Proteomic Profiling Of Human Liver Mitochondria & Multiple Regression Analysis On MRNA/Protein Abundance Correlation

As an important organelle, mitochondria generate the majority of ATP in the cell, and their dysfunction has been implicated in many different diseases. Although mitochondria have been studied for a long time, its composition is still unclear to some degree. Proteomics technologies provide a powerful tool to give a survey of the proteins in the given tissue, cell or organelle, by which we could get more information about it.Organelle separation is one of the most important steps for establishment of subcellular protein expression profile. The liver of C57BL/6J mouse was chosen as the model to find the optimum method for subcellular preparation. The method we selected could obtain the multiple fractions including plasma membrane, mitochondria, nucleus, endoplasmic reticulum, and cytosol from a single homogenate. We systematically evaluated the purity, efficiency, and integrity of mitochondria by protein yield, immunoblotting, and transmission electron microscopy. Subsequently, highly purified mitochondria from human liver were used to compile a protein expression profile with strategy that combined SDS-PAGE separation with liquid chromatography gas phase fractionation (GPF) tandem mass spectrometry analysis. 95% confidence and minimum two peptides match (95P2) were used for protein identificion.To predict the subcellular location of all identified proteins more precisely, a step by step method was employed to evaluate the protein localization. Firstly, based on the quantitation result, KNN (k nearest neighbour) algorithm was used to classify the data. Secondly, we used bayes algorithm to give an evaluation of protein localization, which integrate the results of KNN, pTAGET, Proteome Analyst, WoLFPSORT, TargetP, MitoPred and NUCLEO. By Bayes algorithm, 774 proteins were localized to mitochondria, which is the largest subcelluar protein localization data in human liver to date. We gained 96 protein new localization and 291 new proteins localization information.To mine the potential function of newly localized proteins, Gene Ontology, Pfam and protein interaction data were used for subsequent analysis. The distribution of GO terms in our data showed that proteins involved in energy metabolism and electron transporter were in larger percent, which is the feature of physiological function of liver mitochondria. Proteins involved in cell communication were identified scarely in mitochondria before. New localization of 25 cell communication proteins would refresh our knowledge about signal transduction in mitochondria. Of them, G proteins were thought to be localized in cell membrane. Recent evidence suggests that receptors acting through G proteins also transfer signals across the nuclear membrane. Here, eight subunits of G protein were identified in mitochondria, implicating that G protein receptor signaling may be a common feature to all membranes.95% confidence, two-peptide or above match (95P2) and 99% confidence, one-peptide or above match (99P1) were both reliable criteria for protein identification. Based on protein identifications of human liver from 7 batch experiments, we calculated the possibility that a protein could be identified by one experiment with 95P2 criteria among different N_obsbl (the number of observable peptide per protein) range. Compared with average protein identified possibility (8.9%), low N_obsbl (<35) proteins were of low identification (0.85%). although proteins identified in 95P2 dataset were more 25% than those in 99P1 dataset, for low N_obsbl proteins, identifications in 95P2 dataset were lower than that in 99P1 dataset, which could provide the explaination for the low identification of OXPHOS proteins in our data, and also help us to understand MS-based false negative identification.Integrating of transcriptomics and proteomics could provide the proper context for interpreting gene expression data from high-throughput analyses. As mRNA is eventually translated into protein, one might assume that there should be some sort of correlation between the level of mRNA and that of protein. However, as cells have adopted elaborate regulatory mechanisms at the levels of transcription, post-transcription translation and post-translation, transcript and protein abundance measurements may not be concordant. The discordant of mRNA/protein expression level may reveal additional post-transcriptional regulatory junctures as candidates for the design of therapeutics. The influence of some factors on mRNA/protein correlation variations would make the correlation between mRNA and protein more clearly.Variation of mRNA/protein correlation was effected by not only biological factors, but also technological reasons. However, technological limitations are poorly investigated in prior literatures. In the study, based on the high quality data from human liver proteome project (HLPP), we deduced a technological parameter RIP_pro (protein relative identification possibility), and the reliability and rationality of RIP_pro as technique parameter were also validated. We used multiple regression to analyze the impact of multiple factors on mRNA/protein correlation variations in human liver. Our results showed that the variation of protein abundance was mostly affected by mRNA abundance (36%), followed by RIP_pro, protein half-life period and protein measurement variation (11%, 8% and 2%, respectively). We also found that mRNA/protein correlation was different among different cellular functional categories, and the variation of which was most affected by protein abundance variation (74%), followed by RIPpro (5%).Pearson's correlation analysis of mRNA/protein was performed in distance hierarchy. Compared with 100% data derived correlation value (r=0.59), the correlation of mRNA/protein abundance from 75% data were better(r=0.75). To further substantiate the relationship between mRNA/protein correlation and the protein physical/chemical properties, we analyzed 94 proteins (5% of 1,881 proteins) from which the distance to the trendline was longer than average as a "outline" group. Group1, protein abundance was higher than that estimated by single regression model from mRNA abundance; Group2, protein abundance was lower than that estimated by single regression model from mRNA abundance.Hypergeometry distribution analysis of "outline" proteins about their functional categories and RIPpro were carried out. There appear two features: firstly, proteins from "metabolism" functional categories were enriched in the Group1, but deleted in the group2; while proteins from "signal" functional categories were deleted in the group1, but enriched in the group2; secondly, low RIPpro proteins were significantly enriched in group1. Considering the discordant of mRNA/protein expression level may mean candidates for therapeutics, OMIM (Online Mendelian Inheritance in Man) annotion were performed for 94 "outline" proteins, and demonstrated that 90 of them were related to disease.mRNA/protein correlations were diverse among different functional categories. In "metabolism" proteins, 40% of the protein abundance variability was explained by mRNA abundance; while in "signal" proteins, the variation of protein abundance was mostly affected by RIPpro (22.38%), followed by mRNA abundance (17%). Those results suggest that "signal" proteins would adopt more elaborate regulatory mechanisms at the levels of post-transcription and post-translation than those of "metabolism" proteins.In conclusion, we provided a strategy that could separate multiple subcellular fractions from a single homogenate. The proteome of human liver mitochondria were profiled for first time and Bayes algorithm was employed to give accurate localization for identified protiens. In proteome and transcriptome data integrating, we provided RIPpro as technological parameter to analyze the mRNA/protein correlation in human liver. The study provides the first comprehensive quantitative analysis of the mRNA/protein correlation in human liver, and adds new insights into the RIPpro in mRNA/protein correlation research. The methods for data mining of this investigation will be helpful for data processing in other large-scale datasets.