Proteome Turnover in the Spotlight: Approaches, Applications & Perspectives

Affiliations: 1: Department of Biological Sciences, Columbia University, New York, NY 10027, USA 2: Proteomics; Max Planck Institute of Biophysics, Max von Laue Strasse 3, 60438 Frankfurt am Main, Germany 3: Proteomics; Max Planck Institute for Brain Research, Max von Laue Strasse 4, 60438 Frankfurt am Main, Germany * Corresponding authors: Julian David Langer – julian.langer@biophys.mpg.de Marko Jovanovic – mj2794@columbia.edu Mol Cell Proteomics Papers in Press. Published on November 30, 2020 as Manuscript R120.002190

Proteostatic mechanisms are some of the cell's most essential processes, as they ensure that functional proteins are maintained at their correct concentrations and in the proper locations needed for cellular activities to proceed (8)(9)(10). These processes also ensure that misfolded, aged, or damaged proteins are removed from the cellular protein pool as needed (11). Accordingly, disruption of proteostasis contributes to the pathophysiology of a variety of disease states, most notably neurodegenerative disorders and cancer (11).
Probing the kinetics of proteome-wide protein turnover lends insight into how cells perform crucial functions such as differentiation and stress response in both normal and disease contexts, and can illuminate the guiding principles that underlie the regulation of protein turnover across protein families, cell types, and species.
Protein turnover is monitored and regulated by several cellular surveillance systems. While protein production includes all of the processes that precede mRNA translation, including RNA transcription, maturation, and processing, in this review we will focus on the time frame between protein synthesis and degradation. mRNA translation is controlled by regulatory motifs in mRNA nucleotide sequences; these sequences are bound by RNA binding proteins (RBPs) and small RNA guides (such as microRNAs) to modulate their expression (12)(13)(14)(15)(16). Molecular chaperones, insertases, and translocases control maturation of nascent polypeptide chains, and post-translational modifications are added to proteins in the secretory pathway or through signaling cascades (17)(18)(19).
7 other hand, is a first-order process whose rate corresponds to the fractional removal of proteins from an existing pool in the cell. The degradation rate constant kdeg is therefore quantified with the dimension of time only (1/time). As the amount of protein decreases at a rate (kdeg) proportional to its current value, the amount of protein lost follows an exponential decay function. As such, the amount of protein produced over a certain time window depends only on the integration of its synthesis rate constant, ksyn: (1) dP/dt (syn) = ksyn while the amount of protein lost over that timeframe depends on the existing protein pool multiplied by its degradation rate constant, kdeg (24): (2) dP/dt (deg) = [P] * kdeg or if expressed as an exponential decay function: (3) P(t) = P0 * e -Kdeg*t Equations 1 and 2 together define the total change in protein amount over time: (4) dP/dt (total) = ksyn -[P] * kdeg Turnover rates in steady state vs. dynamic systems In this review, we distinguish between cells at steady state and those undergoing a dynamic change. At steady state, we can make two basic assumptions: (1) for all proteins, the net change in protein levels is zero, which means that (2) the number of protein molecules produced is equal to the number of proteins lost. The turnover rate of a protein is often defined as the time needed to both degrade and re-synthesize half the proteins present in a specific cellular state. At steady state, however, due to the equivalency described in equation 6, the turnover rate is simply equal to the time it takes to remove half of the existing protein pool -and as such, relative turnover rates can simply be expressed through the degradation rate constant kdeg.
During dynamic processes, protein levels often change over time. Protein level changes may be due to changes in protein production rates, protein degradation rates, or both ( Figure 1C). When a new steady state is reached after the perturbation, a protein may be expressed at a very different abundance than before, but its turnover rate will only differ if its degradation rate constant, kdeg, and therefore its half-life, has changed. In other words, based on the definitions above, protein synthesis changes alone will not affect a protein's turnover rate, but only its abundance ( Figure 1C).
Modeling true changes in turnover rates during dynamic processes requires considerably more mathematical manipulation than modeling turnover rates at steadystate (25) -in fact, achieving an accurate model of dynamic turnover rate changes remains an open challenge in the field. So far, dynamic changes in turnover rate constants have only been approximated using linear rate change assumptions, which do not likely fully represent the true physiological behavior of dynamically adjusting proteomes (25).
However, the above steady-state assumptions can and have been used effectively to by guest on January 10, 2021 compare relative end-point synthesis and degradation rates between conditions as we will describe below.
