Combining heavy isotope labeling with complement reporter ion quantification enables high-quality protein turnover measurementsWe wanted to measure protein turnover rates and evaluate how these rates vary across growth conditions. To simplify our measurements, we grow E. coli in chemostats, where we can control the cell doubling time and enforce steady state (Fig. 1A). After cells reach steady state, we change the inlet medium from unlabeled nutrients to 15N-labeled ammonium. Over time, the 15N-ammonium concentration in the reactor increases, and newly synthesized proteins incorporate more heavy isotopes. We can monitor the shift in the isotopic envelope of peptides by taking samples after the media switch using proteomics (Fig. 1B). With the knowledge of a peptide’s chemical composition and the fraction of heavy isotopes over time, we can calculate the turnover rate of the corresponding protein. In practice, however, obtaining such measurements of isotopic envelopes in the MS1 spectrum is quite challenging, particularly at later time points when the isotopic envelopes spread out and overlap with those of other peptides. MS1-based label-free quantification relies on accurate assignment of peaks to peptide elution profiles, which is made more challenging by the complexity of spectra. Additionally, traditional search algorithms struggle to identify peptides with a significant fraction of heavy isotopes, so missing values at later time points are a severe limitation of such approaches42. To overcome these limitations, we labeled samples at each of the acquired eight-time points with TMTpro isobaric tags and combined them for co-injection into the mass spectrometer46,47. While MS1 spectra are still extremely complex using this approach, by combining isotope envelopes of peptides with and without 15N, we ensure that the pseudo-monoisotopic peak (M0) is always present for isolation and fragmentation in the MS2, increasing peptide identification rates and alleviating the missing value problem. Multiplexing experiments with highly complex MS1 spectra inevitably increase co-isolation of multiple peptides in a single MS2 spectra. Analyzing these extremely complex samples with standard low m/z reporter ion quantification would lead to severe ratio distortion and measurement artifacts37,38,48. We overcame this limitation by quantifying the balancer-peptide conjugates (complement reporter ions) that remain after reporter ions are cleaved off in the MS2 spectra. Complement ions have peptide-dependent m/z ratios that are typically slightly different than co-isolated peptides. Therefore, using the complementary ions for quantification reduces ratio distortion effects compared to both MS2 and multi-notch MS3 reporter ion quantification46,47(Fig. 1C).Fig. 1: Combining heavy isotope labeling with an accurate multiplexed proteomics method (TMTproC) enables high-quality measurements of unperturbed protein turnover.A Experimental setup. E. coli cells were grown in chemostats with a defined doubling time. After reaching steady state, the chemostat feed was switched to a medium with 15N-labeled ammonium. Newly synthesized proteins will increasingly incorporate heavy isotopes. Proteomics samples were collected at various time points to determine the protein turnover rate98. B Theoretical isotopic envelopes of an example tryptic peptide, which is assumed to be stable (protein is removed from the vessel only through dilution). Over time, the increasing fraction of heavy ammonium in the peptide’s structure shifts the isotopic envelope to higher masses. Peptides were labeled with isobaric tags (TMTpro) to encode different time points. C Top: theoretical MS1 spectrum for a single peptide species after combining labeled peptides from all the time points. The mass spectrometer was set to isolate the monoisotopic peak (M0) and fragment the peptide. Bottom: the resulting complement reporter ions (peptide plus broken tag) enable accurate quantification of the relative abundance within the M0 peak over time. D Example measurements for the stable OmpF protein and rapidly degrading RpoS protein. Each dot indicates the peptide quantification relative to the median level measured when the feed was switched. The size of each point is proportional to the number of measured ions. Fitting the observed data with the theoretical decay profile for M0, we can extract the total half-life for each protein (solid curve). The dotted curve shows the theoretical decay for a stable protein (n = 8 time points). E Scatter plot of measured protein total half-lives for biological replicates of carbon-limited E. coli grown with a 6-h doubling time. Dotted lines indicate the cell doubling times. The solid line marks the 1:1 line. The total half-lives for each protein were calculated from the fits shown in (D). Median standard deviation for the total half-lives between the replicates is 0.3 h.Figure 1D shows the peptide-level quantification for a stable protein, OmpF, and an unstable protein, RpoS, from carbon-limited chemostats with 6-h doubling times24,49,50. We built a differential equation model51 for the dynamics of M0 with a parameter of kD (active degradation