Accounting for Protein Subcellular Localization: A Compartmental Map of the Rat Liver Proteome

Abstract

Accurate knowledge of the intracellular location of proteins is important for numerous areas of biomedical research including assessing fidelity of putative protein-protein interactions, modeling cellular processes at a system-wide level and investigating metabolic and disease pathways. Many proteins have not been localized, or have been incompletely localized, partly because most studies do not account for entire subcellular distribution. Thus, proteins are frequently assigned to one organelle whereas a significant fraction may reside elsewhere. As a step toward a comprehensive cellular map, we used subcellular fractionation with classic balance sheet analysis and isobaric labeling/quantitative mass spectrometry to assign locations to >6000 rat liver proteins. We provide quantitative data and error estimates describing the distribution of each protein among the eight major cellular compartments: nucleus, mitochondria, lysosomes, peroxisomes, endoplasmic reticulum, Golgi, plasma membrane and cytosol. Accounting for total intracellular distribution improves quality of organelle assignments and assigns proteins with multiple locations. Protein assignments and supporting data are available online through the Prolocate website (http://prolocate.cabm.rutgers.edu). As an example of the utility of this data set, we have used organelle assignments to help analyze whole exome sequencing data from an infant dying at 6 months of age from a suspected neurodegenerative lysosomal storage disorder of unknown etiology. Sequencing data was prioritized using lists of lysosomal proteins comprising well-established residents of this organelle as well as novel candidates identified in this study. The latter included copper transporter 1, encoded by SLC31A1, which we localized to both the plasma membrane and lysosome. The patient harbors two predicted loss of function mutations in SLC31A1, suggesting that this may represent a heretofore undescribed recessive lysosomal storage disease gene.

Fig. 1.

Subcellular fractionation and organelle assignment. A, Fractionation scheme showing differential centrifugation (Left) and orthogonal fractionation of L1 (Right). Organelles enriched in given fraction are indicated. Note that pellets are washed by resuspension and recentrifugation, with pooled supernatants used for the next step (Experimental Procedures and Supplemental Materials). B, Marker enzyme analysis. Fractions are labeled on plots showing the lysosomal marker enzyme. Marker enzymes for different compartments are as follows: mitochondria, cytochrome oxidase; lysosome, β-galactosidase; peroxisome, catalase; ER, alkaline α-glucosidase; PM, alkaline phosphodiesterase; cytosol, dipeptidyl-peptidase III. Differential and Nycodenz Centrifugation, data from Expt A; L1 Sucrose Centrifugation, data from Expt C.

Fig. 2.

Distribution of marker proteins used for classification of subcellular compartments. A, Expt A. B, Expt B, Red lines, Profiles of individual marker proteins for indicated compartments (Supplemental Table 2) determined from MS data. Black-yellow dashed lines, consensus profiles for indicated compartments. Other lines, data from marker enzyme activity assays. A: black solid lines (mitochondria, cytochrome oxidase; lysosome, β-galactosidase; peroxisome, catalase; ER, alkaline α-glucosidase; PM, alkaline phosphodiesterase; cytosol, dipeptidyl-peptidase III) or in black dashed lines (lysosome, TPP1; peroxisome, urate oxidase; ER, NADH cytochrome c reductase). B: as above, except in the lysosome/peroxisome panel, the peroxisomal maker catalase is denoted by a blue solid line and urate oxidase was not quantified.

Fig. 3.

Comparison of classification coefficient in Expts A and B. Point estimates for the 2952 proteins meeting the criteria of having ≥2 peptides and ≥3 spectra in each of the two experiments (Table I) are plotted against each other and fit using linear regression. For Lyso+Perox, the sum of the two classification coefficients in Expt A is plotted against the combined lysosome/peroxisome classification coefficient from Expt B. For ER+Golgi+PM (microsomal distribution), the sum of the three classification coefficients is plotted against each other. For All*, the sum of the ER, Golgi and PM coefficients are used instead of the individual coefficients.

Fig. 4.

Principal component analysis. Plots were made for all proteins with at least two peptides and three spectra in Expts A (A) and B (B). Proteins with point estimates <0.7 are shown in gray, all others are represented by indicated symbol. Plots were generated using the “prcomp” function in R version 3.2 to compute the principal components, with data centered and scaled to have a unit variance prior to the analysis. Explained variances are indicated on the plots.

Fig. 5.

