Protein degradation kinetics measured by microinjection and live-cell fluorescence microscopy

The fundamental discovery in the 1930s¹ that a protein’s intracellular concentration is based on a continuous interplay between synthesis and degradation resulted in an ever-growing interest in methods to measure the concentration of a given protein as a function of time. Progressively, an array of ingenious approaches was developed^2,3, paving the way for insights into the fundamental processes of cellular protein degradation, present in all kingdoms of life. Eventually, this has led to the discovery of the eukaryotic ubiquitin-proteasome system⁴, whose modulation represents the basis of novel therapeutic approaches, including proteasome inhibition and selective induction of protein degradation using proteolysis-targeting chimeras (PROTACs)⁵.

Several experimental approaches to quantify the degradation kinetics of a protein in the living cell have been developed based on detection of the proteins of interest by labels (radioactive, mass, or fluorescent), or specific binding (e.g. antibodies)³. Nonetheless, all these methods need to express the labeled protein or introduce it into the cell, and thereby entangle the kinetics of biosynthesis or cellular uptake with degradation. While pulse-chase approaches can mitigate some of these issues, all of these methods observe a population of cells.

To bring the protein of interest into the cytosol, either recombinant expression or methods to translocate through the plasma membrane (and/or endosomal membrane) are needed. Both strategies typically suffer from having additional slow kinetic components, including when using viral delivery systems, making the observation of “pure” degradation kinetics not straightforward. Several translocation strategies⁶ across the membrane have been described, including (a) passive membrane diffusion by alteration of physicochemical properties; (b) encapsulation using nanoparticles or liposomes and subsequent membrane fusion; (c) engagement of biological transport mechanisms such as bacterial toxins, or pore-forming particles, (d) mechanical membrane perturbation including electroporation, sonoporation, shear stress, and lastly, microinjection.

We set out to develop a method that allows to follow degradation kinetics at the single-cell level, at the same time assessing concentration and distribution of the protein of interest, and that has a clearly defined starting time of degradation. Microinjection has the unique property of being completely independent of physicochemical properties, as long as solubility is given. Furthermore, it is the only method allowing for simultaneous delivery of multiple entities at a specific ratio, which also serves as the basis for precise determination of injection volume.

The predecessor of today’s impressive number of microinjection-based methods⁷ was first published in 1911⁸. An inherent characteristic of it is a large cell-to-cell variability of injection volume. This is minimized, but still encountered in today’s commercially available devices, despite technological advances including robotic micromanipulation, precise pressure systems, as well as specialized mass-produced glass pipettes. While a limitation in many contexts, the resulting wide range of intracellular concentrations can also be useful, provided it is combined with injection volume determination, which can be quantified by correlating the fluorescence output, and thus the estimated volume, using microscopy. We have established a slightly modified approach involving 3D-reconstruction using confocal microscopy, to also obtain insights into the concentration-dependency of degradation rates. Moreover, continuous microscopy permits to simultaneously observe individual cells for their morphology, cell-cycle state, viability, and subcellular localization of the protein of interest, which can then be correlated with total cell fluorescence as a reporter for the amount of protein present.

To establish a robust assay for the determination of degradation rates on the single-cell level using microinjection and live-cell microscopy, we first used an in-depth analysis of the characteristics of eGFP degradation as a proof-of-principle analyte. As explained below, while native eGFP is almost not degraded, the addition of two N-terminal amino acids, Gly-Ser, made it a very convenient model substrate (termed from now on GS-eGFP). We further assessed multiple fluorescent proteins for their suitability as genetically fused reporters.

To quantify non-fluorescent proteins, clearly of wider interest, we established a labeling strategy and compared different tags and dyes. We hypothesized that a dye attached to the protein of interest, where the conjugate can be introduced to the cell upon microinjection, can faithfully monitor this protein’s degradation, provided the dye leaves the cell more rapidly than the actual degradation and is thus not rate-limiting for fluorescence loss. For this purpose, we assessed cellular export rates of different dyes and their dependency on lipophilicity. Upon confirming that site-specific fluorescence labeling can be used to quantify the half-life of GS-eGFP, which shows an average of 7 h, we further challenged our system by accelerating protein turnover to yield average half-lives of under 1.5 h. This was accomplished by using an in-house developed bioPROTAC technology⁹, which selectively binds a target protein, as well as components of the ubiquitin-proteasome system (UPS), to eventually induce ubiquitination and rapid subsequent proteasomal degradation.

We further validated the technologies developed here to re-evaluate the N-terminal amino acid’s role¹⁰ and its acetylation with high precision and minimal cellular perturbance. For this purpose, we have established protocols for image acquisition and analysis for state-of-the-art fluorescence live-cell confocal and wide-field microscopy, allowing for the observation of fluorescent entities, quantitatively introduced to the cytosol using microinjection. We report here on the possibilities and limitations of this method in conjunction with cytosolic delivery. Lastly, to further validate our results, our method was compared with two orthogonal methods to quantify protein degradation rates of the same two proteins, and we found good agreements.

Manual microinjection procedure

To trigger the degradation process at a defined concentration of the protein of interest and a defined time point, we used microinjection for cytosolic delivery. The widely used semi-automatic approach was replaced with a simpler, manual procedure, as visualized in Supplementary Fig. S1. This strategy was chosen due to the semi-automated method’s dependence on a correctly set vertical distance between capillary and cell ___location (Z-value), which frequently required time-consuming prior adjustments, caused by slight inclinations of microscope, dish, and microcontroller. Major advantages of the manual method included manual control of the micromanipulator in all three axes and constant analyte outflow, as opposed to pressure pulses encountered in semi-automatic procedures upon membrane penetration. Manual injection started with confirmation of constant outflow in newly mounted capillaries, visually determined by observing temporary contrast changes surrounding the tip after a high-pressure pulse. In most cases, breakage of the capillary to increase or change the shape of the opening was necessary for efficient injection. This was achieved by a sliding motion along the glass surface until fragments were visible. Capillaries were then placed above target cells, followed by lowering along the Z-axis. At the moment of membrane penetration, a distinct temporary contrast change was observed within the cell, upon which retraction to the starting position was performed, as shown in Supplementary Video 1.

Microinjection speed and cell viability

Injection time per cell was generally between 1 and 5 s, but additional time was necessary for capillary filling and mounting, as well as for initiation of image acquisition. For usual experiments, involving approximately 200 injected cells and three analytes, this unobserved time period lasted around 30 min, but could be reduced to 5 min by using only a single analyte and a reduced number of cells. This period was therefore unique for each cell and was determined by recording the injection time with an accuracy of ± 1 min, combined with programmatic extraction of image acquisition at a resolution of ± 1 s.

Amongst successfully microinjected cells, 75% (s.d. = 14%, n = 631) were available for single-cell analysis after 12-h observation, as evaluated by comparing data from over 30 independent experiments. Excluded cells were generally comprised of approximately 3% of cells undergoing mitosis, 15% exhibiting altered motility and shape, and 7% detaching from the cell culture dish. It should be noted that a high stringency was employed in assessing cell viability, thereby rather erring on the side of excluding viable cells than including cells incorrectly classified as viable. Importantly, these quality controls show that the cells do not undergo significant stress during the time of the experiment, neither by physiological effects on the cells nor by the imaging itself (see below). This was supported by experiments with longer duration of up to 44 h, as shown in Supplementary Fig. S2 and Supplementary Video 2.