rate) and the known variable D (dilution rate) (Supplementary Note, Supplementary Data 8). For every protein, kD + D (total turnover rate) is obtained by fitting this model to the experimentally measured signal of peptides. For a protein that is not actively degrading (kD = 0) and only diluting with D set entirely by the chemostat, we can estimate the dynamics of M0. This is called the dilution curve (dotted). Fitting kD to the measured signal of OmpF peptides yields kD + D ~ D which indicates that this protein is not actively degrading. In contrast, the deduced total half-life for RpoS is much shorter than the cell doubling time. We obtain half-lives for ~2.6k E. coli proteins per experiment with a median standard deviation of 0.3 h (Fig. 1E, Supplementary Table 1). Having established this technology, we acquired similar measurements for 13 different growth conditions, each with two biological replicates, quantifying the turnover rates of ~3.2k proteins in at least one condition (Table 1, Supplementary Data 1, Supplementary Data 7). We then used this resource to investigate how E. coli adapts protein turnover under various growth conditions.Table 1 Summary of the 13 growth conditions for which we measured protein turnover rates
E. coli recycles its cytoplasmic proteins under nitrogen limitationBuilding on our method to measure protein turnover, we wanted to compare protein turnover rates under various nutrient limitations. To this end, we compared carbon (C-lim), phosphorus (P-lim), and nitrogen (N-lim) limitation measurements from chemostats with 6-h doubling times. We found that most proteins in C-lim are stable with a measured total half-life close to the theoretical dilution time (Fig. 2A). Using biological replicates to identify degrading proteins with high confidence (Fig. 2B), we found that 15% of the proteome is actively degraded in C-lim (p-values < 0.05). Protein half-lives under P-lim have a similar distribution and a similar percentage of proteins that degrade with high confidence. However, in N-lim we found that 43% of proteins are actively degraded (Fig. 2B, p-value < 0.05).Fig. 2: E. coli recycles its cytoplasmic proteins when nitrogen is limited.A Histogram of protein total half-lives for E. coli grown in chemostats under C-lim, P-lim, and N-lim. The vertical line marks the dilution limit set by the 6-h doubling time. Total half-lives greater than doubling time indicate measurement noise. Under C-lim and P-lim, most proteins have total half-lives equal to the doubling time, suggesting they are stable. However, under N-lim many proteins are actively degraded. B Separation of the proteome into stable proteins (grey) and actively degrading proteins (yellow). All proteins with p-value < 0.05 are called confidently degrading (t-test, one-sided, n = 2). In C-lim and P-lim, 15% of the proteome turns over with high confidence. In contrast, under N-lim, 43% of the proteome turns over. C Distribution of total half-lives for proteins from different subcellular localizations overlayed against the entire proteome. Most proteins are stable under C-lim and P-lim, irrespective of localization. However, nearly all cytoplasmic proteins slowly degrade under N-lim while the membrane and periplasmic proteomes are largely stable. D Scatter plots of protein total half-lives in different nutrient limitations. The dotted black lines mark the dilution limit, the solid black line denotes perfect agreement. Contour plots contain 50% of the probability mass for each subcellular compartment. The contour plots of membrane and periplasmic proteins are centered around the dilution limit in all the binary comparisons, indicating that most of these proteins are stable under all limitations. However, the shift in the contour plots of the cytoplasmic proteins on comparing N-lim with P-lim and C-lim suggests that the cytoplasmic proteins are degraded in N-lim. E Measurement of protein turnover rates under complete nutrient starvation in batch. In batch, like the N-lim chemostat, the cytoplasmic proteins are degrading with high confidence as compared to the membrane and periplasmic proteins (ANOVA, p-value = 9E–16, n = 2114 proteins). The box extends from the first quartile to the third quartile of the data, with a line at the median. The whiskers extend from the box to the farthest data point lying within 1.5× the inter-quartile range from the box.We found that the increase in protein degradation in N-lim could be attributed to the active degradation of a wide range of cytoplasmic proteins (Fig. 2C, D). The mode protein total half-life for membrane and periplasmic proteins in all three conditions is very close to the theoretical dilution limit. In contrast, the mode protein total half-life for cytoplasmic proteins is significantly shorter under N-lim than C-lim or P-lim. We estimate that 56% of cytoplasmic proteins are actively degraded in N-lim, while only 13% of membrane proteins and 4% of periplasmic proteins undergo active degradation in this condition. Due to measurement noise and low sample sizes, we expect to be statistically underpowered and that these estimates are likely lower bounds of the true extent of protein