"Old" proteins and non-exponential decay Different methods described in the text below each confer particular advantages and disadvantages for tracking protein degradation and synthesis rates at steady-state (Supplemental Table 1). Modeling turnover rates according to the definitions at steady state described above is relatively straightforward, with one notable exception. We assume that the first order process of protein loss is stochastic and all proteins from the same species have the same probability of getting degraded. Under this assumption, a newly synthesized protein has the same probability of being degraded as a preexisting protein of the same species that has been around for a long time (24) and as such protein loss follows an exponential decay function as described above. This also explains why the synthesis signal from metabolic labels appears logarithmic rather than linear, despite protein synthesis following zero-order kinetics as described by equation 1 above ( Figure   2). However, recent studies have demonstrated that the assumption of exponential decay does not hold true for all cellular protein populations. For certain subsets of proteins, the probability that any given protein molecule is degraded can change as a function of its molecular age, with newly synthesized proteins being typically less stable than "older" proteins. The loss of these proteins follows a pattern of non-exponential behavior (see below text for more details) (26)(27)(28)(29).

Cell division and protein turnover
As outlined above, the critical protein turnover parameter, kdeg, corresponds to the time it takes for a cell's preexisting protein pool to be reduced by half. This is certainly by guest on January 10, 2021 https://www.mcponline.org Downloaded from true for non-dividing cells, but for dividing cells, the preexisting protein pool will be reduced to half with every cell division even without active protein degradation. In dividing cells, therefore, the reduction of a preexisting protein pool occurs due to a combination of dilution due to cell division and true protein degradation. In such a system the cell division rate has to be taken into account and should be included as its own term, kdil. The total rate of protein loss, kloss, measured in such a system is defined by the following equation: Consequently, kdeg can be determined by measuring kloss and the division rate, kdil, of the studied system. Taking the division rate into account is extremely important, not just for comparing systems for which cell division rates vary greatly, but for accurately detecting the kdeg of proteins that turn over slowly, which can be confounded by kdil (24). More detail about the non-trivial relationship between protein synthesis, degradation rate, and cell division rate are discussed in depth in a recent publication by the Busse group, who emphasize that taking the cell division rate into account significantly improves the "hit rate" of differential gene expression profiling (30).

Radioactivity, drugs, and fluorescent lightseven before the 70s
In the first protein turnover studies more than 80 years ago, 15 N-isotope-labeled amino acids were fed to mice in order to analyze protein synthesis and degradation, with detection based on mass spectrometry (2)(3)(4)(5)(6). These groundbreaking studies showed that cellular proteins are not static, but rather are in constant flux of production and loss. In the following decades, the most commonly used reporters were amino acids with radioactive isotopes of carbon, hydrogen, or sulfur, with subsequent detection in proteins by guest on January 10, 2021 using scintillation counting (31). Radioactive decay-based detection of synthesis and degradation of specific proteins enabled direct analyses of their half-lives, particularly in combination with antibody-based purification of the target proteins (32). Later on, the combination of autoradiography with 1D and 2D gels allowed for more comprehensive differential turnover studies, as dozens of different proteins can be separated on such gels (33,34). However, these analyses were limited by the relatively high radioactivity doses required (and their associated effects on cell and animal health), the significant protein loss during sample preparation, and the challenging identification of candidate proteins displaying differential synthesis or decay kinetics (35).
Subsequently, the application of biochemical tools, such as small molecule inhibitors of synthesis (e.g. cycloheximide, puromycin) or degradation (e.g. MG-132, bortezomib, lactacystin), alongside the invention of genetically engineered proteins, enabled new insights into proteome stability and turnover. Fusion proteins tagged with constructs such as GFP or the TAP-tag allowed for comprehensive, highly-multiplexed studies with specific and sensitive detection of protein synthesis and degradation; with these tags, no introduction of tracer amino acids was required (36)(37)(38). For example, a systematic study in yeast, done in the year 2006, reported degradation rates for more than 3750 TAP-tagged proteins after inhibition of protein synthesis by cycloheximide, with detection of protein loss over time determined by immunoblotting (36). Two years later, the stability of more than 8000 proteins was profiled in HEK293T cells using a combination of GFP-tagging, flow cytometry and microarrays (37). These studies all required the construction of thousands of reporter-tagged strain or cell line collections -a mammoth task to accomplish in both resources and manpower.
Technical advances in mass spectrometry-based proteomics then allowed for more generalized protein tracing, without the need for tagged constructs. In combination with translation inhibition, shotgun proteomics facilitated the tracking of protein degradation rates. This approach has been applied alongside subcellular fractionation and proteasome inhibition to quantify the differences in subcellular proteome turnover and match degradation pathways to each cellular compartment (39).
While all of these tools provided valuable new insights, they incur considerable limitations (Supplemental Table 1). First, the inhibition of protein synthesis or degradation (e.g. by drugs) may lead to compensatory and off-target effects that can make determining physiological turnover rates challenging (40). Second, protein tags can potentially compromise physiological protein function and half-life. This is particularly critical for small proteins and membrane proteins (41)(42)(43). Third, the construction of large libraries of tagged protein constructs is both time-and resource-intensive. Nevertheless, these approaches are still widely used, particularly in targeted studies exclusively examining either protein synthesis or degradation. They are also still commonly used for specific and sensitive detection of a particular protein of interest with techniques such as Western blotting or immunofluorescent imaging to determine subcellular localization.

