six phase separation predictors (Table S2).

    Firstly, we adopted a comparison between those six tools on human proteome by plotting the receiver operating characteristic curve (ROC). As shown in Figure 2A, catGRANULE,PScore, PLAAC, and PSPer have nearly the same values for area under the curve (AUC), which are > 0.7, while the performance of R + Y and LARKS is not as good as that of the other four tools.

    To test the influence of IDR contents on the scores of these phase separation predictors,we scored the proteins in set P by IUPred with default setting(Table S3).As shown in the scatter plots,phase separation scores of the first five predictors are significantly correlated with IDR scores(Figure 2B).In R+Y,a lower score means a higher phase separation potential, and therefore its prediction scores are significantly negativelycorrelated with IDR scores. For LARKS that did not provide prediction scores, proteins in set P were divided into LARKS High and LARKS Low groups,according to whether the protein is in the list of LARKS enriched proteins.The IDR scores of proteins in the LARKS High group were significantly higher than those in the LARKS Low group (Figure 2C).The aforementioned comparison processes were also performed for all human proteins, scores of all phase separation predictors are significantly correlated with IDR scores (Figure S1). These results demonstrate that all phase separation predictors prefer proteins with high IDR contents.

    Discriminating the scaffold proteins from the client proteins remains challenging. Therefore, it is possible that some low IDR proteins in set P are client proteins.However,taking the 12 phase-separating proteins for example,heterochromatin protein 1 homolog alpha (HP1A), nucleophosmin 1 (NPM1),PML,and NCK1,which are all scaffold proteins,were rankedoutside of the top 10% of the human proteome by all the six predictors, while PML was ranked outside of the top 20%(Figure 2D).

    Table 3 Information of four LLPS databases

    Figure 2 Comparison of phase separation predictors on human proteomeA. ROC curve for each predictor. Since PScore, PLAAC, and PSPer have restrictions on the length of protein sequence, only catGRANULE returned scores for all human proteins.Except for catGRANULE,the AUC scores of the remaining tools were calculated on the subsets of P and N sets.Prediction of LARKS was shown as a point since it did not provide scores for each protein.B.Scatter plots of predicted values for proteins in set P,with one axis being the IDR score and the other axis being the phase separation score.Spearman correlation coefficient with P < 0.05 indicates significant correlation. C. Proteins in set P were divided into LARKS High and LARKS Low groups, according to whether the protein is in the list of LARKS-enriched proteins. P value is calculated through the two-sided Mann–Whitney U test. D. Ranking scores of 12 specific LLPS proteins by six phase separation predictors. ROC, receiver operating characteristic;AUC,area under the curve;LARKS,low-complexity aromatic-rich kinked segment;LLPS,liquid–liquid phase separation.

    In summary, a range of useful bioinformatics tools have been developed to predict phase-separating proteins,however,most of which are designed for screening IDR-containing proteins. Proteins with modular domains account for a considerable part of phase-separating proteins as well, yet a corresponding computational tool to identify such proteins is not available.It may not be so challenging to identify proteins with multiple modular interaction domains. However, some proteins with few modular domains can also participate in phase separation via assembly, like HP1A [13]. More comprehensive features besides sequence composition might be required for phase separation predictors.

    Other features potentially used to predict phaseseparating proteins

    Features other than sequence composition could provide crucial information for identifying possible phase-separating proteins.This section will discuss three new features that could be utilized in phase separation prediction: PPI networks, PTMs,and IF images.

    PPI networks

    As described in the preceding section, the driving force of phase separation is the multivalent interactions between molecules. In some cases, especially for those with low IDR contents, proteins cannot undergo phase separation alone. Take the nephrin–NCK–N-WASP system in Figure 1B for example.The pTyr–SH2 and SH3–PRM interactions among cooperated proteins are the driving force of the phase separation process,and NCK cannot phase separate under the conditions with nephrin or N-WASP abscent[17].Another example is the clustering of T cell receptor (TCR) signaling pathway molecules.Upon TCR activation, the tyrosine kinase ZAP70 phosphorylates the transmembrane protein LAT, and phospho-tyrosines on LAT can bind to the SH2 domain on Grb2. Two SH3 domains within Grb2 further interact with Sos1,and the phosphorylated LAT,Grb2,and Sos1 together form the LAT complex, which can coalesce into T cell microclusters that show phase separation behaviors[15].Similar to the aforementioned example, components in this TCR pathway cannot phase separate without interactions among cooperated proteins.