degradation in N-lim.We then tested whether cytoplasmic protein degradation in N-lim chemostats extends to the more physiologically relevant case of batch starvation. We grew E. coli cells in minimal medium until they reached an OD600 of ~0.4. We then switched the exponentially growing cells into medium depleted of nitrogen (Fig. 2E). Once again, many cytoplasmic proteins are degraded under nitrogen starvation, and membrane/periplasmic proteins are largely stable. Thus, E. coli cells slowly degrade their cytoplasmic proteins when nitrogen is scarce in both chemostats and batch cultures. About 2/3 of the cell’s nitrogen is stored in proteins52. The degradation of proteins upon nitrogen starvation likely allows the regeneration and recycling of scarce amino acids and enables E. coli to produce new proteins to adapt to new environments.Measuring protein turnover in knockout mutants allows the identification of protease substratesNext, we were interested in discovering the protease(s) responsible for the large-scale turnover of cytoplasmic proteins in N-lim. Combining protein-turnover measurements with genetic protease knockouts allows us to investigate protease-substrate relationships on a proteome-wide level. Since unfolding and degrading stably folded cytoplasmic proteins requires energy, we focused on assigning substrates to the ATP-dependent proteases. In E. coli, there are four known cytoplasmic ATP-dependent protease complexes: ClpP (in complex with ClpX or ClpA), Lon, HslV (in complex with HslU), and FtsH1. We identify putative substrates for the first three of these proteases by comparing the protein half-lives in protease knockout (KO) with wildtype (WT) cells (Fig. 3A). We were able to validate several known protease-substrate targets and identify degradation pathways using these experiments. For example, the unfoldase ClpA is completely stabilized by knocking out clpP, consistent with previous literature53. We identified Tag and UhpA as putative substrates of Lon and HslV, respectively. However, many proteins still degrade in the three protease KO lines, e.g., the phosphatase YbhA—which contributes to Vitamin B6 homeostasis—still rapidly turns over with a total half-life of ~1 h in each knockout strain54. Surprisingly, even the proteins that increase their half-lives in single KOs are often not completely stabilized. Additionally, bulk cytoplasmic proteins are still degraded in all three single KOs.Fig. 3: Measurement of protein turnover in protease knockout strains enables proteome-wide identification of protease substrates.A Scatter plots of protein total half-lives of N-limited wild type (WT) compared to ΔclpP, Δlon, and ΔhslV knockout (KO) cells. Dotted lines mark the dilution limit, the solid black line indicates perfect agreement. Substrates (black x) increase their total half-lives in KOs with high confidence (t-test, one-sided, p-value < 0.10). ClpA (pink), Tag (teal), and UhpA (purple) are the substrates of ΔclpP, Δlon, and ΔhslV, respectively. However, the protein YbhA (orange) is still degraded in individual KOs. Contour plots containing 50% of the probability mass for the cytoplasmic (red) and membrane (green) proteins indicate that individual KOs degrade bulk cytoplasmic proteins. B Since ΔftsH cells cannot grow in chemostats, we repeated the batch starvation assay as in Fig. 3E. The box extends from the first quartile to the third quartile, with a line at the median. The whiskers indicate 1.5× the inter-quartile range from the box. Results indicate that the ΔftsH cells, like the WT, also degrade their cytoplasmic proteins under nitrogen starvation (t-test, two-sided, p-value = 2E–12 for WT and p-value < 2E–16 for ΔftsH, n = 1519). C Scatter plots of protein total half-lives of WT and ΔclpP Δlon ΔhslV cells in N-lim. The substrates (black x) increase their total half-lives in the triple KO (t-test, one-sided, p-value < 0.09). Many proteins are still degrading in the triple KO, e.g., LexA and YoaC (blue). In fact, the bulk cytoplasm is still degraded. However, many more proteins are stabilized in the triple KO compared to the individual KOs, indicating redundancy among substrates, e.g., YbhA (orange). D Comparing the shifts in the WT and KO strains’ total half-lives, we assign each protease’s contribution to active protein turnover. The bar graph represents examples from each of the six categories—turnover explained predominantly by ClpP, Lon, HslV, additive contributions, redundant contributions, and actively degrading proteins in the triple KO. E Bar graph for the number of substrates and the % of the proteome assigned to each of the six categories described in (D). F Comparison of the substrates from our categories in E with previous proteome-wide substrate-protease assignment studies. ClpP trapped substrates significantly overlap with the identified ClpP substrates (Fisher test, two-sided, p-value = 6E–9), and previously identified substrates of HslV and FtsH show a significant overlap with redundant and additive substrates (Fisher test, two-sided, p-value = 1E–3). G Comparison of the percentage of active turnover per hour across the protease KOs under N-lim. Even after knocking