Dynamic SILAC approaches
In the early 21 st century, the use of non-radioactive isotopes in combination with mass spectrometry became popular in proteome turnover studies. The combination of high-resolution liquid chromatography, nano-electrospray ionization, and ultra-highresolution tandem mass spectrometry with fast MS/MS cycles enabled the quantitative analysis of thousands of peptides and proteins in a few hours. The required tracer isotopes were initially introduced via carbon sources (44), in which heavy isotopes were incorporated into proteins by sugar/carbon metabolism. However, the incorporation of heavy isotopes via metabolic pathways normally lead to a variation in the degree to which heavy labeled amino acids were incorporated into proteins, such that full labeling was often not achieved. This made the separation of overlapping isotopic envelopes, and therefore quantification of the differently labeled peptides, very challenging (1,(44)(45)(46).
This was largely overcome by the addition of amino acids with a defined and selectable number of stable isotopes into culture media or food sources, allowing for comprehensive and systematic proteome turnover studies in a variety of organisms (47).
These "heavy" amino acids were initially used for quantitative studies as part of Stable Isotope Labeling by Amino acids in Cell culture (SILAC), in which proteome abundance differences in unlabeled and fully-labeled samples are compared (48,49). Since specific amino acids -normally lysine and/or arginine -were labeled with a fixed mass in SILAC, the isotopic envelopes of "light" and "heavy" peaks are separated by pre-defined shifts (e.g. 13 C6-Lysine or 13 C6 15 N4-Arginine), greatly facilitating data acquisition and interpretation. Coupling lysine/arginine labeling with trypsin as a protease for sample preparation guarantees that each peptide has a labeled amino acid.
This quantitative approach was then re-purposed to study proteome turnover by making use of "pulsed-only" experiments. In these so-called dynamic SILAC experiments, cells are switched from unlabeled medium to a medium containing isotopically-labeled amino acids, still typically heavy lysine and/or arginine (50). Samples are then measured via LC-MS/MS over a time course. The rate at which a heavy amino acid-labeled peptide by guest on January 10, 2021 signal appears corresponds mainly to that peptide's rate of synthesis, while the rate at which a light-amino-acid-containing peptide decreases represents its rate of degradation.
The ratio of heavy to light peptide signal thus directly reflects protein turnover ( Figure 2).
It should be noted here that dynamic SILAC and a similar term -pulsed SILAC (pSILAC) -are often used interchangeably in the literature. However, while both approaches are "pulsed-only" experiments with a similar setup, dynamic SILAC actually refers to experiments that determine proteome-wide protein turnover rate using only two SILAC channels, "light" and "heavy." pSILAC, on the other hand, originally referred to a labelling approach in which two "light" cell populations are pulse labelled with either "medium-heavy" or "heavy" amino acids to quantify relative differences in de novo protein synthesis. The term was first coined by Selbach et al., who used pSILAC to assess the impact of microRNAs on protein synthesis (51), and subsequently described in more detail by Schwanhäusser et al., 2009 (52). Although the terms are now often used interchangeably, we will honor their original definition and refer to dynamic SILAC in all studies that measured protein turnover (the majority of studies described here), and use pSILAC only in the manner that it was originally intended -to measure relative differences in de novo protein synthesis.
