One of the most striking findings of this study is that caspase substrates as a whole tend to physically interact with one or more other caspase substrates, either in protein complexes or networks. We interpret this as an indication that caspases target a limited set of biological pathways to elicit programmed cell death, as opposed to indiscriminately targeting the entire cellular proteome. These data also suggest that caspases target protein complexes that are hubs for cell viability in essential processes such as transcription, and that targeting of multiple components in each complex is required for a full commitment to apoptosis. In this regard, it is notable that active caspases are dimeric, which is rare for proteases. A dimer is well equipped for semi-processive activity consistent with targeting multiple components of protein complexes. Another reported example of targeted proteolysis of a protein complex is the cleavage of SET, HMG-2, and Ape1, three components of the SET complex, by the cytotoxic lymphocyte protease granzyme A (Lieberman and Fan, 2003
). Interestingly, granzyme A is also dimeric.
The discovery that several components of the N-CoR/SMRT transcriptional corepressor complex are targets of caspase proteolysis presents a remarkable example of multiple cleavages in a single protein complex or pathway during apoptosis. Six proteins that are part of, or interact with, the N-CoR/SMRT complex are fully cleaved during etoposide-induced apoptosis in Jurkat cells, including the corepressors N-CoR and SMRT themselves. This finding was made possible by our large-scale discovery-oriented proteomic approach, as opposed to a more typical focused hypothesis-driven approach. Inactivation of the N-CoR/SMRT complex during apoptosis may achieve a result similar to the effect of HDAC inhibitors, with decreased histone deacetylation leading to transcriptional upregulation of proapoptotic genes (Bolden et al., 2006
). Interestingly, HDAC 7 has recently been implicated as a physiological substrate of caspase-8, with its proteolytic inactivation leading to upregulation of Nur77 (Scott et al., 2008
Our studies indicate that a change in function of proteins targeted by caspases during apoptosis must be rationalized by one or occasionally a few cuts per protein. We have found that caspase cleavages occur inside functional domains and between functional domains at approximately equal frequencies. In either case, relatively stable products must be produced following cleavage of the substrates since we detected them. Stability of these products is also consistent with the strict P1′ glycine, serine, and alanine specificity we observe for the cellular caspase-like activity, which creates fragments conforming to the N-end rule (Varshavsky, 1992
). In addition to functional disruption of the substrate protein, such cleavages may result in products that function as dominant negatives. For example, in the case of the N-CoR and SMRT corepressors, the C-terminal cleavage products contain the CoRNR boxes known to interact with nuclear receptors (Hu and Lazar, 1999
). These proteolysis products could thus inhibit interaction between N-CoR/SMRT and nuclear receptors.
By globally identifying caspase-like cleavage sites in the proteome of apoptotic cells, this work presents a large-scale substrate specificity profile of caspase processing of endogenous proteins in intact cells. Importantly, this profile is influenced not only by the primary structure of cleavage sites, but also solvent accessibility, secondary and higher order protein structure, and possibly post-translational modifications of substrates (Tözsér et al., 2003
). Our finding that caspases often target proteins in complexes underscores the value of studying determinants of proteolysis under physiologically relevant conditions. Although the aggregate substrate specificity of the caspase-like activity observed during etoposide-induced apoptosis in Jurkat cells is most similar to the known substrate specificity of executioner caspases, substrate specificity studies of caspases using peptides do not fully account for the observed cellular specificity (Schilling and Overall, 2008
; Stennicke et al., 2000
; Thornberry et al., 1997
). Peptide-centric approaches are best suited for determination of optimal protease substrate specificity, invaluable in development of sensitive synthetic substrates or potent inhibitors. In contrast, a protein-centric methods such as the one presented here is best suited for characterization of endogenous proteolysis in biological samples.