Fluorescence live-cell microscopy

Microinjected fluorescent analytes were observed by using a combination of optical, confocal and wide-field fluorescence live-cell microscopy to simultaneously obtain temporal information on subcellular localization, as well as total cell fluorescence (TCF), and were recorded for 12 h in 20-min intervals. Illumination settings were constant throughout all experiments, allowing for quantitative comparison in three fluorescence excitation channels (488, 561, 640 nm). Fluorescently labeled dextran with an average molecular weight of 10 kDa was co-injected as an injection marker. Its expected homogenous distribution in the cell interior was confirmed using confocal microscopy as shown in Fig. 1a. The quantitative nature of our image analysis, including flat- and dark-field corrections, revealed a complete absence of background fluorescence in all channels. This was expected due to the high dilution of any potential material outside the cells, resulting from the deliberately chosen high volume of surrounding medium (2 ml). The method’s high contrast between injected and non-injected cells allowed for the detection of low-intensity, out-of-focus signal, which manifested itself as blurry regions, often appearing to be outside the physical cell boundaries in recorded images. This additional signal change occurs as a result of changes in the cell’s height, shape, and ___location over time. To include such signal, we have used a low signal threshold of 0.1% of the maximum intensity (pixels with values between 80 and 65535 were included). To determine whether out-of-focus fluorescence is correctly recorded, TCF was recorded with focal planes in the range of 20 μm above and below the dish surface. As summarized in Fig. 2a and b, only minor changes were observed within the relevant vertical range, thus showing that, even if cells move out of focus, recorded images can be used for analysis.

Microinjection workflow with GS-eGFP as an exemplary analyte and AF647-labeled dextran (dexAF647) as an injection marker. (a) Microscopic observation using light and fluorescence microscopy at the beginning and end of a 12-hour experiment, displaying total cell fluorescence intensity reduction over time. Pixels included in further analysis are bordered by a yellow line, and they are identified and calculated from injection marker epifluorescence micrographs for each timepoint. This boundary detection is based on a low signal threshold of 0.1% of maximal signal, thereby ensuring the inclusion of out-of-focus fluorescence. (b) Total cell fluorescence (TCF) data over time, fitted with an exponential decay model of the cell shown in (a). The window between injection and first acquired image is highly dependent on the number of cells injected, and can be reduced to approximately 5 min. (c) Distribution of degradation rates from 3612 cells, overlaid with a cumulative density function of a normal distribution (black line). (d) Bar lengths indicate degradation rate means for GS-eGFP (green), dexAF647 (red) or dexCB (cascade blue-labeled dextran, blue). Error bars indicate the standard deviation. MG132 proteasome inhibitor or E1-enzyme inhibitor TAK-243 were added 1 h before microinjection where indicated. Dye refers to dexAF647 or dexCB (cascade blue-labeled dextran). (e) Histogram displaying the normalized distribution of injection numbers versus injection volumes, calculated from extrapolated injection marker TCF signal from 17,129 cells. A black line indicates the cumulative distribution function after fitting to a Weibull distribution with parameters indicated. (f) Comparison between the injection volume and degradation rate constant for 3612 cells, co-injected with GS-eGFP and injection marker.

Full size image

Determination of injection volume using 3D-reconstruction of confocal micrographs in combination with total cell fluorescence obtained using wide-field microscopy. (a) A droplet containing labeled dextran in PBS, created by microinjection into oil. The droplet’s half-sphere shape is the result of placing the injection capillary onto the dish’s glass surface during injection. This method is useful for long-term observation due to its immobilization. (b) Total integrated fluorescence density as a function of the ___location of the focal plane in both observed channels. The same region of interest was used for calculation in all micrographs. A vertical range of 0–10 μm relative to the glass surface is indicated in grey and indicates the range in which cells are usually observed, independent of changes in morphology. (c) Exemplary 3D-reconstruction of a HEK293 cell, whose shape is influenced by neighboring non-fluorescent cells.

Full size image

Degradation rate determination and distribution of GS-eGFP and injection marker

As seen in an exemplary cell, co-injected with purified GS-eGFP in Fig. 1a and b, an exponential decay model described signal alterations of both analytes well. Curve fitting, and subsequent extraction of degradation rates for individual cells, allowed for population-based analysis using averaged single-cell rates, as seen in bar graphs presented in Fig. 1d. Cells were identified as outliers and removed based on degradation rates lying outside the range of ± 1.5 times the interquartile range. Generally, none or fewer than 5 cells were excluded. Their normal distribution, as displayed in Fig. 1c, justified the use of Student’s t-test for hypothesis testing, which is shown for all analytes mentioned herein in Supplementary Figs. S3-S9.

The opportunity to assess rate distributions, obtained from measuring individual cells, can be used to identify the presence of subpopulations. They may arise for various reasons, for example degradation rates being dependent on cell cycle stages. In an unsynchronized culture, such as ours, the percentage of cells in a specific cell cycle stage at any given time depends on the duration of each stage, which is generally considered to have low variation between common human cell lines (median G1 39%, S 37%, G2 17%, M 6%)¹¹. We note that we observe a normal distribution of rates, and not distinctly separated distributions. This tentatively suggests that the rates between the cell cycle stages are somewhat similar. We can furthermore state that the rate does not correlate with injection volume (Fig. 1f) (see section below). To pinpoint the cause of this distribution of rates will require further investigations correlating them with other parameters, such as cell shape, motility, cell volume, the concentration variation of potentially limiting components of the degradation machinery or regulatory factors, and the number of cell-to-cell contacts, which is beyond the scope of the current manuscript. We furthermore suggest that this spread has often been overlooked due to the limited single-cell resolution in most experimental approaches measuring protein degradation.

To exclude the possible role of photobleaching as a cause of GS-eGFP signal loss, we reduced the number of image acquisition timepoints from 37 to 2, which were recorded at the beginning and end of the experiment. The observed signal reduction of 33% (s.d. = 10%, n = 47) was in a similar range as determined in experiments with 18-fold longer exposure, suggesting that photon exposure is unlikely contributing to the observed signal changes. This was expected, given the microscope configuration, where laser and LED intensities were kept low at approximately 5% of maximum intensity, with an exposure time of 0.15 s, made possible by employment of a sensitive state-of-the-art EMCCD camera and spinning disk technology.

Unexpectedly, AlexaFluor647-coupled dextran (dexAF647) showed a TCF signal reduction of 45% within 12 h (s.d. = 17%, med = 17%, n = 17,975), independent of the presence of other analytes. While this behavior was expected for protein analytes, it was surprising for dexAF647, as its fluorescent properties are derived from a small molecule dye covalently bound to dextran. As depicted in Fig. 1d, no signal reduction was observed for dextran coupled to cascade blue dye (dexCB), nor for GS-eGFP in the presence of proteasome inhibitor MG132. These findings suggest that this unexplained behavior is specific for dexAF647, and that the photon exposure used in our experiments is unlikely to induce stress-influenced alterations of degradation rates.

Possible explanations for the effects observed with dexAF647 include the presence of tightly bound dye not conjugated to dextran in the commercial product, and/or subsequent cleavage within the cell, with the dye then being released from the cell, or participating in cell metabolism.

Importantly, this finding does not rule out the use of dexAF647 or other dyes exhibiting similar signal reduction as injection markers, since in our analysis we implemented back-extrapolation to calculate the signal intensity at the timepoint of injection, as indicated in Fig. 1b. This has allowed us to correct for the variable time difference between injection and the first recorded image on the single-cell level. This method also formed the basis for accurate injection volume determination, regardless of the behavior of such fluorescent markers, as discussed in a dedicated section.