    One issue of the available phase separation predictors is that they are based on sequence-dependent features of individual proteins.These features may not be sufficient,as phase separation driven by complex modular interaction domains or motifs and other multivalent interactions might be missed. A more comprehensive approach that also considers PPI network information could be helpful. It is hard to integrate PPI networks into phase separation predictors directly. One possible approach is network embedding method such as node2vec[70].Taking the adjacency matrix of the PPI network as input,node2vec encodes each node in the network as a vector,which can be used for various downstream machine learning tasks.The distance between vectors reflects the similarity of interaction networks of corresponding proteins. However, it should be noted that PPI networks based on experimental evidence usually bias to well-studied proteins. BioPlex database that provides an unbiased mapping of the human PPIs by affinity-purification mass spectrometry might be more appropriate to be incorporated into phase separation prediction[71,72].

    PTMs

    The interactions required for phase separation can be weak interactions or strong interactions reversible on a short timescale [21]. The reversibility can often be regulated by PTMs,like the interaction between the SH2 domain and the pTyr residue in nephrin–NCK–N-WASP and TCR pathways[15,17].In these examples, PTMs can regulate the reversibility of a binding event by generating/degenerating a modular interaction domain recognition site [21].

    PTMs can also regulate phase separation processes by changing protein physical properties, like the charge state,bulkiness, solubility, hydrophobicity, or binding affinity[10,73–75]. Citrullination of RG/RGG motifs has been reported to increase the solubility of proteins like FUS,EWS, and TAF15, inhibiting the arginine methylation, aggregation, and stress granule formation of these proteins [76].Another example is the PTMs of fragile X mental retardation protein (FMRP) and cytoplasmic activation- and proliferation-associated protein 1 (CAPRIN1). The two proteins do not co-phase separate without phosphorylation.However, co-phase separation occurs when Tyr of CAPRIN1 aromatic-rich regions is phosphorylated, or the C-terminal LCR of FMRP is phosphorylated [10,74].

    Figure 3 Comparison of PTM frequencies between different groups on human proteomeA.Venn diagram displaying the overlap of phase-separating proteins with different PTM types.The PTM datasets were collected from the PhosphoSitePlus database.B.For all PTM types,frequencies for proteins in set P are significantly higher than those for set N.P values are calculated through the two-sided Mann–Whitney U test.C.Distribution of IDR contents of proteins in set P and set N.Most proteins in set N have low proportions of IDRs. D. Set N was resampled into a subset according to IDR content distribution of set P. P value indicates that IDR contents of proteins in sampled set N have a similar distribution with that in set P.E.PTM frequencies of proteins in set P are still higher than those in resampled set N.

    Given the widespread regulatory roles of PTMs in phase separation,we tested whether PTM levels could be used as features to discriminate phase-separating proteins. We collected PTM datasets from the PhosphoSitePlus database[77],including phosphorylation,ubiquitination,acetylation,and methylation (Table S4). As shown in Figure 3A, the Venn diagram displays the overlap of phase-separating proteins with different PTM types.For 278 phase-separating proteins in set P,221 of them contain more than three types of PTMs,and 158 of them possess all four types of PTMs considered. The overlap demonstrates that most phase-separating proteins are modified by multiple types of PTMs. Then we defined the PTM frequency of a protein as the number of modification sites on a sequence divided by the length of the sequence,and calculated PTM frequencies of proteins in set P and set N.For each protein,frequency of all PTM types and frequencies of each specific PTM types with more than 10,000 recorded sites in the database were calculated (Table S5). For all PTM types, frequencies for proteins in set P were significantly higher than those in set N(Figure 3B).However,it is well established that PTM sites are enriched in IDRs, and proteins with high proportions of IDRs tend to have higher PTM frequencies(Figure 3C), which might cause bias. To control the impact of different IDR contents, the set N was resampled into a subset,whose distribution of IDR contents was similar to that of the set P(Figure 3D;see details of the resampling process in Table S6). The PTM frequencies of the set P were still significantly higher than those of the resampled set N (Figure 3E),which indicates that PTM frequency can be regarded as a feature to discriminate phase-separating proteins. Previous studies reported that an increased number of PTMs was correlated with an increased LLPS propensity in predicted phase-separating proteins [10,58], and our results show the similar observations using the actual LLPS proteins from more datasets.