out hslV, lon, and clpP simultaneously, 40% of the WT proteome turnover remains, suggesting that a major pathway of protein degradation in E. coli remains to be discovered.Deleting ftsH is more complicated than the other proteases. One of its substrates, LpxC, catalyzes the committed step in the lipid A synthesis pathway. Lipid A is the hydrophobic anchor of lipopolysaccharides (LPS), a critical outer membrane component. Deletion of ftsH leads to increased levels of LpxC, causing an accumulation of LPS that makes the cells nonviable55. ftsH null cells can be rescued with a mutation of FabZ (L85P), which slows LPS synthesis and compensates for the increased LpxC levels55. Interestingly, we were only able to generate the ΔftsH fabZ (L85P) strain in DY378 background56. Our attempts to knock out ftsH in the NCM3722 background used for the remainder of this paper failed. We are currently investigating which other modifications in DY378 might make ΔftsH fabZ (L85P) viable. These ΔftsH fabZ (L85P) cells are viable, though unfortunately, they grow too slowly on minimal media and are washed out of the chemostat. Therefore, we could not measure protein turnover in a ftsH mutant in a similar manner to the other proteases. Instead, we repeated the batch nitrogen starvation experiments (Fig. 2E). Similar to the WT cells, cells lacking ftsH degraded cytoplasmic proteins. In contrast, membrane proteins are mostly stable (Fig. 3B). This indicates that none of the four known ATP-dependent proteases in E. coli are individually responsible for the large-scale cytoplasmic recycling that occurs under nitrogen limitation.We then asked if proteases might act redundantly, i.e., multiple proteases share a substrate, which could mask the effects of deleting individual proteases. To this end, we measured protein turnover in a triple KO line (ΔhslV Δlon ΔclpP) in nitrogen limitation. A quantitative comparison of protein turnover rates between the triple KO and the individual KOs allows us to assign the contribution of each protease in turning over a substrate (Fig. 3D). We can classify the substrates into six groups: those being degraded predominantly by a single protease, those where the effects of the individual proteases are additive, those that are stabilized more in the triple KO than the combined effect of individual KOs (redundantly degraded), and those that are still actively degraded in the triple KO (Supplementary Data 2, Supplementary Fig. 7).We classified 64 and 14 substrates to be predominantly degraded by ClpP and Lon, respectively. We only assigned one substrate uniquely to HslV: UhpA, a transcriptional regulator that activates the transcription of genes involved in transporting phosphorylated sugars57. Eighty-two proteins are degraded additively, a notable example of which is IbpA, a small chaperone. Previous studies have proposed that Lon degrades free IbpA/lbpB and bound client proteins58. We found that ClpP and Lon contribute approximately equally to the degradation of IbpA, and their contribution is additive.We classify 41 proteins as being redundantly degraded by two or more proteases. For example, YbhA is rapidly degraded in all single KO strains but stabilized in the triple KO, indicating that at least two of these proteases act redundantly. Interestingly, the majority of cytoplasmic proteins are still slowly degraded in the triple KO under nitrogen limitation, and we classified ~100 proteins as still being actively degraded (Fig. 3E). LexA, an SOS repressor, auto-degrades itself under stress and unperturbed growth59,60. Consistent with this, LexA still undergoes degradation in the triple KO. It will be interesting to investigate if other proteins with short half-lives in the triple KO are auto-degrading, degraded by FtsH, or if other mechanisms are at play.To validate our classifications, we compared our protease-substrate relationships with previous proteome-wide measurements. We see a significant overlap (p-value = 6E–9) of our identified ClpP substrates with substrates identified via a trap mutant (Fig. 3F)20. However, we do not observe an overlap of our putative Lon substrates with a previous Lon-trap experiment (p-value = 0.22)21. This lack of overlap is most likely caused by our separating the Lon trap substrates into the different classifications, indicated by a more significant overlap with the substrates that were stabilized in any of our KO strains (p-value = 0.05). This is consistent with previous observations that Lon substrates are often shared with other proteases61. Interestingly, the putative substrates of HslV identified through a microarray study show a strong overlap with the proteins we classify as additive or redundant (p-value = 0.001)23. This is consistent with previous reports that HslV substrates are shared with other proteases26,62. We also found enrichment (p-value = 0.002) between substrates identified in a previous FtsH trap63 study and additive or redundant substrates, consistent with findings that FtsH often degrades proteins that are also substrates for other proteases27. The lack of significant overlap between the proteins still degrading in the triple KO and FtsH-trap substrates