GS-eGFP fluorescence changes are exclusively caused by ubiquitination and subsequent proteasomal degradation

As seen in Fig. 1d, we observed an approximate 7-h half-life of GS-eGFP, which was not altered by the presence of dextran. Engagement of the ubiquitin-proteasome system as the sole cause for signal reduction was individually confirmed by proteasome and E1 enzyme inhibition (MG132 and TAK-243), resulting in a constant signal over 12 h. This validated the loss of fluorescence¹² being caused by proteasomal unfolding and proteolysis upon ubiquitin chain attachment, thereby qualifying GS-eGFP as a candidate for studies on induced turnover changes. Moreover, this observation of constant signal supported a lack of photon-induced bleaching or other detectable changes of the chromophore in GS-eGFP, as well as the completion of chromophore maturation during protein production in E. coli and subsequent purification, i.e., before injection. In order to exclude possible influences on degradation rates from co-injected impurities, we have established an extensive purification process, involving four chromatography steps, as well purification tag removal, which is summarized in Supplementary Fig. S10. This procedure for protein purification was used for all protein-based analytes mentioned herein. Quality control (QC) generally involved mass spectrometry and SDS-PAGE, which is summarized in Supplementary Fig. S11-17 for all samples discussed in the present work. QC for GS-eGFP is shown in Supplementary Fig. S14. Degradation rate differences between eGFP (i.e., not containing the N-terminal Gly-Ser) and the variant GS-eGFP used here (see sequence in Supplementary Fig. S18b), and the influence of the altered N-terminal sequence in general is discussed later.

Injection volume

Determination, and possibly optimization, of the injection volume is necessary to ensure intracellular analyte concentrations that are high enough for protein-protein interactions to occur, but at the same time low enough to limit changes in cell physiology caused by possible alteration of pH, osmolarity, cell volume, and introduction of contaminants. The calculation of actual injection volume was based on correlating the total fluorescence and 3D-reconstructed volume of droplets containing non-diluted injection marker solution, formed upon microinjection into oil¹³. With this as a reference, the injection volume could be directly calculated from total cell fluorescence. Placing the injection needle onto the glass dish during injection allows for droplet immobilization, which can be convenient for time-course experiments, as shown in Fig. 2a. To take into account possible signal changes before the acquisition of the first image, back-extrapolation to the time of injection was performed using parameters determined by fitting to an exponential decay model, as exemplified in Fig. 1b. A global analysis across a large number of experiments revealed a Weibull distribution with a most frequently injected amount of 25 μm³ and a 90% chance of being in the range of 8 to 155 μm³, as shown in Fig. 1e. Considering the published HEK293 cell volume¹⁴ of 1400 μm³, this represents an increase in cell volume between 0.5 and 11% (most frequently 1.7%, which represents a dilution factor of 56). As all protein samples were adjusted to a protein concentration of 25 µM in PBS prior to injection, the most frequent intracellular analyte concentration was 450 nM, with 90% of injections resulting in concentrations between 134 and 2752 nM. This large variation, while at first undesired, allowed for additional insights into concentration-dependencies of signal changes, showing that the rates measured for GS-eGFP and dexAF647 were not influenced within the tested concentration range, as shown in Fig. 1f. The ability to assess absolute intracellular concentrations, as opposed to mere relative injection volume differences, is particularly useful when multiple interacting molecules are co-injected while in complex, such as is the case for designed protein binders or small molecules in complex with their target proteins.

Fluorescent proteins show large differences in degradation rate and subcellular localization

To quantify degradation rates using our approach, the introduction of fluorescent properties is necessary. A commonly used method is the genetic fusion to fluorescent proteins. Many variants with different spectral properties usually share a common beta-barrel fold, originally found in GFP¹⁵. Despite their molecular weight in the range of 30 kDa, which is often larger than their fusion partner, they are often considered bystanders. Multiple examples, however, indicate influences on experimental outcomes^{16,17,18,19,20,21}.

We therefore investigated the degradation and subcellular localization differences of a range of fluorescent proteins, and subsequently assessed whether their characteristics are retained when genetically fused to other proteins. For this purpose, we compared GS-eGFP to derivatives of other evolutionarily distinct proteins, including mRuby3²² (GS-mR3), mNeonGreen²³ (GS-mNG), which are more distant in sequence, as well as two closely related GFP variants, superfolder GFP²⁴ (GS-sfGFP) and mClover3²² (GS-mC3) (see sequences in Supplementary Fig. S18a, b and c). Both groups share the same nine N-terminal amino acids. Protein production was performed as described in Supplementary Fig. S10, with protein integrity and purity analysis displayed in Supplementary Fig. S11.

As shown in Fig. 3b, amongst fluorescent proteins of the GFP-lineage, superfolder GFP (GS-sfGFP) showed slower degradation compared to GS-eGFP, whereas GS-mC3 had the same degradation rate, despite 16 mutations compared to GS-eGFP (see alignments in Supplementary Fig. S18b and c). GS-sfGFP has been specifically designed for improved folding kinetics, and yields^24,25, which possibly influences intracellular turnover rates, especially with regard to altered proteasomal unfolding efficiency²⁶.

Degradation rates and subcellular localization of fluorescent proteins and the influence of tags. (a) Structure of eGFP³¹ (PDB 5MA3), commonly referred to as Clontech-eGFP, with native N- and C-termini (MVSK… and …LYK, respectively, see sequence in Supplementary Fig. S18b) indicated as blue and red spheres. Lysines, representing possible ubiquitination sites are shown in blue. Tags investigated are listed next to the structure with names and sequences. (b-h) Bars indicate the mean degradation rate constants from fitting an exponential decay model on the single-cell level, with error bars showing the standard deviation. Pairwise statistical analysis of differences between rate averages was performed using Student’s t-test, as shown in Supplementary Fig. S3-S7. Additional MG132 controls are shown in Supplementary Fig. S22. The purple bar refers to the reference construct GS-eGFP (starting with GSMVSK). Since the y-axes of the panels are different, a purple dashed line indicates this degradation rate. (b) Comparison of fluorescent proteins, starting either with glycine-serine (group 1), or preceded by His₈-biotinylated Avi-tag (group 2, with biotinylation being indicated by red star). Successful purification was confirmed as shown in Supplementary Fig. S11 and 14. (c) Influence of tags. Samples are denoted as shown in a, with sample quality represented in Supplementary Fig. S12. (d) Tags introducing free cysteines for subsequent maleimide-thiol conjugation with or without the conjugated dyes TMR, Cy5-2 and Cy5-4³². Quality control is shown in Supplementary Fig. S13. (e) Influence of the N-terminal amino acid X on eGFP degradation, where the sequence starts with XSMVSK. The variants were produced by TEV cleavage in vitro before injection. Quality control is represented in Supplementary Fig. S14. (f) Influence of N-terminal acetylation on XS-eGFP variants. The presence of the N-terminal acetyl group was verified using mass spectrometry (Supplementary Fig. S15). (g) Degradation rate analysis of eGFP without N-terminal GSM, and without initiator methionine, thus starting with VSK (termed [-M]-eGFP). Proteasomal inhibition was performed using MG132. The rightmost [-M]-eGFP sample was expressed and purified from the cytosol of HEK293 cells using anion-exchange chromatography, while all others were purified from the cytosol of E. coli. (h) Co-injection of degradation-inducing bioPROTAC (see main text) with indicated samples. (i) Subcellular localization of fluorescent proteins, determined by fluorescence confocal microscopy two hours after injection, showing homogenous signal distribution, except for untagged GS-mNG and GS-mR3. Time-resolved export of GS-mNG upon microinjection is shown in accompanying Supplementary Figs. S19 and S20.

