Ordered Organelle Degradation during Starvation-induced Autophagy

EXPERIMENTAL PROCEDURES

Cell Culture—

MCF7-eGFP-LC3 cells (24) were cultured in Dulbecco's modified Eagle's medium (Invitrogen) supplemented with 100 units/ml penicillin, 100 μg/ml streptomycin, glutamine, 10% dialyzed fetal calf serum (Invitrogen) and labeled with either l-arginine and l-lysine, l-[U-¹³C₆,¹⁴N₄]arginine and l-[²H₄]lysine, or l-[U-¹³C₆,¹⁵N₄]arginine and l-[U-¹³C₆,¹⁵N₂]lysine (Cambridge Isotope Laboratories, Andover, MA; Sigma-Aldrich) (2 × 15-cm cell culture dishes per condition; ∼95% confluent). Cells were washed three times in PBS before they were amino acid-starved in Hanks’ balanced salt solution containing calcium, magnesium, and 1 g of sucrose/liter (Invitrogen) for 0, 6, and 18 h, respectively. A second identically labeled set of MCF7 cells was starved for 3, 6, and 36 h. All cell populations were serum-starved for 36 h in total (serum + amino acid starvation), and all treatments were carried out at 37 °C. The cells were scraped in cell scraping buffer (0.25 m sucrose, 1 mm sodium ortho-vanadate, 5 mm NaF, 5 mm β-glycerophosphate, and protease inhibitor mixture (Complete™ tablets, Roche Diagnostics)) and normalized by cell counting, and the 6-h amino acid-starved [¹³C₆]Arg/[²H₄]Lys cells served as a common reference point. To inhibit the different degradation pathways during 12-h amino acid starvation (2 × 15-cm cell culture plates per condition; 95% confluent) the following inhibitors were used (all from Sigma-Aldrich): MG132 (1 μm), epoxomicin (2 μm), and 3-methyladenine (10 mm).

Mixed cells were centrifuged for 5 min at 1800 rpm and lysed in 6 m urea and 2 m thiourea, and 2% Benzonase (Merck) was added before samples were concentrated on spin tubes (cutoff, 500 Da). Protein mixtures were separated by SDS-PAGE (4–12% bis-Tris gradient gel, NuPAGE, Invitrogen), gel lanes were cut into 15 slices, samples were in-gel digested (25), and resulting peptide mixtures were STAGE-tipped as described (26).

Confocal Microscopy—

Cells were grown on coverslips as described above and rinsed three times with PBS before they were fixed in 3.7% paraformaldehyde for 15 min. Coverslips were mounted on glass slides and imaged using a Zeiss LSM 510 Meta laser-scanning confocal microscope. For colocalization analyses cells were analyzed by live cell imaging using a POCmini-Chamber-System (LaCon, Staig, Germany). MitoTracker red and LysoTracker green were used according to the manufacturer's recommendations (Molecular Probes, Leiden, Netherlands). Each experiment was performed as two biological replicates, and three pictures were taken per experiment. Autophagosomes were counted with Image J software where an area between 0.07 and 0.65 μm² was set.

Mass Spectrometry—

Peptide mixtures eluted from STAGE tips were analyzed by on-line C₁₈ reversed-phase nanoscale liquid chromatography tandem mass spectrometry essentially as described previously (27) with a few modifications. Briefly experiments were performed on an Agilent 1100 nanoflow system (Agilent Technologies, Boeblingen, Germany) coupled to a linear ion trap Fourier transform ion cyclotron mass spectrometer (LTQ-FT-ICR Ultra, Thermo Fisher Scientific, Bremen, Germany) equipped with a nanoelectrospray ion source (Proxeon Biosystems, Odense, Denmark). The LTQ-FT-ICR Ultra was operated in data-dependent mode. For each cycle, a survey spectrum of 3 × 10⁶ ions between m/z 350 and 1600 was measured at a resolution of 100,000. In the linear ion trap, tandem mass spectra of the five most abundant multiply charged ions were recorded using a collision energy of 30% and a target value of 30,000. Mass spectrometric parameters were as follows: spray voltage, 2.3 kV; no sheath and auxiliary gas flow; ion transfer tube temperature, 150 °C.