implies that FtsH is likely not involved in the degradation of these substrates.Surprisingly, 40% of active protein degradation in nitrogen limitation in wild-type cells persists upon knocking out the three canonical ATP-dependent cytoplasmic proteases (Fig. 3G, details of calculation in the supplementary note and supplementary data 6). We could not generate a viable quadruple KO with ftsH deletion, so we cannot rule out the possibility that all four proteases act redundantly as an explanation of the remaining protein degradation. However, the results from the individual ftsH knockout (Fig. 3B) and the lack of overlap between degrading proteins and the FtsH-trap experiment (Fig. 3F) are evidence against FtsH being responsible for the remaining degradation. Regardless, a major pathway for degrading proteins in E. coli remains to be discovered: either FtsH plays a much bigger role than is currently believed, or a completely new mechanism degrades cytoplasmic proteins under nitrogen starvation.Analyzing features of rapidly turning over proteinsWe found that most short-lived proteins have similar half-lives regardless of nutrient limitation (Fig. 4A, Supplementary Data 3). With gene-set enrichment64, we found that rapidly degraded proteins were enriched in transcriptional regulators (Benjamini–Hochberg adjusted p-value = 4E–4). A protein’s response time depends on its turnover rate5. Proteins involved in transcriptional regulation might need to rapidly adjust their levels to changing growth conditions.Fig. 4: Features of proteins with short total half-lives.A Scatter plot of protein total half-lives for C-lim, P-lim, and N-lim conditions. The total half-lives of rapidly degrading proteins are typically similar under different nutrient limitations. Three proteins with the shortest average total half-lives are marked. Short-lived proteins are enriched for transcriptional regulators (BH p-value = 4E–4). B The 24 proteins with the shortest mean total half-lives. For eight of these proteins (in blue), we could not find any prior literature evidence for degradation, and six (marked with *) contain Fe–S clusters (BH p-value = 0.048). (C, E, F) For all the boxplots, the box extends from the first quartile to the third quartile of the data, with a line at the median. The whiskers extend from the box to the farthest data point lying within 1.5× the interquartile range from the box. C Smaller proteins have shorter total half-lives. Left: Box plot of total half-lives averaged over all the nutrient limitations for low and high molecular weight (MW) proteins (n = 2864 total proteins, p-value = 1E–11, Mann–Whitney U, one-sided). Right: Fold enrichment of low versus high MW proteins as a function of total half-life. D Comparison of the proteome-wide N-terminus amino acid residues obtained from this study and prior literature. The size of the marker indicates the number of proteins with a particular residue. E No correlation between the N-terminal protein residue and protein total half-lives (N-end rule). Left: Distribution of the N-terminus residues for actively degrading and stable proteins. Right: Box plots of total half-lives for the destabilizing and stabilizing residues on their N-terminus (p-value: 0.99, Mann–Whitney U, one-sided, n = 373 total proteins). F Disordered proteins tend to have shorter total half-life99. Left: The Espritz algorithm classified proteins as ordered or disordered. Right: Box plot of total half-lives for disordered and ordered proteins (n = 2865 total protein, p-value = E–18, Mann–Whitney U, one-sided). G Fold enrichment of disordered versus ordered proteins as a function of total half-life for the WT (olive) and triple protease KO (brown) cells. Rapidly turning over proteins are enriched for the disordered category. This enrichment becomes more pronounced when three proteases are knocked out.Using our data set, we validated examples of degradational regulation that had previously been reported and also uncovered targets. Of the 24 proteins with the fastest average turnover rates, 17 were previously reported to be degraded (Fig. 4B). Seven proteins—ThiH, YgaC, SixA, YciW, CbI, ThiG, EpmB—had no prior evidence in the literature for degradation. Interestingly, six rapidly degrading proteins—ThiH, BioB, IscA, IscR, EpmB, Fnr—contain Fe–S clusters, which is significantly higher than expected by random chance (BH p-value = 0.048). Flynn et al. previously proposed that Fe–S binding proteins are degraded under aerobic conditions, likely because the Fe–S clusters are oxidized, destabilizing the protein20.Multiple metabolic enzymes such as PatA, LpxC, and HemA are also rapidly degraded. Rapid degradation allows for immediate and direct control over intracellular protein levels based on cellular demand. PatA (Putrescine-Aminotransferase) is involved in putrescine (polyamine) degradation (KM = 9 mM) and is unstable under standard growth conditions with high putrescine levels, in which another enzyme (PuuA – glutamate-putrescine ligase) dominates usage of putrescine. PatA is expected to stabilize in specific growth conditions with low putrescine concentrations65. LpxC, a protein required for lipid A