Full size image

Both GS-mNG and GS-mR3 showed significantly longer half-lives of almost 40 h, surpassing even the slow degradation rate of GS-sfGFP. Surprisingly, this correlated with a gradual loss of nuclear fluorescence signal, which was not observed for other eGFP-derived fluorescent proteins (see Fig. 3i). While a homogenous signal distribution across the entire cell was seen immediately after microinjection, only the nuclear signal continuously reduced, whereas the cytosolic signal remained constant. An exemplary cell injected with mNG is shown in Supplementary Fig. S19. Nuclear signal reduction followed an exponential decay model, which allowed for a half-life estimation of 70 min, as shown in Supplementary Fig. S20. While an immediate transport into the nucleus upon cytosolic microinjection is expected due to the small size of both fluorescent protein and injection marker, in principle, an increased nuclear signal could also be resulting from nuclear membrane disruption caused by microinjection. To exclude this possibility, we assessed nucleus integrity by cytosolic injection of labeled dextran that has an average molecular weight of 155 kDa, and is therefore expected not to pass the nuclear membrane and be retained in the cytosol, provided the nuclear membrane is intact. This was indeed confirmed, as no signal was observed in this case in the nucleus (see Supplementary Fig. S21).

These insights, together with the observed small change of total cell fluorescence over time, supports the notion that active nuclear export may be involved, which might be explained by so far unrecognized nuclear export sequences or domains.

Interestingly, all observed differences between fluorescent proteins in degradation rate and subcellular localization were abrogated when genetically fused to the biotinylated N-His₈-Avi tag, an unstructured 35 residue-long peptide, as shown in confocal imagery in Fig. 3i. The tag’s primary sequence is displayed in Fig. 3a, and successful purification of constructs in Supplementary Fig. S11e-h and k. Upon tag-fusion, GS-eGFP turnover was increased, while it was decreased for GS-mC3, GS-mNG and GS-mR3, finally resulting in equivalently low half-lives of approx. 14 h (rates are shown in Fig. 3b). This was unexpected, as unstructured regions are prone to increase turnover by potentially serving as E3-binding or additional ubiquitination sites²⁷.

The tag’s impact on degradation rates cannot solely be attributed to the engagement of different N-degron pathways, however, that would result from a different N-terminal amino acid (initiator methionine instead of glycine): When comparing the N-His₈-Avi-TEV fusion to just a N-His₆ fusion, which start with the same amino acids (see sequences in Fig. 3a and degradation rates in Fig. 3c), a lower degradation rate is found for the longer tag, which carries both charged and large hydrophobic amino acids in the intervening sequencethat have been shown to generally influence degradation rates in the human proteome^28,29.

The relevance of conjugated biotin was assessed by comparing GS-eGFP turnover in the presence of the N-His₈-Avi tag, with or without conjugated biotin, which resulted in equivalent degradation rates. This was evaluated due to the possible presence of biotin-binding proteins in the cytosol³⁰.

It appears that the elongated unstructured peptide tag may inhibit nuclear export by an as yet undefined mechanism, while in its absence features on the barrel surface cause the difference between mNG and mR3 and others (cf. structural alignment in Supplementary Fig. S18a). We can currently not ascertain whether the different nuclear export properties and degradation rates are causally related or whether similar recognition properties for both processes play a role.

N- and C-terminal genetic fusion to protein purification and detection tags impacts degradation rates, independent of known N-degron pathways.

The strong influence of the N-terminal biotinylated N-His₈-Avi fusion prompted additional experiments to characterize other commonly used purification and detection tags, as well as a set of linkers containing single-cysteines for subsequent site-specific labeling. We genetically fused C-terminal FLAG, V5, hemagglutinin (HA), and single cysteine-containing tags to GS-eGFP (see sequences in Fig. 3a). Successful production of tagged GS-eGFP variants was confirmed as described in Supplementary Fig. S12.

C-terminally fused FLAG, V5, and HA tags significantly increased turnover rates, with HA and V5 tags exhibiting the most pronounced effect by reducing GS-eGFP half-life from 7 to 3.5 (HA) and 4.3 (V5) hours, respectively, and the FLAG tag causing a slightly lower decrease from 7 to 5.6 h. The addition of linkers consisting of multiple glycine and serine residues, as well as a single cysteine, generally showed no (C-C-short) or only a slightly reduced degradation rate (C-C-long). The latter altered the half-life of GS-eGFP from 7 to 8.9 h.

A common feature among the tags that increased turnover in the C-terminal position is the presence of both hydrophobic and negatively charged residues, albeit in different ratios. Recent studies focusing on the influence of grafting C-termini of the human proteome to eGFP found a lower abundance of eGFP if tagged with hydrophobic linkers, and higher amounts when using linkers that contained charged amino acids²⁹. These insights were obtained using global protein stability (GPS)-profiling³³, where protein turnover is estimated by comparing the abundances of eGFP and a reference fluorescent protein, DsRed. Our observed correlation of increased degradation with hydrophobicity (HA > V5 > FLAG) (Fig. 3c, d) is therefore generally in agreement with these findings, but the increased stability of a neutral linker, such as the single cysteine-containing tags tested in our experiments, was not predicted by the suggested C-terminal rules. The expected reduction in degradation rates associated with a C-terminally located FLAG tag, as mentioned by the authors, could therefore not be confirmed (Fig. 3c). While the lack of influence of C-terminal V5 addition was reported before², it seems likely that the discrepancy between our findings and the authors’ results might be due to the more limited sensitivity of the western blot-based assays used in the reported assays².

Changing the ___location of the FLAG tag from the C-terminus, where it had almost no effect, to the N-terminus of GS-eGFP, drastically decreased GS-eGFP half-life from 7 to 1 h, thereby representing one of the strongest influences reported in our study. It was surpassed only by the addition of a dipeptide (Phe-Ser) N-terminal linker (or by induced degradation with a bioPROTAC, as discussed further below). This finding that the N-terminal FLAG tag has an accelerating effect on degradation of GS-eGFP is in stark contrast to the reported strongly reduced degradation by the N-terminal addition of a three-tandem FLAG tag on protein Parkinson-related DJ-1 carrying the mutation L166P². A key difference is that this mutation strongly hampers the intrinsic folding propensity of DJ-1³⁴, while GS-eGFP is very stable.

Another difference to our experiments, however, is the likely presence of a methionine preceding the tag, as the following Asp residue is expected to inhibit cleavage by methionine amino peptidases during translation^10,35, while our constructs are obtained from TEV cleavage and start with the indicated amino acid. Reduced degradation, if Met is located as the first amino acid, is predicted by GPS-profiling experiments, similar to those mentioned above, where N-termini of the human proteome were genetically fused to eGFP²⁸. According to their findings, both possible N-termini (MDY or DY) would be predicted to decrease turnover.

The stark increase in degradation by FLAG tag addition observed in our experiments can therefore most likely be explained by an undescribed, potent, N-terminally located degron, which is different from the actual N-terminal amino acid Asp. Our later discussed experiments involving short dipeptide N-terminal extensions (replacing the Gly of GS-GFP e.g. by Asp) with varying first amino acids, showed that Asp as the N-terminal amino acid is not per se accelerating degradation: The reason is that the Arg/N-degron pathways, requiring deamination and arginylation, must be engaged during translation, but are not involved when the mature, folded protein is introduced to the cytosol by microinjection.