    It should be noted that although a considerable number of PTM sites have been collected into the database,most of them are phosphorylation sites.Poor data quality and low coverage of the PTM dataset might affect the performance of incorporating PTM frequency as a feature.A high-confidence dataset,including 119,809 phosphorylation sites, has been reported recently [78], which could provide a high-quality PTM feature for identifying possible phase-separating proteins.

    IF images

    Identification of spherical droplet structures through IF images represents the most common approach for validation of phase-separating proteins. Phase-separating proteins usually appear as spherical droplet-like structures in IF images,which allows them to be distinguished from non-phaseseparating proteins. Thus phase-separating proteins can be identified if we screen out the proteins which appear as spherical structures in IF images.The Cell Atlas(part of the Human Protein Atlas) database provides antibody-based profiling by IF confocal microscopy for 12,073 proteins [79], allowing the screening of phase-separating proteins based on IF images.

    It should be noted that the formation of phase separation is condition-dependent. Phase separation proteins that do not appear as droplets in the inquiry IF images cannot be identified. In addition, some phase-separated condensates are too small to be detected in IF images, such as the ‘‘transcription hubs” in transcriptional activation [80], and some membrane-bound organelles like vesicles have similar spherical appearance in IF images.Therefore,IF images alone are insufficient to determine whether a specific protein can undergo phase separation or not. However, IF images with dropletlike structures could provide evidence that the labeled proteins aggregate in cells, which allows us to screen out the potential phase-separating proteins.In a previous study,we built a convolutional neural network (CNN) classifier to identify IF images with droplet-like structures and found that the aggregation evidence extracted from IF images is useful in screening phase-separating proteins with low IDR contents [81]. Besides deep learning methods, CellProfiler[82] that can segment droplets and cells in the IF images should also be useful in extracting aggregation evidence from IF images.The outputs of both the CNN classifier and the CellProfiler can measure the aggregation state of the labeled proteins in IF images,and these outputs can be used as features for various downstream machine learning tasks.

    Conclusion and perspectives

    Experimental studies have enabled significant progress in improving our understanding of phase separation.Researchers have also noticed that some sequence features are closely related to the phase separation behavior, and several sequence-based computational tools have been developed accordingly. These computational methods facilitate studies on the phase separation phenomenon by providing predictions and proteome-scale screening of phase separation candidates.Furthermore, to examine the roles of different domains of specific proteins in phase separation, truncation mutants are usually constructed to detect segments crucial in forming liquid-like droplets. Computational methods that provide residue-specific predictions also assist phase separation studies in screening critical fragments.However,a meta-predictor that integrates predictions from existing phase separation predictors is in need to reduce the complexity of sequence analysis,like D2P2 and MobiDB in IDR analysis.

    Furthermore, current phase separation predictors are mostly designed for IDR-containing proteins and are based on sequence-dependent features of individual proteins. More comprehensive features should be incorporated into phase separation predictors. As multivalent interactions are critical to phase separation,PPI networks can be useful in phase separation computational analysis. PTMs and IF images are also available features that can be utilized in the prediction of phase-separating proteins. PTMs are predominant regulators of phase separation behavior. Accordingly, in this review, we also find that phase-separating proteins tend to have high PTM frequencies. Components of phase-separating condensates usually appear as spherical droplets in IF images, allowing the screening of phase-separating proteins. We expect that incorporating PPI networks,PTMs,and IF images into prediction algorithms will lead to more effective and unbiased phase separation analytic tools.

    CRediT author statement

    Boyan Shen: Methodology, Investigation, Writing - original draft, Writing - review & editing. Zhaoming Chen: Methodology,Formal analysis,Writing-original draft,Writing-review& editing. Chunyu Yu: Methodology. Taoyu Chen: Writing -review & editing. Minglei Shi: Writing - review & editing.Tingting Li: Conceptualization, Methodology, Supervision,Writing - review & editing. All authors read and approved the final manuscript.

    Competing interests

    The authors have declared no competing interests.

    Acknowledgments

    This work was supported by the National Key R&D Program of China (Grant No. 2018YFA0507504) and the National Natural Science Foundation of China (Grant Nos. 61773025 and 32070666).

    Supplementary material

    Supplementary data to this article can be found online at https://doi.org/10.1016/j.gpb.2020.11.003.