While classical dynamic SILAC experiments have many advantages, they do have some limitations (Supplemental Table 1). It should be noted that due to the re-use of existing, light amino acids, the true synthesis rate may be higher than that measured by the increase in the heavy-labeled peak. This "recycling issue" can be addressed by monitoring peptides with missed cleavage sites and correcting for their uptake of light amino acids (53). Additionally, because each sample over a time course must be by guest on January 10, 2021 harvested separately, it is difficult to make absolute comparisons of either the heavy or light isotope peak intensities over time due to experimental differences in sample preparation and data acquisition. In general, dynamic SILAC data yields protein turnover information (half-lives at steady state -see "definition of terms" section), but does not allow clear separation of the contribution of synthesis and degradation to the measured turnover rate without further internal standards or quantification strategies (see below, dividing NIH 3T3 mouse fibroblast cells. They found that mRNA levels and mRNA translation rates contributed the most to the final protein levels, while mRNA and protein stability only had a minor global effect. Moreover, they found that mRNA and protein turnover rates themselves showed no correlation to one another. This study illustrates one of the most straightforward applications of dynamic SILAC-based technology, which is to probe proteome-wide turnover rates, then match these rates with bioinformatic by guest on January 10,2021 analysis to other parameters (e.g. primary sequence, motifs, mRNA turnover, etc.) in order to identify the generalizable parameters that determine turnover rates under the measured conditions (53) ( Figure 3A). More recently, Martin-Perez et al. (2017) also reported such a study, in which they measured total proteome turnover in exponentially growing yeast and determined which parameters had the most influence on protein turnover rates (56). They determined that proteome turnover depended upon functional characteristics such as subcellular localization, membership of a protein complex, and GO process more than it did on sequence-intrinsic or biochemical features and expression levels. Surprisingly, in contrast to the above-mentioned study by the Selbach lab, they discovered a strong relationship between mRNA turnover and protein turnover rates (56). With the cell-type specificity of protein half-lives established, dynamic SILACbased approaches can furthermore elucidate how proteome turnover relates to the ways that specific cell types perform their higher-order, specialized functions. For example, studies measuring proteome turnover rates in cultured neurons have identified mechanisms of proteostasis with direct implications for neurobiological processes such as memory formation and aging. The Ziv lab used dynamic SILAC to explore synaptic processes by measuring the half-lives of synaptic proteins and the influence of proteostasis on metabolic load. They found that protein turnover rates were not significantly different for pre-and post-synaptic proteins, nor for proteins whose corresponding mRNAs have been found to localize to dendrites (63). Later work in neuronal cultures sought to determine which proteins are degraded by the UPS by identifying those whose half-lives increased upon proteasome inhibition. While some by guest on January 10, 2021 proteins, including those related to glutamate receptor trafficking, were slowed by UPS inhibition, most synaptic proteins were not affected, indicating that they may be degraded by alternate pathways. They also found that inhibition of the proteasome also led to a profound blockage in the synthesis of a large number of synaptic proteins, indicating that there may be crosstalk between protein production and degradation pathways (64).
Additional studies in neurons found that not only does cell type influence proteome turnover, but cellular microenvironment does as well. Dörrbaum et al. (2018) used dynamic SILAC to profile rat hippocampal neurons in neuron-enriched and glia-enriched cultures, and found that proteins from glia cells had shorter half-lives than the same proteins in neurons (65). Moreover, they found that the presence of glia in co-culture changed the turnover of proteins in neurons. This indicates that not only is cell identity an important determinant of protein turnover, but cell-cell signaling influences turnover rates as well.
A more recent study measured proteome-wide turnover using dynamic SILAC in both naïve and memory T-cells. This study revealed that despite the quiescent state of naïve T cells, not all cellular states are inert and that they contain a subset of highly turned over proteins, such as certain key transcription factors that both help maintain the quiescent state of naïve T cells and facilitate a rapid transition into an activated state through their rapid depletion after stimulation. Additionally, the authors found that, despite not being at all dependent on glycolysis, naïve T cells maintain high levels of glycolytic enzymes with very slow turnover rates, which allows naïve T cells to jumpstart glycolysis upon activation. With this data, the authors elucidated mechanisms by which the turnover by guest on January 10, 2021 rates of certain proteins is optimized in naïve T cells to prime them to efficiently exit quiescence after activation and maintain their new cell identity (66).
Proteome turnover analysis can be used to study disease states such as cancer and their potential treatments (67)(68)(69)(70)(71)(72)(73). One such study published by the Wiita lab examined proteome turnover in MM1.S multiple myeloma cells (72). They did not apply a global dynamic SILAC approach, but rather targeted proteomics (selected reaction monitoring (SRM)) coupled to dynamic SILAC to acquire high accuracy turnover data for 272 selected proteins. Due to the high accuracy of the SRM measurements, the authors provided quantitative data for the heavy and light channel over time separately and therefore estimate protein production and degradation separately. They compared the dynamic SILAC-determined protein synthesis data with ribosome footprinting data.
Ribosome footprinting is a next generation sequencing based method to estimate the ribosome density on any given mRNA, and therefore is considered a good proxy for protein production (74)(75)(76) Lastly, dynamic SILAC can be used in whole animal studies to elucidate in vivo proteome turnover rates by feeding animals isotopically labeled amino acids. This approach has been successfully applied to model organisms like C. elegans (77,78), be simply explained by differences in cell division rates, and give important insight about condition and cellular specific protein turnover (85).
An isotopic labeling strategy has recently been applied to study protein turnover in human ventricular cerebrospinal fluid (CSF) from patients who had suffered a subarachnoid hemorrhage (86). The protocol was based on isotopically labeled leucine, a method dubbed whole proteome stable isotope labeling kinetics (wpSILK), rather than the arginine and lysine-based dynamic SILAC approach, but the experimental principle is the same. Lehmann et al. detected proteins from multiple cell types, including neurons and immune cells, and found a link between protein turnover rates and cell of origin, again echoing results from previous in vitro investigations (86). This approach could potentially be applied for biomarker discovery and indicates a potential crossing over of dynamic SILAC-like approaches from biology to medicine.