synthesis, is rapidly degraded under slower growth to balance LPS production with cellular demand55. HemA, involved in porphyrin biosynthesis, is degraded when the media lacks heme as an iron source66. DnaQ, the proofreading exonuclease of the stable DNA polymerase III core enzyme [DnaE][DnaQ][HolE], is rapidly degraded (t1/2 = 1.2 h). DnaE is more stable with a total half-life of 5 h, whereas HolE is undetected in our data set, likely due to its short length. Free DnaQ is unstable but stabilized on complexation with HolE67.We next tested whether these rapidly degrading proteins share attributes such as their physiochemical properties, sequence features, or structural characteristics. We found that smaller proteins (MW < 10 kDa) have significantly shorter half-lives regardless of the nutrient limitation (Fig. 4C, Supplementary Fig. 2A). This enrichment was more pronounced at lower total half-life cutoffs (Fig. 4C). On the other hand, charge and isoelectric point were not significantly correlated with half-lives under P-lim and C-lim (Supplementary Fig. 2B, C). However, both the charge and the isoelectric point of a protein were correlated with half-lives under N-lim (Supplementary Fig. 2B, Cp-value = E–20). This is most likely because cytoplasmic proteins are short-lived under N-lim while membrane proteins are typically stable. Membrane proteins tend to have higher isoelectric points and more positive charge due to their interaction with negatively charged phospholipids68,69.One obvious sequence feature to investigate is the N-end rule, which relates a protein’s stability to its amino-terminal residue70. Amino-terminal arginine, lysine, leucine, phenylalanine, tyrosine, tryptophan, and formylated N-terminal methionine (fMet) are believed to be destabilizing residues, whereas the other residues are believed to be stabilizing70,71. First, we determined the in vivo N-terminal residue of ~600 proteins using a label-free proteomics data set (Supplementary Data 4). MS2 spectra were searched with the Sequest algorithm72 considering all possible N-terminal tryptic subfragments for a protein. Encouragingly, when we compared a small subset of the identified N-termini with previous data in literature obtained using Edman Degradation73we found nearly perfect agreement (55/61 proteins) (Fig. 4D). Surprisingly, however, our rapidly degrading proteins showed no enrichment for the previously reported destabilizing N-terminal amino residues (Fig. 4E). Interestingly, few proteins detected in either dataset had destabilizing N-terminal residues. It is possible that proteins with destabilizing N-termini are immediately degraded and therefore difficult to detect. Proteins which had destabilizing residues exposed via cleavage would similarly be short-lived and low abundant. In this case, the main-determinant of protein half-life would be the rate at which destabilizing N-termini are exposed. Either way, our results suggest that the N-terminus of E. coli proteins is not the primary determinant of proteins’ in vivo stability.The SsrA-tagging system is another known degron used for marking polypeptides for degradation whose translation has stalled74To investigate how much of the observed protein degradation could be attributed to the system, we knocked out the smpB gene (codes for a protein in the SsrA tagging complex75), and measured gene-by-gene protein turnover in nitrogen limitation with a 6-h doubling time in duplicate. We did not observe significant changes in protein turnover compared to the wild-type strain (Supplementary Fig. 10). It’s possible that the primary targets for the SsrA-tagging system are low abundant relative to stably-fold proteins, or that another pathway (e.g., ArfA-mediated76) for releasing stalled ribosomes is redundant with the SsrA system.Another sequence feature previously shown to affect protein stability in bacterial and eukaryotic cells is intrinsically disordered protein segments77,78. To this end, we determined the percentage disorder for all the proteins using the Espritz algorithm79. Disordered proteins had significantly shorter half-lives than ordered proteins (Fig. 4F, p-value = E–208). Interestingly, this enrichment further increases when we use protein half-lives measured in the triple protease knockout cells (ΔhslV Δlon ΔclpP) (Fig. 4G). This is consistent with ATP-dependent proteases being able to unfold and digest structured proteins. Once these proteases are removed, the remaining proteins with short half-lives should be enriched for those that are unstructured and therefore prone to degradation by energy-independent proteases.Analysis of turnover for functionally related proteinsNext, we investigated protein turnover for functionally related protein modules, such as multiprotein complexes, operons, and metabolic pathways. We calculated each module’s coefficient of variation (CV) and compared this to the CV distribution when proteins were randomly assigned to sets. We observe that the functionally associated modules exhibit significantly lower variance than if the proteins were randomly assigned to each module (Fig. 5A), suggesting that functionally associated proteins tend to exhibit