Proteome-wide remodeling of protein location and function by stress

Abstract

Protein location and function can change dynamically depending on many factors, including environmental stress, disease state, age, developmental stage, and cell type. Here, we describe an integrative computational framework, called the conditional function predictor (CoFP; http://nbm.ajou.ac.kr/cofp/), for predicting changes in subcellular location and function on a proteome-wide scale. The essence of the CoFP approach is to cross-reference general knowledge about a protein and its known network of physical interactions, which typically pool measurements from diverse environments, against gene expression profiles that have been measured under specific conditions of interest. Using CoFP, we predict condition-specific subcellular locations, biological processes, and molecular functions of the yeast proteome under 18 specified conditions. In addition to highly accurate retrieval of previously known gold standard protein locations and functions, CoFP predicts previously unidentified condition-dependent locations and functions for nearly all yeast proteins. Many of these predictions can be confirmed using high-resolution cellular imaging. We show that, under DNA-damaging conditions, Tsr1, Caf120, Dip5, Skg6, Lte1, and Nnf2 change subcellular location and RNA polymerase I subunit A43, Ino2, and Ids2 show changes in DNA binding. Beyond specific predictions, this work reveals a global landscape of changing protein location and function, highlighting a surprising number of proteins that translocate from the mitochondria to the nucleus or from endoplasmic reticulum to Golgi apparatus under stress.

Keywords: DTT and MMS; bioinformatics; dynamic function prediction; protein translocation; systems biology.

Fig. 1.

Proteome-wide prediction of conditional locations and functions under stresses. (A) Generally known protein functions, including 18 subcellular locations, 33 biological processes, and 22 molecular functions. (B) Yeast protein–protein interactions accumulated from several databases. (C) Static information of proteins, including sequence, chemical properties, motifs, and GO terms (single-protein features). (D) Model generation after generating static single-protein (denoted S) and network features (denoted N and L) up to network distance D = 2. The best combination of features is selected for each functional category using a divide-and-conquer k-nearest-neighbor method classifier. (E) Stress-specific interaction networks in individual conditions are generated by assigning different functional coherence scores to each interaction of a protein depending on the interactor’s similarity in time series gene expression profiles. (F) After generating the selected features from D using the condition-dependent networks from E, the prepared 73 classifiers compute a conditional functional map for the protein, indicating the quantitative possibility that the protein is in each function under each condition. Dynamic functions under stressful conditions are identified by calculation of significant differences in the possibility degree in a stress condition. (G) The fraction of protein pairs having the same process, function, or location (rows) shown for protein pairs involved in physical protein interaction (column 1), high coexpression (columns 2–4; NEGATIVE, negative correlation; NO, no correlation; POSITIVE, positive correlation), and both physical interaction and high coexpression (columns 5–7). Gray sectors indicate random expectation resulting from 100 permutation tests. (H) The average performance (area under the ROC curve (AUC) value) of S, N, or L feature sets for process, function, and location. Composite indicates the performance of the selected feature sets for individual functional categories.

Fig. 2.

Experimental validation of predicted locations in a stress-free condition. (A) Forty-two previously unidentified locations but correctly predicted cases as shown in new validation experiments. (Left) The heat map is the location prediction for Rrp12. By prediction, Rrp12 had the strongest signal (0.82 possibility) at nucleolus (NO). (B) Example validation experiments for 42 cases, including Rrp12, Emp24, and Vps33. Protein is marked in green, and location is marked in red. Yellow indicates high overlap between the corresponding proteins and the location markers. Red squares indicate the area that is magnified 4×. RFP-tagged Nop56, Sec66, and Snf7 were used as location markers for the nucleolus, ER, and endosome, respectively. (C) Seventeen previously misidentified locations but correctly predicted cases as shown in new validation experiments. (Lower) The heat map is the prediction for Bud20. Although originally localized to the ER, Bud20 had an almost zero possibility in this location, a prediction that we validated. (D) Some examples of the new experiments for 17 cases, including Bud20 and Acf2. Protein is marked in green, ER location is marked in red, and nucleus is marked in blue. (Scale bar: 5 μm.) (E) The coherence-mapped interaction network in location prediction of Rrp12. The node and edge sizes are proportional to the summed coherence score. The numbers in parentheses of the nodes indicate the numbers of neighbors in corresponding localizations. Full names for the abbreviations of locations and functions are shown in SI Appendix, Table S1.