Combinatorial approaches to determine quantitative, differential proteome turnover data Ideally, quantitative analyses of proteome turnover and protein half-lives include data on both protein synthesis and degradation rates separately. In conventional dynamic SILAC experiments, it is challenging to separate data for these two processes due to experimental variabilities in sample preparation and LC-MS/MS data acquisition (54,55).
Several different approaches have been implemented to specifically enable the determination of separate synthesis and degradation rates. One approach relies on the addition of an internal standard to the dynamic SILAC experiment (Figure 2), a method first used by the Lamond group in 2012 (54). This additional channel confers three by guest on January 10, 2021 advantages: it can be used for normalization, it provides information about protein abundance, and most importantly, it can be used to track protein synthesis and degradation separately (Figure 2). Through this strategy, one can determine if protein and also turnover rate changes are due to changes in protein synthesis rates, protein degradation rates, or the combined effect of both. Therefore, such an extended dynamic SILAC approach is best suited to characterize dynamic changes in these rates. They found that while LPS-induced protein production changes were primarily driven by transcriptional changes, proteome remodeling of pre-existing proteins, often so-called housekeeping genes, occurred at the level of mRNA translation and protein degradation (25). The same approach was used recently to examine proteome turnover changes during synaptic scaling, a type of homeostatic plasticity, in primary neurons (88). In this study, over half of the synaptic proteins in both pre-and post-synapses showed changes in their turnover rates in different forms of synaptic plasticity, using different mechanisms to adjust turnover in up-and down-scaling experiments.
by guest on January 10, 2021 The second strategy to measure protein production and degradation separately and track their changes upon perturbation is to combine dynamic SILAC with isobaric labeling ( Figure 2) (25,63,65,83). It is important to note that, in this context, the dynamic range of the mass spectrometer used for data acquisition provides a hard limit on the shortest timepoint that protein synthesis or degradation will be detectable. Poor ionization efficiency of a target peptide and co-elution of highly-abundant peptides may further confound detection (Supplemental Table 1). It is thus important to always use appropriate control experiments and data analysis validation steps. A good starting point is to perform a mock by guest on January 10, 2021 dynamic SILAC incubation, i.e. to treat an unlabeled sample as a dynamic SILAC sample in data analysis; in particular for short time points, this control allows for the identification of false-positive and background signal levels (94). Conversely, the use of a fully heavylabeled sample as a booster signal in combined dynamic SILAC-TMT protocols can increase detection of nascent peptides at early time points (94). This booster channel significantly increases the chance that heavy isotope MS1 peaks are selected for fragmentation (55,89,94), which subsequently increases the likelihood that lowabundance, newly synthesized, heavy-labeled peptides are quantified (94). Finally, reversed-SILAC channel experiments -in which heavy SILAC-labeled cells are pulsed with light SILAC amino acids as an additional experimental replicate -can help exclude signals inferred by confounding parameters such as isotopic envelope overlap (90).
A further consideration for dynamic SILAC experiments is that many primary cells require specific media and display significant sensitivity to medium changes -for example, neurons, which require conditioned media (Supplemental Table 1). This poses a further challenge for analyses of short time points in these systems, as it would be impossible to tell if protein production rate changes at early time points are also affected or induced by the media change. Some groups bypass this issue by adding heavy amino acids in excess to the preconditioned media at time point zero (63), or make use of preconditioned media that was already generated from a culture grown in heavy amino acids and therefore minimizes the adverse effects of the medium change on the cells (65,88). These issues become even more challenging in whole animal studies, where heavy amino acids are injected into or ingested by the animals. In contrast to cell culture systems, the in vivo system does not get "flushed" by the number of heavy isotopes, therefore heavy isotope by guest on January 10, 2021 incorporation will be slower. Moreover, the heavy isotopes may not enter all of the cells at equal rates, causing noisiness in early time-point signals (83). It is important to correct for these biases by carefully monitoring heavy isotope incorporation (84).

Artificial amino acids using affinity purification
The difficulties detecting low-abundant, nascent proteins using pulse-only SILAC experiments are particularly salient in post-mitotic cells. To overcome these challenges, the Schuman and Tirell labs developed "Bio-Orthogonal-Non-Canonical-Amino-acid Tagging" (BONCAT) (95). BONCAT makes use of natural amino acid surrogates, typically methionine mimetics, that can be chemically targeted for purification. They usually carry an azido-or alkyne functional group and can thus be immobilized on a solid phase using click chemistry and affinity purification. While initially developed for neurobiological applications (95)(96)(97)(98), the technique has since been applied to multiple systems, including primary cells (99), tissue sections (98), and in vivo in a variety of organisms, including bacteria (100), archaea (101) , plants (102), zebrafish (97), and other higher eukaryotes (103,104). BONCAT can also be used to visualize overall proteome synthesis in cells using fluorescent tags (FUNCAT) (105), or to measure synthesis of target proteins in a spatially resolved manner using a proximity ligation assay (106).