Transfer of the above-mentioned Cys-containing linkers for later site-specific conjugation from the C- to the N-terminus of eGFP further decreased protein degradation, altering GS-eGFP half-lives from 7 to 10 (N-C-short) and 19 h (N-C-long). Given that both linkers start with the identical Gly-Cys sequence, we assume that differences are caused by the internal linker containing exclusively glycine and serine residues, which are predicted to decrease degradation rates, as shown in above-mentioned experiments by other investigators using GPS-profiling²⁸. The strong stabilizing effects of N-terminally fused tags (except FLAG) is surprising, given that the first amino acid is glycine, and therefore the same as GS-eGFP. As a result, we opted for the use of C-terminal versions of Cys-containing linkers in our experiments (see below), which do not have a significant effect on degradation rates (Fig. 3d).

Evaluation of degradation rate influences based on the N-terminal amino acid

To examine the consequences of altering only the N-terminal amino acid in the absence of additional properties introduced by different tags, we altered the Gly to other amino acids in the GS-eGFP construct (Fig. 3e), which we can access through TEV cleavage before microinjection (Supplementary Fig. S10). This has allowed us to compare these results with the classic investigations reported by Bachmair et al.³⁶ in 1989 that had investigated the degradation rates of beta-galactosidase in living yeast and lysates of rabbit reticulocytes. In contrast, we are using here mature, folded proteins devoid of post-translational modifications, and the degradation machinery of a human cell at the single-cell level, with known intracellular analyte concentration rates. This further allows for direct testing of the influence of acetylation at the N-terminal α-amino group, as this can be achieved chemically. Of all proteinogenic amino acids used, proline had to be excluded due to its incompatibility with TEV-protease cleavage, and cysteine was omitted due to unwanted post-translational modifications during E. coli production. All degradation rates are shown in Fig. 3e.

N-terminal amino acids leading to degradation through the Arg/N-end pathway can be distinguished based on whether direct interaction (primary N-degrons) with ubiquitin ligases is possible, or whether arginylation (secondary) is necessary (Asp, Glu, Asn, Gln, Cys) or whether, before this step, prior deamination (tertiary) must occur (Asn, Gln), or oxidation in the case of Cys¹⁰. Primary degrons are further grouped according to their binding sites on downstream E3 ligases of UBR-type, and are designated type 1 (Arg, Lys, His) and 2 (Leu, Phe, Tyr, Trp, Ile)¹⁰. Placement of large aliphatic amino acids at the N-terminus, thereby creating type 2 N-degrons, generally showed the expected increase in protein degradation rates, consistent with previous studies (Fig. 3e), with isoleucine ranking amongst the lowest rates, and phenylalanine showing the fastest degradation rate recorded, resulting in a half-life of only 30 min. Of the positively charged type 1 residues, arginine, and to a slightly lower extent lysine, increased degradation, consistent with what was shown before in both yeast and reticulocyte extracts³⁶. Histidine had a low turnover, also as expected. Variations within these two groups of primary N-degrons may arise due to differences in affinities to ubiquitin ligases, as it was found that lysine and histidine bind with lower affinity than arginine to ubiquitin ligase UBR2³⁷, with other investigators reporting further evidence that UBR1 and 2 may have higher affinity to bulky hydrophobic amino acids than to basic ones^38,39.

Primary and secondary N-degrons that require modification before interaction with ubiquitin ligases were amongst the most stable terminal residues observed (Fig. 3e), thereby indicating that both deamination and arginylation did not occur efficiently in living HEK293 cells after cytosolic delivery of the mature folded protein. This may at first be surprising, as increased degradation rates were observed in cell-free expression experiments using reticulocyte extracts, reported by Varshavsky and colleagues³⁶. However, in their case, fusion proteins with N-terminal ubiquitin were expressed and added to a rabbit reticulocyte lysate, where both de-ubiquitination and further processing might have occurred, while in our case, the mature protein was delivered.

The remaining set of N-degrons (Ala, Ser, Thr, Val, Gly) is characterized by amino acids with a smaller side-chain volume. When these smaller residues are present at the second position, they become exposed after co-translational cleavage of the N-terminal methionine, due to the specificity of the exopeptidase¹⁰. Subsequent acetylation is observed in most proteins (and is also possible for Met) and can influence subcellular localization, protein-protein interactions, structural integrity, as well as protein turnover rates⁴⁰. While initially discovered as an inducer of protein degradation, recent insights show that acetylation most frequently stabilizes proteins⁴⁰. In our experiments, involving in-vitro acetylated N-degrons (see confirmation by mass spectrometry in Supplementary Fig. S15), we measured high degradation rates in the unmodified form, whereas strong stabilizing effects were observed upon acetylation, yielding rates only slightly above those observed in the presence of the proteasome inhibitor MG132 (see Fig. 3f in comparison to Supplementary Fig. S22, including proteasomal inhibition). This remaining degradation may be due to inevitable minor amounts of unacetylated N-termini still present. Incomplete acetylation was also observed during efforts to optimize acetylation procedures using the method of Mikami et al.⁴¹. It should be noted that valine in the unacetylated form showed lower degradation rates than other hydrophobic residues, but not to the extent observed in the experimental setup used by Varshavsky and colleagues³⁶. In summary, the possibility of injecting acetylated and non-acetylated forms of the protein allows one to gain additional insight into the influence of this modification — independent of its biosynthetic introduction.

Site-specific conjugation of dyes as reporters for protein degradation

When using microinjection, conjugation to small-molecule dyes becomes a viable alternative to fluorescent protein fusions to study the degradation of non-fluorescent proteins⁴². Important advantages include the possibility to vary the position and number of attachment sites, as well as the wide range of dyes with different biophysical properties, including size, charge, and lipophilicity. Moreover, many strategies for site-specific labeling are available. Small-molecule dyes furthermore often show higher quantum yields, chemical stability, and favorable spectral characteristics.

For changes of total cell fluorescence to occur upon proteasomal degradation, fluorophore-containing fragments, resulting from proteasomal cleavage, need to be exported from the cell interior. Provided that the extracellular fragment concentration remains low, both the export and the protein degradation rate are expected to follow exponential decay kinetics. Changes in total cell fluorescence signal can therefore be described by a model combining both processes, as illustrated in Fig. 4a.

Site-specific coupling to small-molecule chromophores as reporters for protein degradation. (a) A model describing total cell fluorescence (TCF) changes as a result of protein degradation and subsequent cellular export of fragments. P refers to the labeled protein, F to the labeled degradation fragments, k₁ to the degradation rate constant and k₂ to the overall export rate constant. (b) TCF of single cells upon microinjection of dyes containing a maleimide group, quenched before injection with excess β-mercaptoethanol. The black line indicates the averaged degradation rate from an exponential decay model with a non-zero plateau parameter. (c) eGFP degradation was followed simultaneously by its intrinsic fluorescence and by an attached dye (indicated next to the orange arrow). eGFP fluorescence as a function of time is shown with dashed blue lines, while solid orange lines indicate dye signal. Black lines indicate simulated signal changes for a cell with averaged degradation rate (dashed, eGFP fluorescence; solid, respective dye fluorescence). The simple exponential decay model was used, except for eGFP coupled to Cy5-4 and co-injected with a bioPROTAC (see main text), where a free plateau parameter was introduced for fitting. Top row, Cy5-2; middle row, Cy5-4; bottom row, TMR. (d) Correlation between the degradation rate determined by loss of GFP fluorescence and from loss of fluorescence of small-molecule dyes attached to GFP. A perfect correlation would be indicated by the dashed grey line going through zero, while the linear regression line is shown in black. Top row, Cy5-2; middle row, Cy5-4; bottom row, TMR. (e) Simulated TCF data assuming a fixed degradation rate constant of a protein with a 30 min half-life and a range of dye export rates. Top, actual kinetics; bottom, apparent kinetics, if the observed traces are normalized at the first measured point after 0.5 h. (f) The difference between true and fitted simulated data using a first order exponential decay model is plotted against the dye export rates. The result of fitting with the exclusion of the first 30 min after injection is indicated in black, whereas fitting of data from the injection start is shown in grey. (g) Measured dye export rates. (h) Bar heights indicate averaged degradation rate constants based on eGFP (blue, left bar) or dye signal (orange, right bar); “Degr.” denotes bioPROTAC (see main text). (i) Calculated partition coefficient (cLogP) versus observed dye export rate.