Fig. 3.

Performance of predicted conditional locations under DTT and MMS conditions. (A) The tested 100 proteins under DTT condition. The heat map indicates the possibility degrees of the proteins for individual locations. The predicted locations are indicated by red +, and validated locations are white −. FN, false negative; FP, false positive; TN, true negative; TP, true positive. (B) Some examples of validated proteins under DTT condition. Green, corresponding proteins; red squares, 10× magnified. (C) The tested 100 proteins under MMS condition. (D) Some examples of validated proteins under MMS condition. (Scale bar: 10 μm.) (E and F) The summarized performance of the experimentally tested 100 proteins under the (E) DTT and (F) MMS conditions shown in A and C, respectively.

Fig. 4.

The landscape of dynamic locations and functions of yeast proteins under stresses. (A) The numbers of predicted dynamic functional events regarding biological processes, molecular functions, and locations integrated over 17 stresses. (B) Example showing the predicted functions of Par32 across diverse conditions, including untreated stress-free normal, diauxic shift, heat shock, and RNA stability conditions. The prediction shows that Par32 is likely to have a protein-binding (PB) function under stress-free conditions and transferase activity (TF) in all three stresses. H in the heat map indicates the function with the highest signal for each condition. The blue cells in the Previous column indicate that this protein had no previously known functions in GO. (C–E) The landscape of changing protein state in terms of (C) location, (D) function, and (E) process across all dynamic yeast proteins. Proteins were considered dynamic if they were significantly different (P value < 0.01) between untreated and stress conditions. Each peak (z axis) corresponds to the percentage of these proteins disappearing from one function (x axis) and appearing in another function (y axis). Colors along the x and y margins represent the total percentage of proteins with functions disappearing or appearing, respectively. Full names for the abbreviations of locations and functions are shown in SI Appendix, Table S1.

Fig. 5.

Experimental validation of dynamic locations under DTT and MMS conditions. The validated dynamic locations of (A) Tsr1, (B) Caf120, and (C) Dip5 under MMS condition and (D) Skg6, (E) Lte1, and (F) Nnf2 under DTT condition. All white cells in the “Previous” columns indicate previously known locations. The red squares indicate the areas that are magnified 10×. Full names for the abbreviations of locations and functions are shown in SI Appendix, Table S1. BD, bud; BN, bud neck; CP, cell periphery; CY, cytoplasm; NO, nucleolus; VO, vacuole. (Scale bar: 10 μm.)

Fig. 6.

Experimental validation of dynamic functions under DTT and MMS conditions. (A) PCR analysis to check the association of Rpa43 with the rDNA regions (25S, NTS1, NTS2, and 5.8S regions of the rDNA). The Rpa43-associated DNA was prepared by ChIP and amplified using primer sets located in the indicated regions. PCR with the primer set within the CUP1 region is used as an internal control. No tag indicates samples from nontagged cells. (B) The association of Rpa43 with the NTS1 region of rDNA under DTT treatment using quantitative real-time PCR. Amplification efficiencies were validated and normalized against CUP1, and fold increases were calculated using the comparative cycle threshold (C_t) method. Values are the mean of three independent experiments, and error bars indicate SDs. (C) Western blot analysis showing the protein levels of Rpa43 under DTT treatment. (D) Fluorescence images showing the localization changes of Rpa43 under DTT treatment. Arrows indicate Rpa43 spreading out from the nucleolus. RFP-tagged Nop56 was used as a nucleolar marker. Red squares indicate the area that is magnified 5×. DIC, differential interference contrast images. (Scale bar: 5 μm.) (E) The association of Ino2 with the promoter region of ARG4 was measured using a ChIP assay under DTT treatment. Relative fold enrichment refers to the relative ratio of PCR products amplified from immunoprecipitated DNA to products from input DNA. PCR amplicons used in ChIP assays are indicated below the promoter region of the target gene. (F) The expression of ARG4 was measured under DTT treatment. Cells were treated with 2.5 mM DTT for the indicated times at 25 °C, and the amount of ARG4 was analyzed by quantitative real-time RT-PCR. (G) The association of Ids2 with the promoter region of SPS1 was measured using a ChIP assay under MMS treatment. NT, no tag. **P value < 0.01 using a Student t test.