Recently, artificial amino acid incorporation was genetically targeted to specific cell types using modified tRNA-synthetases (107) in living mice (108,109), Drosophila (110) and zebrafish (111). This method enables specific analysis and imaging of the synthesis of a cell-type-specific proteome in its physiological environment without prior cellular isolation.
by guest on January 10, 2021 The main challenge in click-chemistry-based strategies is the biochemical purification of the labeled proteomes, as often only small fractions of the experimental sample are labeled, and background adsorption to the affinity resin can be substantial (Supplemental Table 1). This is particularly true for hydrophobic tissues such as brain lysates. These issues have been addressed through several different strategies, including covalent immobilization of nascent proteomes to enable stringent washing (99), or the use of cleavable crosslinkers to allow specific elution of labeled proteins (108,109,112,113). An essential step to "quality-test" a workflow for nascent or cell type-specific proteome analysis is to perform a control experiment using methionine instead of AHA or ANL and determine the experimental background proteome.
A logical extension of BONCAT was to combine it with a pSILAC approach. This approach improved the accuracy of nascent proteome analyses by the incorporation of heavy isotope-labeled amino acids in cell culture (114), macrophages (115,116) and T cells (117). This combinatorial approach also improved the signal to noise ratio as the background will be dominated by peptides produced before the isotope pulse and can now be easily distinguished from the BONCAT labeled proteins, which have to have the heavy isotope incorporated. The combined pSILAC and BONCAT labeling was also recently combined with TMT labeling to also provide the advantage of sample multiplexing (118).
While the above-mentioned studies used AHA labeling to look at production differences at shorter time windows or in specific cell types, this approach can also be reversed to study protein degradation, as demonstrated by the Selbach group (29). In this study, they applied a one hour AHA pulse followed by a cold methionine chase to NIH 3T3 cells for several different time periods. By combining this pulse-chase AHA experiment with SILAC labeling of samples with varying methionine chase lengths, they were able to precisely measure protein loss. This revealed that for ~15% of proteins, protein degradation does not follow the predicted exponential decay function, but rather undergoes a two-state model in which newly synthesized proteins are more likely to be degraded than older proteins (29). This process, dubbed "non-exponential degradation" (NED; see also above in the "definition of terms" section), was found to be more common in subunits of complexes produced in superstoichiometric amounts ( Figure 3B).

Integrative multi-omics approaches
Above, we laid out tracer-based approaches that enable direct measurement of protein turnover parameters under steady state and dynamic conditions. A potential alternative are multi-omics approaches using RNA levels, protein levels and, optionally, ribosome density as measured through ribosome footprinting, which can be analyzed together to estimate protein synthesis and degradation rates. Due to the lack of absolute protein estimates in steady-state proteomics measurements, these approaches work best on dynamic expression data, where relative changes can be very precisely measured on both RNA and protein levels (119)(120)(121)(122).
A good example of this type of integrative analysis is from Peshkin et al. (2015), who looked at the mRNA-to-protein relationship during Xenopus embryonic development (120). The authors could model protein synthesis and degradation by mass action kinetics, and found that there are two major behavioral classes of proteins in the early embryo: one group with relatively stable expression levels, which were primarily inherited by guest on January 10, 2021 from the maternal cell, and a second group produced by the zygote that displayed greater "dynamicity" but lower abundance, and had strong correlations with mRNA level changes.
This indicates that proteome changes in early Xenopus development are primarily driven by changes in mRNA (120).
Another example of a multi-omics approach is from Eisenberg et al. (2018), who used matched RNA-sequencing, ribosome profiling, and TMT-based proteomics to look at the temporal changes in gene expression during yeast meiosis (123). Here, ribosome footprinting was used as a proxy for protein production instead of direct metabolic labeling of the proteins themselves. As reported previously in other systems (124)(125)(126)(127), the authors found that members of the same protein complex showed stronger correlations with one another at the ribosome footprinting level than at the RNA level. However, by comparing the quantitative protein measurements with the ribosome-footprint-based protein production proxies, they found that changes in protein levels of protein complex members matched one another significantly more closely than ribosome footprinting changes. Taken together, this implies that, although members of protein complexes can be synthesized at ideal stoichiometry (124,125,127), often they are synthesized at imprecise stoichiometry and their levels are adjusted by protein degradation (123). These results are very much in line with the above mentioned study by the Selbach group, where a subset of newly synthesized proteins were found to have significantly shorter half-lives than older proteins, leading to the conclusion that these proteins are members of protein complexes, are synthesized superstoichiometrically, and that excess proteins not incorporated in the protein complex are rapidly degraded (29).