Analysis of Triton-shift experiments. A, Distribution of select lysosomal, peroxisomal and mitochondrial proteins following sucrose density centrifugation of L1 fractions from control (C) and Triton-treated (T) rats. Proteins were selected as having ≥4 peptides in Expts C and D that have been localized to mitochondria (from MitoCarta v2.0 (29) with a FDR ≤0.001); lysosomes (from compendium reviewed in (53) omitting MPO (54)); and peroxisomes (PeroxisomeDB2.0 (32) omitting SOD1(17)). B, Distribution of proteins with at least 2 peptides and 3 spectra (gray points) with other symbols indicating marker proteins from supplemental Table S2. C, As in Panel B, with other symbols indicating assignments from Expt A for proteins with a predominant compartment lower confidence interval ≥ 0.7 in Expts A and B. Lines indicate boundaries for proportional assignment between lysosomal and other locations (dashed line 1:1; dotted lines, 2:1 and 1:2). D, Comparison of Expts C and D.

Fig. 6.

Benchmarking protein location from the Prolocate high stringency data set. A, Receiver operating characteristic (ROC) curves showing ability of Expt A classification coefficients (symbols as in Fig. 5) to correctly predict location using proteins with single compartment assignment in the CDbBS (23) as reference sets. Sensitivity is true positive rate and 1-specificity is false positive rate. Dashed line indicates ROC curve for an uninformative test. B, As above, using the curated CDbBS (see Text). Area under the curve for data sets shown in Panel A/Panel B are as follows: Mitochondria, 0.9782/0.9994; Lysosome, 0.9930/1.000; Peroxisome, 0.9708/0.9758; ER, 0.9550/0.9978; Golgi, 0.8353/0.9995; PM, 0.8191/0.9996; Cytosol, 0.7499/0.9929; Nucleus, 0.6825/0.9964.

Fig. 7.

Comparison of Prolocate high stringency data with other databases. A, Mito classification coefficient for 2640 proteins listed in the MitoCarta 2.0 all mouse genes database (30) that are found in our high stringency classification set, broken down by MitoCarta false discovery rate (FDR); B–E, All classification coefficient values are plotted for each protein found in our high stringency classification set that was found in the following databases: B and C, MitoCarta 2.0 list of designated mitochondrial proteins with (B) FDR ≤ 0.1 (394 proteins) and (C) FDR > 0.1 (226 proteins); D, MitoMiner, v3.1 - 2015_04 set using the IMPI filter (31) (998 proteins); E, PeroxisomeDB v2.0 (32)(58 proteins); and F, CLEAR network (35) (159 proteins). Red bars represent mean classification coefficient value.

Fig. 8.

Localization assignments for dynamic multiprotein complex proteins and a disease gene candidate. A, COG proteins: COG1, open circles; COG2, filled circles; COG3, open squares; COG4, filled squares; COG5, open triangles; COG6, filled triangles; COG7, open diamonds; COG8, filled diamonds. B, AP1 complex: AP1B1, open circles; AP1G2, filled circles; AP1M1, open squares; AP1S1, filled squares. AP2 complex: AP2A1, open circles; AP2A2, filled circles; AP2B1, open squares; AP2M1, filled squares; AP2S1, open triangles. AP3 complex: AP3B1, open circles; AP3B2, filled circles; AP3D1, open squares; AP3M1, filled squares; AP3S2, open triangles. AP5 complex: AP5B1, open circles; AP5M1, filled circles; AP5S1, open squares. C, SLC31A1.

Fig. 9.

Compound heterozygosity of SLC31A1 in a patient with a previously unsolved lysosomal storage disease. Yellow arrows indicate direction of sequencing of plasmids containing PCR-amplified DNA from each mutant allele. Mutant and wild-type nucleotides on electropherograms are boxed in red and black, respectively. The schematic was generated by the UCSC genome browser Custom Tracks tool, with exons shown as thick solid boxes. Exome sequencing and confirmatory PCR/Sanger sequencing revealed that patient 82RD265 was heterozygous at two positions in SLC31A1 (ENSG00000136868): 9,116021039 C/T, flanking sequence GCC [C/T]GA GAG); 9,116022721 G/T, flanking sequence GCA [G/T]TG GTA. These result in two changes in the protein: Arg90Stop and Val181Leu (numbering based on (ENSP00000363329). To determine if these were on same or different copy of the SLC31A1 gene, long range PCR was conducted to amplify 2.1 kb fragments (forward primer, CAAGCAGTCTGACCAAAAGGT; reverse primer, CTTCAACAACTTCCCACTGCA) containing both regions of interest. After subcloning, sequencing of individual plasmids indicated compound heterozygosity.