Full size image

As lipophilicity is known to be a contributing factor for cell-permeability, we first assessed the cellular export rates of three dyes with different computationally predicted partition coefficients (cLogP), as detailed in Supplementary Fig. S23⁴³. The presence of maleimide functional groups, used for conjugation to cysteines, required incubation with β-mercaptoethanol prior to injection, in order to abolish maleimide reactivity, which could otherwise cause permanent cytosolic retention by possible reaction with cytosolic proteins (data not shown). The observed first-order export kinetics shown in Fig. 4b and g allowed for the estimation of half-lives in the range of 4 (Cy5-2), 13 (TMR) and 22 (Cy5-4) minutes. Likely due to incomplete maleimide-inactivation, the signal reached a plateau slightly above zero, thereby making the use of an exponential decay model with a plateau parameter necessary.

As shown in Fig. 4i, a weak correlation between export rate and cLogP was observed, suggesting that high values can predict membrane-permeability. However, this trend should be interpreted with caution due to the notoriously inaccurate prediction of cLogP values. Nevertheless, the likely difference in lipophilicity, due to the presence of two additional sulfonic acid groups, being the only structural difference between the dyes Cy5-2 and Cy5-4 (see Supplementary Fig. S23), may well be the cause for their contrasting cell permeability. A limitation of dyes with high lipophilicity might be constituted by their partial insolubility in solvents required for site-specific labeling, during protein purification and in the buffers for cellular assays. Thus, depending on the protein to be labeled, as well as the expected degradation rate, the use of dyes with slower export rates might thus still be preferred.

To assess the approximate range of dye export rates necessary for quantification of degradation rates using our experimental approach, we simulated expected signal changes based on the model indicated in Fig. 4a. This model contains a fixed degradation rate, equivalent to a half-life of 20 min, and varying dye export rates. The results of these simulations are illustrated in Fig. 4e. The upper section assumes recording of the first image at the exact moment of microinjection, whereas the lower part emulates a more realistic scenario of a 30-minute delay between injection and the first data acquisition.

The noise in real data makes fitting to the combined model (i.e., simultaneous estimation of both parameters) difficult, due to the low curvature at the beginning of experiments. Nonetheless, with a delayed observation start, assuming a first-order exponential decay model with a much higher export than degradation rate, one can still use this approach to estimate degradation rates. The difference between the true and calculated rate constants is shown in Fig. 4f, where the dye export rate in the single-parameter model was fitted to the data. As is intuitively obvious, the protein degradation rate is generally underestimated when dye export rate decreases, but certain rate combinations can result in overestimation as well. In summary, except in the most extreme scenarios, the loss in fluorescence is a good measure of the actual degradation rate when using chemically labeled proteins.

To confirm that labeled peptides are indeed removed from the cytosol, and the dye faithfully reports proteasomal degradation, we have site-specifically conjugated all three dyes to eGFP at a C-terminal tag containing a single cysteine, denoted C-C-long-eGFP (see sequence in Fig. 3a), with the aim to correlate the loss of intrinsic eGFP fluorescence with signal loss based on labeled fragments. The completeness and stoichiometry of coupling was evaluated using mass spectrometry, and confirmation that only the linker-located cysteine was available for maleimide conjugation was achieved by comparing migration distances in SDS-PAGE after polyethylene glycol coupling (see Supplementary Fig. S13g). The presence of only one band with the expected higher molecular weight confirmed this. Following microinjection of dye-labeled eGFP, we simultaneously recorded signal changes for both dye and eGFP. Single-cell time-resolved data are displayed in Fig. 4c, and calculated rate averages in Fig. 4h. Both channels showed similar results for all three dyes, with only a slight rate reduction, compared to uncoupled C-C-long-eGFP. Experiments involving proteasomal inhibition excluded the presence of non-covalently bound dye, or photo-induced chromophore changes. Here, a slight increase in signal was observed for eGFP coupled to Cy5-2, likely due to minor amounts of initially unformed eGFP chromophore, which is in line with a lack of signal increase in the dye’s channel. This confirms our expectation that all three dyes are equally suitable for determination of the seven-hour half-life of C-C-long-eGFP, as its half-life is much longer compared to the dyes’ cellular export rate.

To assess the dye’s usefulness also for the precise determination of faster degradation rates, we have induced rapid degradation of the same labeled eGFP discussed above by co-injecting a bioPROTAC, designed to selectively induce ubiquitination and subsequent proteasomal degradation of eGFP in the living cell. Briefly, our design is based on an eGFP-binding DARPin⁴⁴, genetically fused to the E3 enzyme CHIP (carboxyl terminus of Hsc70-interacting protein), similar in design as reported by others⁴⁵, and its characterization will be reported in detail elsewhere. The induced short half-life of 1.7 h for C-C-long-eGFP was similarly well detected with all dyes, as shown on the right side of Fig. 4c and h.

Rates obtained from the eGFP (488 nm excitation) and the dye channel (561–640 nm excitation) of individual cells are correlated (Fig. 4d), suggesting that the large rate differences within cells, injected with the same analyte, are likely of biological origin. A perfect correlation would be indicated by a dashed grey line with a 45° angle going through the origin. For the slow rates, the experimental data are in an acceptable agreement with all dyes, and for the faster rates (right column), Cy5-2 appears to give the most robust correlation, possibly because of its fastest export rate. Importantly, a seemingly good correlation is indicated in Fig. 4d for Cy5-4, as a the visible trendline follows a 45° angle. This, however, is caused by fortuitous mutual cancellation of strong outliers, which are visually clearly identified, but mathematically not justified to be excluded. Close inspection of the raw data points, however, reveals a trend close to what was observed for TMR.

To summarize, we have shown that our bioPROTAC-based approach can be used to assess the “degradation rate detection limits” for individual dyes, which was only possible due to the large biological distribution of degradation rates within cells that were injected with the same analyte. This was based on the observation that underestimation of degradation rates happened only above a certain degradation rate cutoff, which was lower for dyes with low lipophilicity and thus low cell permeability. While it was clearly shown that dyes with fast cellular export are generally preferred, we have shown that more slowly exported dyes can still be used, depending on the observed protein’s degradation rate and experimental context. We recommend using our approach to screen more dyes and increase the number of dyes suitable for reporters of protein degradation.