    ORCID

    0000-0002-1574-1958 (Boyan Shen)

    0000-0002-6810-5445 (Zhaoming Chen)

    0000-0002-3519-1332 (Chunyu Yu)

    0000-0002-8966-1264 (Taoyu Chen)

    0000-0002-1108-4233 (Minglei Shi)

    0000-0003-4266-0317 (Tingting Li)

    免费观看性生交大片5| 免费观看的影片在线观看| 男人和女人高潮做爰伦理| 中文字幕制服av| 亚洲丝袜综合中文字幕| 国产女主播在线喷水免费视频网站| 国产成人免费观看mmmm| 欧美成人一区二区免费高清观看| 午夜激情福利司机影院| 日本欧美国产在线视频| 激情五月婷婷亚洲| 精品国产三级普通话版| 国产精品成人在线| 99热网站在线观看| 美女福利国产在线 | 欧美最新免费一区二区三区| 菩萨蛮人人尽说江南好唐韦庄| 日本爱情动作片www.在线观看| 亚洲精品日本国产第一区| videos熟女内射| 国产精品一区二区在线观看99| 国产精品偷伦视频观看了| 一级a做视频免费观看| 春色校园在线视频观看| 观看免费一级毛片| 国产男女超爽视频在线观看| 国产精品一及| 国模一区二区三区四区视频| av黄色大香蕉| 成年美女黄网站色视频大全免费 | 日产精品乱码卡一卡2卡三| 各种免费的搞黄视频| 久久午夜福利片| 纯流量卡能插随身wifi吗| 中国美白少妇内射xxxbb| 日韩av不卡免费在线播放| 国产一区亚洲一区在线观看| 毛片女人毛片| 99热这里只有是精品50| 亚洲av欧美aⅴ国产| 大码成人一级视频| 亚洲精品日本国产第一区| 老女人水多毛片| 网址你懂的国产日韩在线| 成人黄色视频免费在线看| 丰满迷人的少妇在线观看| 国产精品久久久久久精品电影小说 | 高清毛片免费看| 日本黄色日本黄色录像| 日产精品乱码卡一卡2卡三| 男女国产视频网站| a级毛色黄片| 蜜桃在线观看..| 男女啪啪激烈高潮av片| 国产男女内射视频| 青春草国产在线视频| a级一级毛片免费在线观看| 成年美女黄网站色视频大全免费 | 26uuu在线亚洲综合色| 青青草视频在线视频观看| 久久久久国产精品人妻一区二区| 中国美白少妇内射xxxbb| 亚洲精品久久久久久婷婷小说| 97在线视频观看| 国产淫片久久久久久久久| 国产日韩欧美在线精品| 永久网站在线| 亚洲国产欧美人成| 成年免费大片在线观看| 不卡视频在线观看欧美| av在线观看视频网站免费| 国产高清国产精品国产三级 | 精华霜和精华液先用哪个| 校园人妻丝袜中文字幕| 日本黄色片子视频| kizo精华| 精华霜和精华液先用哪个| 日日摸夜夜添夜夜爱| 妹子高潮喷水视频| 欧美激情极品国产一区二区三区 | 国产黄频视频在线观看| 亚洲精品国产色婷婷电影| 水蜜桃什么品种好| 尾随美女入室| 久久99蜜桃精品久久| 国产亚洲欧美精品永久| 久久久久国产精品人妻一区二区| 在线精品无人区一区二区三 | 99热这里只有是精品50| 午夜福利网站1000一区二区三区| 日韩av免费高清视频| 国产老妇伦熟女老妇高清| 亚洲精品一区蜜桃| 十八禁网站网址无遮挡 | 最近的中文字幕免费完整| 日本wwww免费看| av在线观看视频网站免费| 亚洲怡红院男人天堂| 亚洲欧洲日产国产| 欧美zozozo另类| 国产亚洲精品久久久com| 国产精品一二三区在线看| 丝袜喷水一区| 久久鲁丝午夜福利片| 大话2 男鬼变身卡| 日本色播在线视频| 777米奇影视久久| 国产高清有码在线观看视频| 国产人妻一区二区三区在| 