This work indicates that the widely used primary structural determinants of caspase in vitro
substrate specificity are insufficient to predict physiological caspase cleavage sites. However, the cellular cleavage sites we have identified experimentally double the number of annotated caspase cleavage sites, significantly expanding a dataset that can be used to train algorithms for predicting cleavage sites. Indeed, a proof of principle is provided by an accurate prediction of caspase cleavage sites by our preliminary HMMs. In addition to demonstrating that caspase cleavage sites are most commonly found in solvent accessible loop regions, as shown for other proteases (Hubbard et al., 1991
), our analysis also indicates that a number of cleavage sites appear in partially solvent inaccessible regions and α-helices. This information could also be incorporated into predictive algorithms. Finally, based on our protein interaction analysis, predictive algorithms may also benefit from scoring that considers physical interactions of candidate substrates with other caspase substrates.
Common approaches for the study of proteolysis in complex mixtures employ gel electrophoresis and mass spectrometry for analysis of proteins in cells, typically identifying tens of protein substrates at a time (Machuy et al., 2005
). These approaches do not usually identify specific cleavage sites. In contrast, modern proteomic methods using positive enrichment of phosphorylated, glycosylated, or ubiquitinated polypeptides can lead to the identification of hundreds or thousands of post-translationally modified sites on proteins (Collins et al., 2007
; Peng et al., 2003
; Vosseller et al., 2006
). However, selective capture of the products of proteolysis is not facile. Gevaert et al. and McDonald et al. have reported methods for negative selection of N-terminal peptides, while Timmer et al. have reported an approach for positive selection of N-terminal peptides (Gevaert et al., 2003
; McDonald et al., 2005
; Timmer et al., 2007
). These chemical approaches require two consecutive and quasi-orthogonal derivatization steps, the first to block lysine ε-amines and the second to label terminal α-amines. We believe the success of our method is based on the advantage of achieving great selectivity for α-amines in a single labeling step through use of the enzyme subtiligase.
The incomplete overlap between cleavage sites and protein substrates identified in our separate experiments is not uncommon for tandem mass spectrometric analysis of complex mixtures, in which analysis of many species, whether peptidic or not, precludes complete sampling (Elias et al., 2005
). The number of caspase substrates we have identified is thus likely smaller than the total number of caspase substrates in apoptotic Jurkat cells. We identified 50 of approximately 361 previously reported human caspase substrates (Supplemental Figure 3C
), 48 of approximately 227 previously reported caspase substrates for which cleavage sites are known (Supplemental Figure 3D
), and 50 of approximately 307 previously reported human caspase cleavage sites (Supplemental Figure 3E
) (Lüthi and Martin, 2007
). In addition to incomplete proteomic sampling, three additional factors likely account for the modest overlap with previously identified substrates.
First, only cleavage sites corresponding to N-terminal semi-tryptic peptides 7 to 40 residues in length (without the additional SY dipeptide label) are generally identified in our studies. With collision-induced dissociation (CID), fragmentation of peptides below this range generally does not provide enough information for unambiguous matching to databases, and most peptides above this range do not fragment efficiently. Approximately half of the 307 previously reported human caspase cleavage sites result in N-terminal semi-tryptic peptides that fall outside this range and would not be identified using our current method (Supplemental Figure 15
). Second, each analytical method, whether global or focused, will have its own associated biases and limitations. For example, a limitation of the subtiligase labeling method is that the enzyme cannot access protein N-termini that are buried or occluded. Third, we employed a single apoptotic inducer in a single cell line, using a single analysis method, whereas the previously reported set of substrates comes from a multitude of studies using varied inducers, cell types, and methods.
Although the dataset of substrates we have identified is not comprehensive, it doubles the number of known cleavage sites in human targets of caspase-like proteolysis in apoptosis. The study of apoptotic pathways has important ramifications for identification of pathways that are critical for cellular homeostasis, and for development of potential anti-cancer therapeutics. A number of caspase targets are active or established drug targets for treating cancer, including topoisomerase II, Bcl-2, Hdm2, MEK1, and Akt, to name a few. Thus, it is possible that the list of substrates we have identified includes new candidate chemotherapeutic targets. The products of caspase proteolysis may also serve as useful biomarkers for assessment of chemotherapeutic efficacy, as demonstrated in the case of cytokeratin-18 for breast cancer (Olofsson et al., 2007
). Along with MS-based quantitation, the technology we describe should enable global analysis of the apoptotic phenotype as a function of time, cellular context, and type of induction. Finally, the technology should also be broadly applicable for global sequencing of proteolytic cleavage sites in other biological settings.