Comparison of protein degradation measurements between orthogonal methods

We next wanted to compare microinjection to existing methods to quantify protein degradation rates. As an example of non-fluorescent proteins, we used two published eGFP-binding DARPins, 3G124 (abbreviated here as DP1) and 3G61 (named DP2)⁴⁴. Sequences are shown in Supplementary Fig. S18d. Fluorophores were either introduced by chemical linkage to the cysteine of a C-terminal linker (sequence shown in Fig. 3a), or by N-terminal fusion to the fluorescent protein mRuby3 (mR3). Purification was performed as described in Supplementary Fig. S10 and confirmation of purity and successful labeling is shown in Supplementary Fig. S16. We compared rate measurements after microinjection with transfection followed by protein synthesis inhibition, and protein detection by using either western blots or luciferase complementation assays at various timepoints⁴⁶. DP1 was identified as more slowly degrading with a half-life of 6 h, while a half-life of only 2 h was measured for DP2, calculated from data fitting to an exponential decay model, as seen in Fig. 5a-e. All methods agreed within the error of the measurements, except that the mR3-fused DARPins were protected from degradation, leading to the same slow rate for both mR3-fused DARPins. Furthermore, these mR3-fusions were not found in the nucleus (Fig. 5g), and thus this intrinsic property of mR3 is bestowed onto the fusion protein, thereby confirming that nuclear exclusion, which was observed for fluorescent proteins alone, was retained when used as a genetically fused fluorescent reporter. The complete inhibition of degradation, as well as the exclusion from the nucleus, therefore disqualifies mR3 from the use as reporter for degradation measurements.

Comparison of methods for degradation rate determination in living HEK293 cells using two representative DARPins (DP1 and DP2, see Supplementary Fig. S18d for sequences). Black lines in a to d show data fitting using an exponential decay model with (a, b) and without a free plateau parameter (c, d). Y-coordinates represent the relative amount of DARPin, normalized to the signal at timepoint zero. X-coordinates represent the time after the addition of cycloheximide (CHX) for western blot (WB, a) and luciferase assay (LUC, b). In microinjection assays using fluorescently labeled DARPin (with TMR, c) and DARPin genetically fused to mRuby3 (mR3, d), the x-axis indicates the time after the first recorded image after microinjection. For WB and LUC assays, data points and error bars shown in grey represent averages and standard deviations of replicates per timepoint. Pairwise statistical analysis for mR3 fusions and TMR-labeled constructs are shown in Supplementary Fig. S9. (f) For WB experiments, three biological replicates, each containing two technical replicates were performed. Bars represent the sum of pixel intensity after background subtraction, normalized to the first timepoint after ribosomal inhibition. Data are further adjusted to the total protein concentration of the samples, which were loaded in identical amounts. Original uncropped blots and regions of interest used for quantification are indicated and further discussed in Supplementary Fig. S24. (e) Extracted degradation rate constants obtained by all four methods are indicated as bars. (g) Fluorescence confocal micrograph for intracellular localization of fluorescently labeled DARPins, obtained six hours after microinjection.

Full size image

Reasons for the differences in degradation rates between DARPins are currently under investigation and will be reported elsewhere. For the slowly degrading DARPin DP1, a slightly faster half-life of 3.4 h was observed using luciferase complementation assays, which is possibly caused by the presence of the 11 additional residues (VSGWRLFKKIS) at the C-terminus that need to complement it with the large luciferase fragment, yet could act as a weak degron. In summary, all methods, except for genetic fusion to fluorescent proteins, allowed for degradation rate estimation of two inherently non-fluorescent proteins, leading to similar values, albeit with lower precision than microinjection, and without single-cell resolution.

Proteasomal unfolding is likely a rate-limiting factor in eGFP degradation

The long eGFP half-life of over 26 h reported by others⁴⁷ is in stark contrast to our findings, where the slowest turnover of all tested N-terminal eGFP variants resulted in a half-life of approximately 7 h. The only difference is the presence of three additional residues at the N-terminus (GSM) in our version (see Supplementary Fig. S18b and UniProt C5MKY7). Our construct is therefore termed “GS-eGFP” and starts with GSMVSK, while the native eGFP primary sequence used in previous studies starts with VSK, since the initiator Met is cleaved off, and we termed it [-M]-eGFP.

As published experiments in the context of degradation are mainly based on the intracellular expression in mammalian cells, we also purified [-M]-eGFP directly from the cytosol of eGFP-expressing HEK293 cells, using anion-exchange chromatography. This sample was then subjected to degradation rate determination after microinjection, which confirmed a lack of degradation within 12 h, thus reproducing other investigators’ observations⁴⁷, as shown in Fig. 3g.

Considering the observed high turnover of VS-eGFP, as well as its drastic reduction upon acetylation (Fig. 3e and f), we hypothesized that [-M]-eGFP might be post-translationally acetylated as well, given that the first two amino acids are equivalent in both protein forms (VSMVSK compared to VSK). To investigate potential acetylation, we used LC-MS-MS to assess the ___location and type of modifications, which indeed confirmed methionine cleavage and N-terminal acetylation, as shown in Supplementary Fig. S25. We detected other deviations from expected peptide masses, but it is likely that some were introduced in the MS work-flow. We therefore also assessed the molecular weight of full-length [-M]-eGFP by performing LC-ESI-MS analysis. No unmodified protein was detected, but two species with positive mass differences of 45 and 117 Da compared to [-M]-eGFP with a formed chromophore were found, as seen in Supplementary Fig. S25b. The smaller one is consistent with acetylation (+ 42 Da) and the larger one is consistent with β-mercaptoethanol.

To clarify the behavior of [-M]-eGFP further, we aimed at producing defined [-M]-eGFP with the only modification being N-terminal acetylation. For this purpose, we employed the E. coli-based production system described in Supplementary Fig. S10. Purity was confirmed as shown in Supplementary Fig. S14, and acetylation in Supplementary Fig. S15. Subsequently determined degradation rates are shown in Fig. 3g. While we anticipated and found a reduced degradation rate for acetylated [-M]-eGFP, we were surprised to find that the unacetylated protein showed similarly slow turnover rates, likely resulting from a compactly folded N-terminus. This could be a result of either a lack of N-degron pathway interaction and/or hindered proteasomal unfolding.

To assess whether the reduced degradation rate of [-M]-eGFP is also observed in other E3-dependent pathways, we have co-injected the above-mentioned degradation-inducing bioPROTAC, and found complete absence of degradation (Fig. 3h). To rule out that this is caused by insufficient ubiquitination, we performed a cell-free ubiquitination assay based on HEK293 lysate, and confirmed equivalent polyubiquitination, as displayed in Supplementary Fig. S25c. These results are therefore consistent with the hypothesis that proteasomal unfolding at the N-terminus is a key rate-limiting factor for eGFP degradation^48,49. This is especially true when ubiquitination is performed by E3 ligases not directly engaging the N-terminus, as is the case for many bioPROTACs. In light of the inaccessible N-terminus of [-M]-eGFP, it is therefore likely that insufficient engagement of N-degron pathways and/or unfolding at the proteasome may both play a role. We should point out that our in vivo results are fully consistent with in vitro studies of proteasomal degradation, examining the degradation of GFP and circularly permuted variants, fused at the N-terminus to either a linear Ub4-degron or the ubiquitin-like ___domain from yeast Rad23⁵⁰. These studies and our in vivo results presented here emphasize the importance of an initial engagement of a linear fragment with the proteasome, achieved here with a handle as short as Gly-Ser.

We have optimized and validated an experimental approach combining microinjection, microscopic observation, and subsequent image analysis, allowing for the precise measurement of protein degradation rates in the living cell, based on fluorescence as a reporter. To use our approach for any protein, we evaluated strategies to introduce chromophores, including genetic fusion to fluorescent proteins and site-specific conjugation to small-molecule dyes. While we have shown that many dyes can likely be used as reliable reporters for protein degradation, we have evaluated their cell-permeability and found that slow export from the cell interior can limit their usage in proteins with high turnover rates. We devised some guidelines for the choice of appropriate dyes.