成人二区视频| 国产淫语在线视频| av福利片在线观看| 欧美日韩视频精品一区| 黑人高潮一二区| av播播在线观看一区| 亚洲精品乱码久久久久久按摩| 欧美日韩精品成人综合77777| 2018国产大陆天天弄谢| 久久精品久久精品一区二区三区| 在线 av 中文字幕| 亚洲精品456在线播放app| 国产日韩欧美在线精品| 久久久久国产网址| 欧美精品一区二区大全| 美女国产视频在线观看| 欧美日韩视频高清一区二区三区二| 久久久色成人| 纯流量卡能插随身wifi吗| 中国美白少妇内射xxxbb| 久久热精品热| 日韩中文字幕视频在线看片 | 男人舔奶头视频| 综合色丁香网| 精品亚洲乱码少妇综合久久| 精品人妻偷拍中文字幕| 久久精品久久精品一区二区三区| 日韩视频在线欧美| 国产精品成人在线| 毛片一级片免费看久久久久| 一级爰片在线观看| 亚洲精品亚洲一区二区| 少妇猛男粗大的猛烈进出视频| av在线app专区| 亚洲国产高清在线一区二区三| 在线观看三级黄色| 亚洲成人一二三区av| www.av在线官网国产| 我的女老师完整版在线观看| 国产乱来视频区| 十八禁网站网址无遮挡 | av网站免费在线观看视频| 欧美成人精品欧美一级黄| 韩国高清视频一区二区三区| 色视频www国产| 国产69精品久久久久777片| 国产成人a区在线观看| 看十八女毛片水多多多| 亚洲一区二区三区欧美精品| 国产欧美另类精品又又久久亚洲欧美| 久久6这里有精品| 少妇人妻精品综合一区二区| 中文字幕人妻熟人妻熟丝袜美| 欧美激情国产日韩精品一区| 高清黄色对白视频在线免费看 | 3wmmmm亚洲av在线观看| 美女xxoo啪啪120秒动态图| 亚洲精品成人av观看孕妇| 国产淫语在线视频| 日韩av不卡免费在线播放| 50天的宝宝边吃奶边哭怎么回事| 欧美精品高潮呻吟av久久| 最近最新中文字幕大全免费视频 | 五月开心婷婷网| 女人精品久久久久毛片| 久久久精品区二区三区| 亚洲欧美精品自产自拍| 国产成人av激情在线播放| 国产在线一区二区三区精| 亚洲国产精品一区二区三区在线| 久久ye,这里只有精品| 亚洲精品国产区一区二| 亚洲九九香蕉| 久久热在线av| 欧美性长视频在线观看| 欧美性长视频在线观看| 亚洲美女黄色视频免费看| bbb黄色大片| 国产精品免费大片| 免费看av在线观看网站| 欧美日韩亚洲综合一区二区三区_| 午夜激情久久久久久久| 久久人人爽人人片av| 日韩,欧美,国产一区二区三区| 女人精品久久久久毛片| 亚洲精品中文字幕在线视频| 久久精品国产亚洲av高清一级| 亚洲 欧美一区二区三区| 亚洲精品国产区一区二| 国产成人精品久久久久久| 亚洲av电影在线观看一区二区三区| 三上悠亚av全集在线观看| 最近中文字幕2019免费版| 啦啦啦在线观看免费高清www| 亚洲欧美日韩另类电影网站| 久久人妻福利社区极品人妻图片 | 少妇精品久久久久久久| 麻豆av在线久日| 黄色毛片三级朝国网站| 国产伦理片在线播放av一区| 狂野欧美激情性xxxx| 久热爱精品视频在线9| 五月开心婷婷网| 一级片'在线观看视频| 另类精品久久| 一级片免费观看大全| 看免费av毛片| 久久亚洲精品不卡| 精品一区二区三区四区五区乱码 | 大陆偷拍与自拍| 国产97色在线日韩免费| 国产视频一区二区在线看| 亚洲av电影在线进入| 黑人猛操日本美女一级片| 亚洲国产欧美日韩在线播放| 国产免费福利视频在线观看| 亚洲成人手机| 久久av网站| 十分钟在线观看高清视频www| 国产精品99久久99久久久不卡| 无遮挡黄片免费观看| kizo精华| 久久久久久久大尺度免费视频| 国产亚洲一区二区精品| 亚洲伊人久久精品综合| 超碰97精品在线观看| 