similar half-lives. For example, twelve of the fourteen proteins involved in phosphonate metabolism and transport are rapidly degraded (average total half-life = 0.7 h) under P-lim (Fig. 5B). These proteins were 16-fold more abundant in P-lim compared to N-lim and C-lim (Supplementary Data 5). Therefore, we were unable to measure their half-lives in C-lim or N-lim, so it’s unclear if they turn over in these limitations as well.Fig. 5: Protein total half-lives of functionally related proteins.For all the box plots in this figure, the box extends from the first quartile to the third quartile of the data, with a line at the median. The whiskers extend from the box to the farthest data point lying within 1.5× the inter-quartile range from the box. A Functionally related proteins tend to have similar total half-lives. Box plots of CVs of total half-lives for functionally related proteins compared with a randomized set of proteins (Mann–Whitney U, one-sided, p value 2E–6, 2E–2, 8E–7 for complexes, operons, and pathways respectively). B Map of the total half-lives for proteins involved in the phosphonate pathway under P-lim. The phosphonate pathway is encoded in a single operon and is involved in the transport and metabolism of organophosphonates (C–P bond). Most of these proteins degrade rapidly under P-lim. C The rut operon consists of seven genes and one repressor (n = 5 detected). Four of the five proteins were rapidly degraded in the wild-type strain and ΔclpP, Δlon, and ΔhslV. These four proteins were stabilized in a triple knockout ΔclpP Δlon ΔhslV. D The wild-type and TKO strain were grown on glucose minimal media with thymidine as the sole nitrogen source. The mutant strain grows to a higher maximum OD600 than the wild-type strain. A LOESS curve was fit to the data with a 99% confidence interval. E Map of the half-lives of flagellar proteins under C-lim. The basal body elements of the flagellum are largely stable, while filament and motor structure components are rapidly degrading. Flagellum schematic adapted from ref. 100. F Scatter plot of protein total half-lives for protein complexes containing two distinct subunits. Data is presented as mean +/− standard deviation. Highlighted in purple are complexes for which the total half-lives of subunits disagree, potentially because their interaction is transient, e.g., ClpA/P or BolA, GrxD. G Box plots of total half-lives for ribosomal proteins compared to all cytoplasmic proteins. Ribosomal proteins are more stable than cytoplasmic proteins in C-lim and N-lim but degrade faster than the median of cytoplasmic proteins in P-lim (n = 30 for ribosomal proteins).One pathway where this is particularly noteworthy is the rut pathway for pyrimidine degradation. The pathway is transcribed by a single operon of seven genes (rutABCDEFG) and has one adjacent transcriptional repressor, rutR. In our database, we detected five of the seven proteins plus the repressor. Of these five proteins, four are rapidly degraded in the wild-type strain and the single protease knockouts. The fifth protein, RutG, is found in the membrane, which likely explains its stability. However, when we knock out lon, clpP, and hslV simultaneously, the four proteins are all significantly stabilized (Fig. 5C).The study that originally characterized the rut pathway showed that when the rut genes are transcriptionally upregulated, E. coli was able to grow on thymidine as the sole nitrogen source in minimal media at room temperature80. We reasoned that by removing their degradation, we could similarly increase protein concentrations and observe the same growth on thymidine. Indeed, the triple protease knock-out strain grows to an OD600 of nearly 0.28 on thymidine while the wild-type strain reaches an OD600 of 0.03 (Fig. 5D).Additionally, proteins associated with flagella show correlated expression levels and half-lives. Surprisingly, most of the proteins forming the basal flagellar body are stable, but the filament (FliC, FliD), motor (MotA, MotB), and sensory proteins (CheA, CheW) are rapidly degraded (Fig. 5E). Future work will be required to decipher the underlying mechanisms and functional relevance.In general, proteins that form a complex tend to exhibit similar half-lives (Fig. 5F). Several complexes whose subunits are degraded at different rates are known to interact weakly or transiently or have subunits which are expressed non-stoichiometrically, suggesting that at least some of these discrepancies might be due to annotation details. For example, ClpA and ClpX are the unfoldases in complex with ClpP. Autodegradation of ClpA is used to regulate the number of ClpAP complexes in the cell and the flow of substrates to ClpAP53. Finally, antitoxins like PrlF are subject to regulated degradation while their toxin counterparts are stable81,82.The ribosome is one of the heterocomplexes which exhibits unanticipated patterns under different nutrient limitations (Fig. 5G). Under both C-lim and N-lim, ribosomal proteins are slightly more stable than the median cytoplasmic protein. However, under P-lim, ribosomal proteins are less stable than typical cytoplasmic proteins. rRNA contains about 50% of the cellular