Assigning Peptide Sequences Using MASCOT and Quantitation Using MSQuant—

Raw MS2 spectra were centroided and merged into a single peak list file using the in-house-developed software DTASupercharge (default values; version 1.18) (44) and searched against the human MSIPI database (version 3.28, 68688 entries; 3.32, 68,404 entries; 3.34, 68,174 entries) (28) using MASCOT 2.0 (Matrix Science, London, UK) with the following parameters: carbamidomethylcysteine was set as fixed modification and methionine oxidation, deamidation of asparagine and glutamine, and protein amino-terminal acetylation were set as variable modifications. Triple SILAC was chosen as quantification mode. Two miss cleavages were allowed, enzyme specificity was trypsin D/DP, precursor mass accuracy had to be within 30 ppm, and the fragment spectra mass accuracy was set at 0.8 Da. The identified peptides were recalibrated using MSQuant (version 1.4.3a89) (44), results were combined using MGF combiner (version 1.05) and re-searched using MASCOT 2.0 with the above mentioned parameters except that the precursor mass tolerance was set to 5 ppm. To determine the number of false-positive peptide hits the data were searched against a human MSIPI decoy database essentially as described previously (29), and the MASCOT peptide score was adjusted to yield a number of false-positive (FP) peptide identifications of less than 1% (calculated as follows: FP rate (%) = reverse hits × 2 × 100/forward hits). For a protein to be counted as identified a minimum of two unique peptides (bold red hits; minimum length, 7 amino acids) had to be sequenced and had to fulfill the determined criteria. Peptides were quantified by MSQuant using extracted ion chromatograms of the monoisotopic signals. Protein ratios were normalized so that the identified histone ratios were 1:1:1. For cluster analyses, protein ratios were normalized to 0 (calculated as x − 1 for protein ratios larger than 1 and 1 − (1/x) for protein ratios smaller than 1).

Data Analysis—

Proteins were clustered using The Institute of Genomic Research MultiExperiment Viewer (TM4) software (30). A self-organizing tree algorithm was used with the following settings: Euclidian distance as matrix with five cycles and cell variability of 0.05. The preservation of important information about the magnitude of change is at the expense of outliers causing undesired correlations between genes. Therefore, four outliers were temporarily removed and afterward manually added to cluster A. High throughput GoMiner (31) was used to investigate the Gene Ontology (GO) identities of proteins in the generated clusters and the regulated proteins in the inhibitor studies. Protein Center (Proxeon Biosystems) was used to combine redundant protein hits with 98% homology between several data sets and to analyze subcellular localizations. Babelomics was used to analyze KEGG pathways. Two biological replicates were analyzed showing essentially the same results.

A search for KFERQ-like motifs in all observed proteins was based on a 2-fold analysis of (a) all experimentally observed KFERQ-like motifs (including reverse motifs) and (b) a search based on all possible combinations of the basic patterns (1143 motifs) matching the very basic motif description supplied previously (32). The distribution of N-terminal amino acids (after removal of N-terminal methionine) in each cluster was analyzed to test whether the observed degradation patterns confer with the N-end rule where the N-terminal amino acid determines the lifespan of the protein.

RESULTS

Protein Dynamics during Amino Acid Starvation—

To follow starvation-induced protein dynamics on a proteome scale the following SILAC experiment (21) was performed. Three populations of MCF7 cells were SILAC-encoded with both arginine and lysine using three distinct isotope forms. After complete isotope incorporation the cells were amino acid-starved for different time periods, whereas the length of serum starvation was kept constant. The presence of 1 g/liter glucose allowed maximum cell survival under these stress conditions. Two time course experiments were combined using 6 h of amino acid starvation as a common time point, providing a five-time point profile (Fig. 1A). Subsequently cells were mixed and lysed, and proteins were separated by SDS-PAGE prior to in-gel digestion (25) and on-line LC-MS on a hybrid linear ion trap/Fourier transform ion cyclotron resonance mass spectrometer (LTQ-FT-ICR Ultra). All labeled peptides elute as triplets where the intensities reflect the relative amount of peptide at the corresponding time points (Fig. 1B). Quantitation was done by our in-house-developed software MSQuant yielding an average relative S.D. of 13%. As starvation leads to a cell cycle arrest (33), we assumed that proteins involved in DNA organization should stay constant during the analyzed time frame. Therefore, histones were used to normalize the data to a mixing ratio of 1:1:1. Fig. 1C shows examples of recorded and normalized profiles of proteins changing in abundance over the analyzed starvation period.

Altogether we were able to record 1486 protein profiles consisting of five time points each (supplemental Table 1) that were clustered by a self-organizing tree algorithm using Euclidian distance as matrix. Six clusters were generated as shown in Fig. 2, A–F. The first five clusters contain proteins decreasing with different kinetics (blue), and the last cluster contains proteins that stayed constant or increased (yellow). We observed a strong decrease of proteins involved in translation, such as ribosomal protein S6 (Fig. 1C), which located to cluster A. Interestingly we also detected a decrease of its upstream kinase mammalian target of rapamycin (mTOR) (FRAP1; Fig. 1C), which is known to be inhibited during starvation conditions (34). Furthermore we noticed a strong decrease in the amount of tRNA synthetases: 11 of the 17 detected were located in clusters A and B containing proteins decreasing very early during starvation.