国产精品人妻久久久影院| 啦啦啦 在线观看视频| 亚洲国产最新在线播放| 久久久久久亚洲精品国产蜜桃av| 久久天堂一区二区三区四区| 丝袜脚勾引网站| 麻豆av在线久日| 日韩视频在线欧美| 亚洲欧美一区二区三区国产| av在线播放精品| 爱豆传媒免费全集在线观看| 少妇人妻 视频| 日韩一本色道免费dvd| 免费看十八禁软件| 嫁个100分男人电影在线观看 | 亚洲精品美女久久av网站| 亚洲av美国av| 91麻豆精品激情在线观看国产 | 国产午夜精品一二区理论片| av视频免费观看在线观看| 中文字幕制服av| 涩涩av久久男人的天堂| 日本五十路高清| av在线老鸭窝| 波多野结衣av一区二区av| 丝袜在线中文字幕| 蜜桃国产av成人99| 18禁观看日本| 亚洲少妇的诱惑av| 亚洲第一av免费看| 精品一品国产午夜福利视频| 赤兔流量卡办理| 国产一区二区三区综合在线观看| 精品人妻熟女毛片av久久网站| 成年人免费黄色播放视频| 国产精品av久久久久免费| 欧美在线一区亚洲| 国产精品 国内视频| 亚洲av欧美aⅴ国产| 男女边摸边吃奶| 一级片'在线观看视频| 最近手机中文字幕大全| 国产高清视频在线播放一区 | 国产高清videossex| 久久 成人 亚洲| 色播在线永久视频| 桃花免费在线播放| 欧美人与性动交α欧美软件| 免费看av在线观看网站| 宅男免费午夜| av网站免费在线观看视频| 精品国产乱码久久久久久男人| 中国美女看黄片| 别揉我奶头~嗯~啊~动态视频 | 在线看a的网站| 一区二区三区精品91| 国产成人av教育| 亚洲精品国产一区二区精华液| 久久精品久久精品一区二区三区| 久久精品国产综合久久久| 亚洲精品日本国产第一区| 国产成人欧美| 久久精品熟女亚洲av麻豆精品| 91精品三级在线观看| 欧美日韩综合久久久久久| 18在线观看网站| 少妇的丰满在线观看| 中文字幕av电影在线播放| 亚洲自偷自拍图片 自拍| 97精品久久久久久久久久精品| 自线自在国产av| 日韩精品免费视频一区二区三区| 国产91精品成人一区二区三区 | 午夜免费鲁丝| 免费在线观看视频国产中文字幕亚洲 | 久久久久久久大尺度免费视频| 天天躁狠狠躁夜夜躁狠狠躁| 亚洲精品美女久久久久99蜜臀 | 久久热在线av| 午夜91福利影院| 美女国产高潮福利片在线看| 亚洲国产欧美日韩在线播放| 女人高潮潮喷娇喘18禁视频| 少妇被粗大的猛进出69影院| 成在线人永久免费视频| 免费人妻精品一区二区三区视频| 国产成人一区二区三区免费视频网站 | 人成视频在线观看免费观看| 美女大奶头黄色视频| av福利片在线| 一边亲一边摸免费视频| 一级毛片女人18水好多 | 十八禁网站网址无遮挡| 男女下面插进去视频免费观看| 极品人妻少妇av视频| 9色porny在线观看| 欧美精品啪啪一区二区三区 | 亚洲欧洲日产国产| a 毛片基地| 美女视频免费永久观看网站| 午夜影院在线不卡| 看免费成人av毛片| 男的添女的下面高潮视频| 久久九九热精品免费| 最新在线观看一区二区三区 | 国产免费视频播放在线视频| 亚洲精品在线美女| avwww免费| 国产成人精品久久二区二区91| av福利片在线| 啦啦啦在线免费观看视频4| 少妇 在线观看| 成年动漫av网址| 国产精品一区二区在线观看99| 国产色视频综合| 亚洲国产中文字幕在线视频| av国产久精品久网站免费入址| 国产成人欧美| 1024香蕉在线观看| 人成视频在线观看免费观看| 菩萨蛮人人尽说江南好唐韦庄| 欧美久久黑人一区二区| 少妇 在线观看| 日韩电影二区| 国产精品av久久久久免费| 婷婷色综合大香蕉| www.