by guest on January 10, 2021 The integration of multi-omics data has long been a major challenge, due to differences in the instruments used to capture the data and the format in which it is generated. However, in the past few years, multiple computational tools have been developed that are relatively easy to implement and allow integration of such multi-omics time course data, estimating key regulatory parameter changes such as changes in protein synthesis and degradation (128)(129)(130)(131). One such ensemble of programs, Protein Expression Control Analysis (PECA) plus (129), can be used as a plugin in the popular "point and click" statistical software Perseus (132). Its various iterations can be used to calculate the probability for changes in mRNA or protein-level regulatory parameters at each time point in matched, large-scale time course data. Specifically, PECA-pS can determine synthesis and degradation rates from dynamic SILAC data, while PECA core can be used to identify change points for protein-level expression and degradation using matched RNA and protein expression data (129). Programs such as PECA now also provide labs with limited computational experience the means to gain considerable regulatory insight from their gene expression data.

Conclusion
Soon after the publication of the first dynamic SILAC publications, a review by Hinkson and Elias (2011) outlined open questions in the field of proteome turnover (1) .
Among these were (1) cataloging the differences in protein turnover rates after activation, between cell types, and across species, (2) understanding to what extent functionally and physically associated proteins are turned over in accordance with one another, and (3) matching proteins to degradation pathways (1). Although several of these questions have by guest on January 10, 2021 been addressed in selected model systems as outlined above, a comprehensive survey of biological systems is yet to be achieved. However, a few patterns regarding proteome turnover have emerged from the data that already exists.
One of the most robust findings across all of the studies surveyed here is that protein half-lives are similar for proteins found in the same complexes, and that proteins known to participate in protein complexes have longer half-lives than those with no known association partners (54,56,61,83,85,89). Complex turnover is not entirely coherent, but typically sub-clustered based on the architecture of multimeric complexes, with more dynamic subunits showing higher turnover than stabilizing "core" subunits, such as has been seen for the proteasome (50, 61, 65, 123) ( Figure 3B). Additional mechanistic studies have revealed that for some complexes, certain subunits may be translated in excess, then degraded down to stoichiometric equivalencies (29,123). The fact that members of large, multimembered protein complexes are more likely to have different turnover rates depending on subcellular location (54,62) implies that these excesses of protein complex subunits may be generated to favor complex formation in one location distinct from the location where the complex functions (54). More generally, the relationship between complex membership and proteome turnover suggests the possibility of coordinated biosynthesis and degradation mechanisms for groups of interacting proteins (56,63). These findings were discovered using different conditions and model systems, validating one another's conclusions and suggesting that the tight coupling of protein turnover to complex membership is a basic feature of biological systems. Future studies could incorporate the use of size exclusion chromatography (133)(134)(135)(136) with dynamic SILAC in order to better understand how proteins that are a part by guest on January 10, 2021 of several distinct complexes and/or are involved in "moonlighting" functions display variation in their turnover rates (137,138). Additional robust findings include that subcellular localization (45,54,63,65,89,139), proteoforms (89,139), protein disorder (53,56,89), protein abundance (54,83,89), GO category (53,54,56,65), and cell type (61) are all correlated with proteome turnover rates to some degree ( Figure 3A).
In contrast to the above-mentioned results that showed general agreement between several studies in different systems, there is a lack of strong agreement regarding the relationships between protein turnover rates and other key molecular features such as mRNA half-lives (53,56,120), N-terminal motifs (50,54,56,65), codon sequences (82), amino acid composition (53,56,82) and other intrinsic properties. It is possible that this disagreement is due to biological differences in different systems. Largescale meta-analyses and literature mining (140) of proteome turnover studies will be useful for reaching more answers, and potentially consensuses, particularly as more data is generated. Additionally, re-examining preexisting datasets with an eye towards unexplored or overlooked parameters will yield greater insight into the underlying biological and biochemical principles driving proteome turnover. A first glimpse into such studies is provided by recent work from the Ghaemmaghami lab. They propose that differences in proteome turnover rates underlie organism-level phenotypes such as longevity, and that mechanisms of protein turnover regulation are linked to metabolic processes. As such, the principles underlying protein turnover rate differences may vary substantially between species (60). independent acquisition (DIA) strategies that recently outperformed data dependent acquisition (DDA) methods significantly in particular for low-input proteomics (146,147), a significant increase in sensitivity could be achieved. Recent studies by the Aebersold and Liu labs already used a dynamic SILAC-DIA approach to monitor proteome turnover at high accuracy and sensitivity (62,73,148).
For in vivo studies, the convolution of signals from different cell types still represents a major challenge. The use of artificial amino acids that are specifically incorporated into a target cell type only by genetic targeting may pose a very attractive option for studying proteome turnover in specific cell types in the future (100,108,111).