To investigate whether the subcellular localization of proteins had an influence on their degradation dynamics, we analyzed the distribution of proteins from specific multiprotein complexes and organelles according to GO (35) using the software GoMiner and Protein Center. As multiple GO terms can be assigned to one protein results have to be interpreted carefully. Although we did not detect the entire proteome, we detected at least 10 proteins per analyzed organelle (on average 48 proteins per organelle), which represents an adequate coverage of organellar proteins to draw biological conclusions. To our surprise, we observed a correlation between the degradation profiles of proteins and their subcellular localization (Fig. 2G, supplemental Table 2, and supplemental Fig. 1). Free cytosolic proteins were rapidly degraded and located mainly in clusters A and B. The large fraction of cytosolic proteins in cluster D was due to ribosomal proteins that also carry the GO term cytosolic. Proteasomal proteins were similarly rapidly degraded and located largely in cluster B. Ribosomal proteins localized to cluster D and showed already a delayed degradation profile. Mitochondrial proteins located mainly in cluster E. As anticipated, nuclear proteins were relatively stable with 42% localized in clusters E and F as compared with 21% in clusters A and B. One-third of the nuclear proteins localized in the first two clusters were also annotated as cytosolic compared with only 10% in clusters E and F, indicating that this group belonged to rapidly degraded cytosolic proteins. Dynamic organelles involved in vesicular transport such as the Golgi apparatus, the ER, and the vacuole were difficult to address as their proteins generally carry multiple GO terms and can localize in different subcellular compartments. Proteins staying at a constant level during starvation localized to cluster F that contained e.g. ER membrane and spliceosomal proteins. A large fraction of vacuolar proteins involved in degradation processes was also found in cluster F, e.g. hexosaminidase B, palmitoyl-protein thioesterase 1, cathepsins D and B (CTSD and CTSB). The last three showed even a slight increase over time (Fig. 1C). Thus, compared with proteasomal proteins, which drop dramatically, some lysosomal proteases stay constant or increase, highlighting their importance during amino acid starvation and indicating that proteins are most likely degraded in the lysosome under such conditions in MCF7 cells. In a biological replicate 1133 proteins were quantified, and organelles showed very similar degradation characteristics (supplemental Fig. 2).

We also investigated whether certain biological processes were overrepresented in the different clusters according to the corresponding KEGG pathway annotations. For example, aminoacyl-tRNA biosynthesis and purine and pyrimidine metabolism located mainly in clusters A and B, whereas the Krebs cycle ended primarily in clusters E and F, supporting our findings regarding the different degradation kinetics of proteins from distinct subcellular localizations (Fig. 2H and supplemental Table 3).

Amino Acid Starvation-induced Autophagy—

The two major intracellular eukaryotic protein degradation pathways are the ubiquitin-proteasome system and the autophagosomal/lysosomal system. The proteasome seemed to play only a minor role in protein degradation during starvation conditions in MCF7 cells as it localized mainly to cluster B, meaning it was rapidly degraded. In addition, we did not find any specific distribution of proteins in clusters in concordance with the N-end rule that states that proteins having a specific N terminus are especially prone to proteasomal degradation (supplemental Table 4) (36).

Autophagy is blocked by mTOR, which was degraded during the starvation period (Fig. 1C), and it is known that autophagy can be induced by starvation in MCF7 cells. To determine autophagy induction, we monitored the level of autophagosome accumulation during 36 h of amino acid starvation in MCF7 cells stably expressing the autophagosome-associated protein LC3 fused to enhanced green fluorescence protein (eGFP-LC3) (24). In control cells eGFP-LC3 stained diffusely in cytosol and nucleus. As expected, during the starvation period the number of dotted eGFP-LC3 increased indicating that autophagosomes were formed in the cells (Fig. 3, A–E). After 6 h a steady-state level of ∼2 autophagosomes/100-μm² cell area was reached (Fig. 3F).