熟女人妻精品国产| 欧美精品啪啪一区二区三区 | 精品国产乱码久久久久久小说| 亚洲国产欧美在线一区| 无限看片的www在线观看| 亚洲av成人不卡在线观看播放网 | 超碰97精品在线观看| 赤兔流量卡办理| 99国产精品一区二区蜜桃av | 黄频高清免费视频| 日韩制服骚丝袜av| 一本久久精品| 亚洲国产看品久久| 成年美女黄网站色视频大全免费| 国产一区二区三区av在线| av天堂久久9| a 毛片基地| 久久精品熟女亚洲av麻豆精品| 久久人妻福利社区极品人妻图片 | 校园人妻丝袜中文字幕| 亚洲av综合色区一区| 99re6热这里在线精品视频| 母亲3免费完整高清在线观看| 久久久国产一区二区| 女人爽到高潮嗷嗷叫在线视频| 好男人电影高清在线观看| 国产成人精品在线电影| 最黄视频免费看| 黄色毛片三级朝国网站| 亚洲精品成人av观看孕妇| 女人高潮潮喷娇喘18禁视频| 久久久国产精品麻豆| 久久久久精品人妻al黑| 国产成人精品在线电影| 大型av网站在线播放| 少妇的丰满在线观看| 9191精品国产免费久久| 91字幕亚洲| 国产亚洲欧美在线一区二区| 久久久久久亚洲精品国产蜜桃av| 亚洲熟女毛片儿| 大陆偷拍与自拍| 国产视频首页在线观看| 亚洲少妇的诱惑av| cao死你这个sao货| 久久精品久久精品一区二区三区| 亚洲精品一区蜜桃| av网站免费在线观看视频| 高潮久久久久久久久久久不卡| 一个人免费看片子| 国产精品久久久久久人妻精品电影 | 久久精品久久久久久久性| 国产成人精品在线电影| √禁漫天堂资源中文www| 三上悠亚av全集在线观看| 人成视频在线观看免费观看| 真人做人爱边吃奶动态| 考比视频在线观看| 国产欧美日韩综合在线一区二区| 亚洲午夜精品一区,二区,三区| 国产精品九九99| 男女边吃奶边做爰视频| 黄色 视频免费看| 精品福利永久在线观看| 欧美日韩国产mv在线观看视频| 久久天堂一区二区三区四区| 人人妻人人澡人人看| 久久国产精品影院| 国产成人精品在线电影| 久久午夜综合久久蜜桃| 久久久国产欧美日韩av| 婷婷色综合大香蕉| 亚洲精品一卡2卡三卡4卡5卡 | 成人国产一区最新在线观看 | 精品福利观看| 亚洲中文av在线| 国产精品免费大片| 成人国产一区最新在线观看 | 欧美日韩一级在线毛片| 99久久综合免费| 亚洲第一青青草原| 91麻豆av在线| 亚洲图色成人| 欧美亚洲 丝袜 人妻 在线| 日韩免费高清中文字幕av| 中文字幕人妻丝袜一区二区| 91麻豆av在线| www.自偷自拍.com| 最新的欧美精品一区二区| 国产精品久久久久成人av| 国产在线免费精品| 亚洲国产精品一区三区| 精品人妻一区二区三区麻豆| 后天国语完整版免费观看| 亚洲午夜精品一区,二区,三区| 高清视频免费观看一区二区| 多毛熟女@视频| 国产亚洲av片在线观看秒播厂| 成人黄色视频免费在线看| 午夜激情av网站| 黄频高清免费视频| 国产成人a∨麻豆精品| 国产成人一区二区在线| 亚洲国产中文字幕在线视频| videos熟女内射| 如日韩欧美国产精品一区二区三区| 亚洲精品国产av成人精品| 亚洲国产日韩一区二区| √禁漫天堂资源中文www| 男女午夜视频在线观看| 国产极品粉嫩免费观看在线| 建设人人有责人人尽责人人享有的| 亚洲av日韩在线播放| 精品久久久精品久久久| 黄色一级大片看看| 我的亚洲天堂| 另类亚洲欧美激情| 高清av免费在线| 97在线人人人人妻| 免费在线观看黄色视频的| 激情视频va一区二区三区| 国产黄色免费在线视频| 老司机亚洲免费影院| 99国产综合亚洲精品| 久久精品久久精品一区二区三区| a级片在线免费高清观看视频| xxx大片免费视频| 亚洲伊人色综图| 别揉我奶头~嗯~啊~动态视频 | 国产精品一二三区在线看| 