phosphorus83. Therefore, cells likely recycle the phosphorus stored in rRNA when phosphorus is scarce, and associated ribosomal proteins might become unstable once their binding partners are lost.Active protein degradation rates typically do not scale with division ratesSo far, we have compared turnover rates under various nutrient limitations but with the same cell division time. We wanted to determine how active protein degradation scales with cell cycle time. The total turnover rate (ktotal) of a protein is a combination of active degradation (kactive) and dilution (kdilution) due to cell division (Fig. 6A). We consider two simple and reasonable models of the relationship between these two parameters. In the first model, kactive scales with kdilution, i.e., the protein total half-life remains a constant fraction of the cell cycle time. In the second model, active degradation rates are independent of growth rate, i.e., the active degradation rate of each protein remains constant regardless of cell doubling time.Fig. 6: Active degradation rates are generally uncoupled from cell division rates.A Two simple models describing the relationship between cell division and protein-specific turnover rates. The total protein turnover rate (ktotal) is the sum of the active degradation rate (kactive) and the dilution rate due to cell division (kdilution). In the “scaled model,” active degradation rates increase proportionally to division rates with a protein-specific constant (αp), i.e., active degradation remains a constant fraction of the total protein turnover rate. In the “constant model,” protein-specific active degradation rates are constant (βp), regardless of changing division rates. In this case, for slower-dividing cells, the contribution of active degradation increases relative to dilution. B t1/2, total is the time taken to replace half the protein. A theoretical plot of t1/2, total from two conditions (i, j) where cell division rates change by a factor of ri,j < 1. In the scaled model, t1/2, total values for all the proteins lie on a straight line with slope ri,j (orange). In the constant model, the t1/2, total values follow a nonlinear relationship between the two doubling times (purple). For proteins with very high active degradation rates, the constant model predicts that t1/2, total will approach the same value for both doubling times, indicated by the slope 1 line (black). For diluting proteins with no active degradation, both models converge to the doubling times of conditions i and j. C Scatter plots of protein t1/2, total for E. coli grown at doubling times of 6 h (C-lim), 3 h (C-lim), and 0.7 h (defined minimal media batch) compared to 12 h (C-lim). The dotted lines represent the dilution limit. We observe a strong statistical preference for the “constant model,” in which active degradation rates are uncoupled from cell cycle duration. Shown are the likelihood ratios (L) of the constant models compared to the scaled models assuming normally distributed errors.The two models have distinct predictions on how the total protein half-time (t1/2, total) should scale with changing cell cycle times. In the scaled model, t1/2, total for each protein linearly increases with cell cycle time (Fig. 6B). In contrast, in the constant model, the dilution rate dominates for rapidly dividing cells while the contribution from active degradation becomes more relevant for slower dividing cells.To test the models’ predictions, we grew E. coli cells with a range of doubling times, including rapidly doubling cells with unlimited growth in minimal medium (0.7 h) and slower doubling cells in carbon-limited chemostats (3 h, 6 h, and 12 h). We found that the data favored the constant model regardless of which cell cycle times we compared (Fig. 6C). This indicates that active protein degradation rates typically remain constant regardless of cell division rates.We noted interesting exceptions to the model (Supplementary Fig. 8), however, particularly when comparing slower-growing cells in the chemostat to cells growing without nutrient limitation. For example, RpoS degrades faster in unlimited growth conditions than the non-scaled model would predict based on chemostat measurements. This is consistent with the previous finding that RpoS is rapidly degraded in exponentially growing cells but becomes stabilized when nutrient-limited84.Because kactive rates are generally constant across cell division rates, we can more accurately measure kactive when kdilution is small. Importantly, the observed constancy of degradation rates, regardless of cell cycle times, allows us to extrapolate active degradation rates from conditions with long cell division times (e.g., chemostats with 6-h doubling times) to conditions with more rapid cell division times, in which separating between active degradation rates and dilution rates is experimentally difficult. Thus, the protein half-lives in this manuscript, primarily obtained in the chemostat, are a valuable resource that can be extrapolated to arbitrary cell division rates.
Global protein turnover quantification in Escherichia coli reveals cytoplasmic recycling under nitrogen limitation