Inhibition of Protein Degradation Pathways—

To investigate the importance of the proteasomal and the autophagosomal/lysosomal degradation routes during starvation conditions we inhibited both pathways by drugs, hypothesizing that proteins being degraded by one specific pathway should increase if the respective pathway is blocked. Thus, we inhibited the proteasome by MG132 and autophagy by 3-methyladenine (3-MA) for a 12-h starvation period. The number of autophagosomes from cells treated with the proteasomal inhibitor did not differ significantly compared with starved cells (Fig. 4, A, B, and D). 3-MA inhibits class III phosphatidylinositol 3-kinases, which are required for the initial nucleation and assembly of the primary autophagosomal membrane (37, 38). Accordingly 3-MA inhibited autophagosomal formation during amino acid starvation (Fig. 4, C and D).

Protein degradation dynamics were analyzed by quantitative MS as follows. SILAC-encoded cells were starved for 12 h in the presence of the inhibitors and mixed with starved control cells. The combined cells were lysed, and lysates were processed as outlined in Fig. 1A. Looking at proteins that were degraded within the first 12 h of starvation in the initial study, namely cytosolic, proteasomal, and ribosomal proteins, it became evident that proteasome inhibition during amino acid starvation had hardly any effect on protein abundance. The mean ratios of all three groups did reflect the mixing ratio of the experiment. The proteasomal inhibitor epoxomicin showed similar results (data not shown). On the other hand inhibition of autophagy by 3-MA led to an increase of proteins from all three localizations (Fig. 4E and supplemental Tables 5 and 6). The means of cytosolic and proteasomal protein abundance rose over 30%, and the mean of ribosomal protein abundance rose over 17%. Upon application of unpaired two-tail Student's t tests to the groups of quantified proteins, the means turned out to be significantly different (p < 0.001). Thus, as already mentioned in the preceding paragraph, it appears that proteasomal degradation during starvation conditions in MCF7 cells is less important than autophagosomal/lysosomal degradation.

As autophagy appeared to be the major degradation pathway during starvation conditions in MCF7 cells we wondered how the ordered nature of degradation we observed could be carried out by a process described as unspecific bulk degradation. One explanation could be that different types of autophagy are active at different time points during starvation. It has already been suggested that macroautophagy is only active during the initial starvation phase and that later CMA takes over (1). We recorded numerous profiles of proteins known to be CMA substrates, such as tRNA synthetases (4) and proteasomal proteins (39). However, in our case these proteins were rapidly degraded, indicating that CMA might play a role in the initial starvation period. To address the impact of CMA on the detected protein dynamics, we analyzed the occurrence of experimentally verified KFERQ-like motifs and the cluster distribution of the corresponding proteins. Substrates for CMA are reported to contain a pentapeptide motif related to the KFERQ sequence (4, 40). We did not find a correlation between protein profiles and the occurrence of KFERQ-like motifs. In fact, potential protein substrates of CMA localized to all six clusters (supplemental Table 7), indicating that CMA-dependent degradation cannot explain the observed cluster distribution of proteins. Thus, macroautophagy itself appears to be responsible for the ordered nature of protein breakdown.

Mitochondrial Degradation through Autophagy—

To further verify that autophagy is responsible for ordered organelle degradation, we followed the subcellular localization of mitochondria during amino acid starvation by microscopy. We compared cells amino acid-starved for 7 and 30 h with non-starved control cells (Fig. 5). Whereas we rarely detected mitochondrial-autophagosomal colocalization after 7 h of starvation (Fig. 5, D–F), we were able to detect extensive colocalization after 30 h (Fig. 5, G–I). The same holds true for mitochondrial-lysosomal colocalization (supplemental Fig. 3, A–I). These observations are in accordance with already obtained results, highlighting that autophagy is regulated at the protein complex/organelle level and that organelles are degraded in an ordered fashion via autophagy in the lysosome.

DISCUSSION

Amino acid starvation represents the classical stimulus to induce macroautophagy, which is still considered to be an unspecific bulk degradation pathway (41). However, this hypothesis has been questioned numerous times as the unspecific elimination of vital organelles could threaten cellular survival. Already in 1986 Kitagawa and Ono (42) noticed that autophagosomes in rat pancreatic exocrine cells contained mitochondria and ER only after long term starvation. In addition, a selectivity toward the substrate ALD6P has been shown in yeast (43). We have analyzed protein degradation dynamics by quantitative MS under autophagic conditions in the human MCF7 breast carcinoma cell line and were able to record protein profiles of almost 1500 proteins consisting of five time points each. Cluster analysis of these profiles revealed that degradation dynamics strongly depend on the subcellular localization of the corresponding proteins. Cytosolic proteins, multiprotein complexes, and organelles showed distinct degradation patterns. Thus, we were able to show by an unbiased proteome-wide analysis that protein degradation during starvation-induced autophagy is regulated at the protein complex/organelle level and happens in an ordered fashion (Fig. 6).