国产成人精品久久二区二区91| 精品卡一卡二卡四卡免费| 日韩,欧美,国产一区二区三区| 国产片内射在线| 久久精品熟女亚洲av麻豆精品| 日韩中文字幕视频在线看片| 97在线人人人人妻| 黄色片一级片一级黄色片| 另类精品久久| 免费高清在线观看视频在线观看| 97人妻天天添夜夜摸| 国产免费福利视频在线观看| 好男人视频免费观看在线| 亚洲av日韩精品久久久久久密 | 悠悠久久av| 欧美日韩视频精品一区| 一边摸一边做爽爽视频免费| 成人亚洲精品一区在线观看| 久久热在线av| 国产精品久久久人人做人人爽| 色综合欧美亚洲国产小说| 在线观看人妻少妇| 一本—道久久a久久精品蜜桃钙片| 成人国产一区最新在线观看 | 老汉色∧v一级毛片| 91国产中文字幕| 一区二区三区四区激情视频| 亚洲,欧美,日韩| 啦啦啦在线免费观看视频4| 亚洲人成77777在线视频| 人妻人人澡人人爽人人| 最近中文字幕2019免费版| 人妻人人澡人人爽人人| 午夜福利在线免费观看网站| 大片免费播放器 马上看| 欧美日韩综合久久久久久| 中文字幕精品免费在线观看视频| 最近中文字幕2019免费版| av国产久精品久网站免费入址| 女人久久www免费人成看片| 中文字幕av电影在线播放| 国产男女超爽视频在线观看| 精品卡一卡二卡四卡免费| 亚洲九九香蕉| 五月天丁香电影| 美女大奶头黄色视频| 国产日韩欧美视频二区| 国产黄频视频在线观看| 亚洲精品中文字幕在线视频| 青草久久国产| 一区二区三区激情视频| 1024视频免费在线观看| 免费在线观看视频国产中文字幕亚洲 | 国产日韩欧美视频二区| 亚洲精品久久久久久婷婷小说| 黄网站色视频无遮挡免费观看| 久久久亚洲精品成人影院| 午夜激情av网站| 国产黄色免费在线视频| 久久人人爽人人片av| 一级,二级,三级黄色视频| 侵犯人妻中文字幕一二三四区| 国产黄色免费在线视频| 亚洲中文字幕日韩| 最新的欧美精品一区二区| 欧美亚洲日本最大视频资源| 操出白浆在线播放| 好男人电影高清在线观看| 在线av久久热| 精品视频人人做人人爽| 夫妻午夜视频| 尾随美女入室| 亚洲精品久久午夜乱码| www.999成人在线观看| 精品第一国产精品| 十八禁网站网址无遮挡| 久久精品久久精品一区二区三区| 国产又色又爽无遮挡免| 亚洲国产精品一区三区| 国产一区二区三区综合在线观看| 啦啦啦中文免费视频观看日本| 国产欧美日韩精品亚洲av| 国产成人免费无遮挡视频| 国产精品.久久久| 欧美乱码精品一区二区三区| 国产高清不卡午夜福利| 老司机深夜福利视频在线观看 | 亚洲九九香蕉| 亚洲欧美一区二区三区久久| videosex国产| 精品少妇久久久久久888优播| 久久亚洲国产成人精品v| 久久久国产一区二区| 亚洲欧美色中文字幕在线| 视频区图区小说| 亚洲午夜精品一区,二区,三区| 99国产精品免费福利视频| 七月丁香在线播放| 水蜜桃什么品种好| 精品国产一区二区三区久久久樱花| 国产精品欧美亚洲77777| 国产日韩欧美亚洲二区| 制服诱惑二区| 亚洲精品在线美女| 在线观看免费午夜福利视频| 999久久久国产精品视频| 国产一区二区在线观看av| 国产真人三级小视频在线观看| 亚洲欧洲精品一区二区精品久久久| www.999成人在线观看| 好男人视频免费观看在线| 免费观看av网站的网址| 国产精品久久久久久精品电影小说| 中国美女看黄片| 国产欧美日韩一区二区三区在线| 在线av久久热| 赤兔流量卡办理| 啦啦啦中文免费视频观看日本| 精品国产一区二区三区久久久樱花| 成年人午夜在线观看视频| 在线观看免费日韩欧美大片| 晚上一个人看的免费电影| 日韩制服骚丝袜av| 青春草视频在线免费观看|