What comes next (Box 1)? A recent study by the Walther group is a great example of the kind of future studies that will shift us from more descriptive, correlative protein turnover studies to investigations that provide comprehensive functional insight into proteostasis and dynamic proteome remodeling. Building on previous work from the same group in which they compared the protein turnover rate between two distant yeast species (58), the authors used proteome-wide protein turnover measurements to match degradation pathways to individual proteins in S. cerevisiae (149). To this end, they combined systematic single gene deletions of over one hundred components of the yeast degradation machinery (e.g. E2-and E3-ligases) with quantitative measurements of protein turnover, thereby mapping protein degradation pathways for hundreds of genes.
This massive effort enabled the authors to identify the endogenous targets of the majority of E2-and E3-ligases, and could serve as a blueprint for future studies about protein turnover (149). Due to the technological advances described above, combined with the by guest on January 10, 2021 availability of genetic perturbation technologies like CRISPR and targeted drug screens, the time is ripe to match protein turnover to functional pathways.
These novel combinatorial approaches could be used to answer some of the remaining open questions regarding proteome turnover and its regulation (Box 1). Such questions include identification of the molecular events mediating the crosstalk between synthesis and degradation pathways (150) and illuminating the mechanisms by which complex assembly is coupled to protein turnover (151,152). Future mechanistic studies could also elucidate how the determinants of protein half-lives, such as proteoform identity, are able to confer such variability in turnover kinetics (89). In recent years, enormous effort has been expended to quantify the proteomes of various cancer cell lines and primary tissues. We predict that similar efforts in categorizing proteome-wide protein turnover rates will be well worth the investment of cost and effort, as they will provide unique, complementary information regarding the principles underlying these essential processes driving protein expression (62).
by guest on January 10, 2021

Box 1: Some open questions and considerations regarding proteome turnover
What are the gene specific regulators of protein production and degradation? E.g. what are the targets of specific E3 ligases?
What are the mechanisms by which synthesis and degradation pathways are coupled? How does inhibition of the proteasome change synthesis rates of many genes and vice versa?
Through what molecular events are subunits of protein complexes coherently turned over?
To what extent can we expect turnover rates to stay constant between two identical cultures? More importantly, how relevant are turnover rates functionally? Are differences in turnover rates between cultures predictive of differences in functional/phenotypic responses between them?
What are the relative strengths and weaknesses of gathering turnover rate information compared to other cell parameters (eg RNA-seq, proteomics, etc) in terms of giving relevant information that provides functional insight into cell behavior and disease, beyond insight into the mechanisms of turnover itself?   isotope (e.g. 13 C6-Arg), typically a semi-/medium-heavy isotope, can also be generated as normalization standard for data analysis. It should be noted that label switches can also be done in a different way than depicted here (e.g. cells good be grown in "heavy" amino acids and then pulsed with "light amino acids) (B) Data acquisition. LC-MS/MS enables direct monitoring of light (red) and heavy (green) peptide signals, which correspond to pre-existing and newly-synthesized proteins, respectively. For dynamic SILAC-TMT experiments, relative quantification of each sample is completed at the MS 2 level (far right). In 3-channel designs, the signal from the constant semi-heavy labeled sample (yellow) spike-in provides an internal normalization standard between different mass spectrometry measurement, allowing for relative signal from light and heavy channels to be quantitated. (C) Data analysis. Here we show data for an example protein measured from a 2-channel dynamic SILAC experiment (left), a 3-channel dynamic SILAC experiment (middle), and a combined 2-channel dynamic SILAC-TMT experiment (right). With 2-channel dynamic SILAC, half-lives and kdeg can be calculated using lntransformed heavy over light (H/L) peak ratios over time, but due to run to run variability during the mass spectrometry measurements it is difficult to separate the contributions of synthesis and degradation. On the other hand, data from 3-channel dynamic SILAC and dynamic SILAC-TMT can be used to determine ksyn separately from kdeg. For 3-channel dynamic SILAC, this can be achieved by plotting heavy isotope over medium isotope (H/M) signal to generate a synthesis curve, while plotting light isotope over medium isotope signal (L/M) generates a curve for protein degradation. In dynamic SILAC-TMT, all the heavy (H) and light (L) signals are measured in the same, which allows for separate synthesis and degradation curves.  degradation. Proteins involved in multimeric complexes tend to have turnover rates that are generally coherent. Although cells are capable of translating complex members at stoichiometric equivalencies, in yeast and mammals they can overexpress one or more members of the complex, then degrade a subset of them post-translationally to achieve stoichiometric equivalency (shown in orange, dashed lines). Some complex members may be synthesized in excess but not degraded, and may perform additional functions either as free subunits or as members of other complexes (shown in teal). These proteins may actually have two different turnover rates depending on complex association or subcellular location, but only one aggregate turnover rate will be measured in standard dynamic SILAC-based approaches. Despite overall agreement in turnover rates, complex by guest on January 10, 2021 Supplemental Table 1: Advantages and limitations of approaches to study proteome turnover discussed in this review.
See supplemental file.