Quantitative MS has become a very powerful tool to follow spatiotemporal dynamics of endogenous proteins and of protein modifications in different experimental settings (44). In the current work, we used SILAC (21) in combination with treatment of cell populations to globally follow protein degradation during amino acid starvation at multiple time points. To our knowledge, this is the most comprehensive data set on protein dynamics during starvation conditions so far. Cytosolic and proteasomal proteins were degraded rapidly followed by ribosomal proteins. On the other hand mitochondrial, ER, spliceosome, and vacuolar proteins lagged greatly behind. One reason for delayed organellar protein degradation could be that cytosolic proteins are more easily accessible than organellar proteins, which are part of larger structures and in many cases are integrated into membranes. Another reason could be that the cell “chooses” to degrade organelles at a later time point as they are energetically more expensive to synthesize and important for cellular survival. The idea of a selective process is supported by the fact that we were able to detect only minor mitochondrial/autophagosomal colocalization during the early phase of starvation, whereas we saw extensive colocalization at a later time point.

One of the major cellular responses to amino acid starvation is the initiation of autophagy. Autophagy is an evolutionarily conserved process thought to be important for the turnover of whole organelles and long lived proteins. However, prolonged autophagy can lead to type II programmed cell death. Autophagy has been linked to several diseases, ranging from myopathies, heart and liver diseases, and neurodegeneration to cancer (9, 45). In all cases autophagy has been described to have beneficial as well as detrimental effects, depending among others on the disease state. This highlights the complexity of autophagy regulation and the need to further define underlying processes. Using confocal microscopy, we detected an induction of autophagy during starvation in MCF7 cells by eGFP-LC3 translocation into autophagosomes that reached a steady-state level after 6 h. To delineate the cellular pathways responsible for protein degradation during amino acid starvation, we inhibited the proteasome and autophagy by drugs. Whereas the proteasome has been shown to be important for cellular survival during an acute decrease in the external amino acid supply (46), the autophagosomal-lysosomal pathway appears to be the major degradative route during long term starvation; this is highlighted by the increase of lysosomal proteases, glycosidases, and lipases. Inhibition of the proteasome led only to minor changes in protein abundance. During autophagy inhibition we observed an increase of ribosomal, cytosolic, and proteasomal proteins, the latter two decreasing rapidly at the beginning of starvation. The strong increase during inhibited autophagy stresses its importance for the degradation of these protein groups.

As protein degradation happened primarily via the autophagosomal/lysosomal route we wanted to outline the influence of different autophagy subtypes and compared CMA with macroautophagy. CMA could not explain the observed cluster distribution of proteins during starvation. Thus, CMA seems to play only a minor role during amino acid starvation in MCF7 cells. However, we cannot exclude that CMA is responsible for the degradation dynamics of single proteins. The ordered nature of protein degradation observed during amino acid starvation could also be influenced by steric properties/hindrance of the respective substrates. Thus, smaller substrates could be degraded rapidly, whereas larger substrates would be delayed. We analyzed the size distribution of proteins in each cluster and did not observe major differences, although the largest substrates localized to clusters E and F. A likely explanation for this could be the general size difference between cytosolic and nuclear proteins. In general, the majority of proteins in each cluster had a length of 200–400 amino acid residues (supplemental Fig. 4). Organelle-specific autophagy types such as ER-phagy (18) and mitophagy (19) could influence the ordered nature of degradation. One could imagine that organelle-specific degradation is initiated at specific time points, which would yield the observed protein profiles. Further we would like to point out that we also detected distinct degradation patterns for different multiprotein complexes resident in the cytosol, such as the proteasome and the ribosome. In addition, protein groups such as tRNA synthetases also showed distinct degradation profiles. This all indicates that autophagy is much more tightly regulated as so far anticipated and that proteins are not arbitrarily degraded via this pathway. Tightly regulated macroautophagy or temporally regulated autophagy subtypes could be responsible for this phenomenon. Future work to delineate underlying mechanisms of substrate specificity of autophagy could help to solve this problem.

Taken together, we have generated a comprehensive data set of cellular protein dynamics during 36 h of amino acid starvation. In an unbiased approach we were able to highlight that proteins with different subcellular localizations show distinct degradation patterns and that degradation is mainly dependent on the autophagosomal/lysosomal pathway. Thus, starvation-induced autophagy seems to be selective and to degrade proteins in an ordered fashion and not in a completely unspecific bulk process. Our study raises the important questions of how the autophagosome decides what to take up and to deliver to the lysosome and how these choices change over time.