A limitation is represented by modifications necessary for site-specific labeling, which we have evaluated for N- or C-terminally located linkers that introduce cysteines for subsequent thiol-maleimide conjugation. Additionally, our results indicate that some commonly used purification or detection tags can either in- or decrease protein turnover, and they should thus be removed by proteolysis. A surprising finding for fluorescent proteins was the nuclear exclusion of mNeonGreen and mRuby3, which was also observed when genetically fused to other proteins. The involvement of the N-terminus in the nuclear exclusion, despite having the same sequence as eGFP, was confirmed, as homogenous distribution was observed again when an N-terminal linker was added.

A particular feature of our system, which can be seen as either a strength or a limitation, depending on the research question, is that the degradation of a completely defined molecular species is studied. We have always measured the mass spectra of the injected molecules and therefore know the precise post-translational modifications present, or their absence, e.g. N-terminal acetylation. If the degradation of a heterogeneously modified natural protein is to be studied, our system would have the limitation that it requires that this protein is isolated from its natural source, but even then, it may still provide unique access to defined protein complexes which can be purified before injection.

To test the sensitivity of our system to small changes, we have further quantified rate changes related to the interaction with N-degron pathways, highlighting the protective role of acetylation. Since we inject the form of the protein whose degradation is to be measured, we can access all precursors as well. This would not be possible when expressing the proteins, as well as placing any amino acid in the first position, independent of the co-translational removal of the initiator methionine, and/or subsequent acetylation. We therefore can access a wider range of reactions in the cell, and access rates and mechanisms involved which are likely specific for each protein.

We therefore consider the measurement of degradation by microinjection and microscopy as an important complementation of existing methods. By the very large number of time points taken, it has an unprecedented time resolution. Since there is a defined starting point, which can be accurately determined by back-extrapolation, the rates are unencumbered by kinetics of biosynthesis or transfer across cell membranes or by cellular changes upon inhibition of biosynthesis or medium exchange. As it measures rates in single cells, it can sensitively detect the correlation of degradation rates with other cellular parameters some of which are simultaneously recorded by light microscopy, and this undoubtedly will open up new research questions. The protein to be examined can be added in any desired state and thus forms can be investigated that would normally be co-translationally modified. Protein complexes can be added at predefined ratios. In summary, the measurement of protein degradation by microinjection and microscopy can be used whenever these above features are of critical importance.

Microinjection

48 h prior to injection, HEK293 (ATCC CRL-1573) cells were cultured in DMEM containing heat-inactivated FCS, 1% (v/v) penicillin, and 100 µg/ml streptomycin at 37 °C with 5% CO₂. Medium for microinjection and subsequent live-cell imaging was exchanged 4 h prior to injection against Live Cell Imaging Solution (Invitrogen) containing 10% FCS. The addition of inhibitors was performed one hour before injection start to yield a final concentration of 30 µM for MG132 and 0.1 µM for TAK-243. Glass microinjection cuvettes (Femtotip II, Eppendorf) were filled with 5 µl analyte at a protein concentration of 25 µM and 0.1 mM labeled dextran with an average molecular weight of 10 kDa. Dextran was conjugated to either Cascade Blue (#D1976, Invitrogen) or Alexa Fluor 647 (#D22914, Invitrogen). For experiments requiring only cytosolic labeling, without diffusion into the nucleus, we have used coupling to tetramethylrhodamine isothiocyanate with an average molecular weight of 155 kDa (Sigma, #T1287). For cuvette filling, elongated tips were used (Microloader, Eppendorf). The micromanipulator setup consisted of InjectMan4 and FemtoJet4i (Eppendorf). Injection and live-cell imaging was performed in air at RT and 37 °C, respectively.

Microscopy

A Visitron CSU-W1 microscope was used for image acquisition, consisting of a Nikon Eclipse Ti-E inverted microscope and a spinning disc system Yokogawa W1 with a pinhole size of 50 μm. For phase-contrast microscopy during microinjection, we used a CFI S PlanFluor ELWD objective (Nikon) with 20-fold magnification. For subsequent live-cell imaging, a CFI PlanApo with 100-fold magnification was utilized for both differential interference contrast (DIC) and fluorescence confocal and wide-field microscopy.

Images for all channels were acquired sequentially at each position and timepoint. Wide-field fluorescence microscopy with LED acquisition was used for total cell fluorescence (TIR) calculation, with the following band-pass filter wavelength windows: 485/20, 560/25, and 650/13 nm. Dichroic mirrors reflected light below 504, 582, and 669 nm. Emission band-pass filters included light at the following wavelengths: 521/30, 607/40, and 700/60.

During confocal microscopy, excitation laser wavelengths of 488, 561, and 640 nm were used, and combined with dichroic mirrors reflecting light below 504, 582, and 669 nm, respectively. Emission band-pass filters included the following wavelengths: 525/50, 609/54, 700/75 nm.

We detected no channel cross-talk for all analytes used for both confocal and wide-field microscopy.

Image analysis

Image timepoint extraction, as well as flat-fielding⁵¹ was implemented using Matlab R2020b. Darkfield images were acquired with equivalent settings as during live-cell imaging for each channel, with the only difference being redirection of the light path towards the shaded eye-piece. Flat field images were acquired using cell culture dishes containing either water for DIC, or glass slides containing 7-diethylamino-4-methyl coumarin (Sigma, #D87759-5G, 405 nm channel) at 50 mg/ml in DMSO, fluorescein sodium salt at 100 mg/ml in water (Sigma, #46960-25G-F, 488 nm channel), rose bengal (Sigma #198250-5G, 561 nm channel) at 100 mg/ml in water, or brilliant blue FCD (Sigma, #80717-100MG, 640 nm channel) at 100 mg/ml in H₂O for wide-field and confocal fluorescence microscopy. Approximately 50 images were obtained at different locations within the dish, and the median was calculated. Image segmentation was performed for each cell over time using CellProfiler 2.2.0 with a boundary detection in the injection marker wide-field fluorescence 640 nm channel. Subsequent data analysis of thereby extracted total cell fluorescence, which included data fitting on the single-cell level, outlier removal, and subsequent calculation of average and standard deviations for bar graph representation, was performed using R software, mostly relying on tidyverse packages⁵².

Cloning of expression constructs

Proteins were expressed from a pQE30 (Qiagen) vector with an N-terminal His₈-tag followed by a TEV protease cleavage site (amino acid sequence ENLYFQX, where X is any proteinogenic amino acid, where indicated), which was inserted via EcoRI/BamHI sites. The eGFP, sfGFP, mClover3, mRuby3 and mNeonGreen constructs were inserted via BamHI/HindIII sites. Exchanges of the amino acid directly following the TEV site (XS-eGFP constructs) were created by full-plasmid PCR using overlapping primers with the respective mutation. DARPin constructs were inserted by BamHI/PstI sites. N-terminal tags and linkers containing single-cysteines were inserted by EcoRI/BamHI sites. C-terminal linkers and single cysteines were inserted by PstI/HindIII sites. Gene fragments were synthesized by Twist Bioscience and Integrated DNA Technologies.

Protein expression and purification

Proteins were expressed in the E. coli strain BL21. A graphical summary of the purification method used is shown in Supplementary Fig. S10. Constructs containing a single cysteine were expressed with 0.4 mM IPTG induction overnight at 25 °C. Other constructs were expressed by autoinduction for 24 h, as described elsewhere⁵³. Cells were pelleted and lysed with a French press in buffer containing 50 mM Tris, 500 mM NaCl, 10% glycerol, 0.5 mM EDTA, 0.2 mM 4-(2-aminoethyl) benzene sulfonyl fluoride hydrochloride (AEBSF), 1.4 µM pepstatin-A, 1 µM leupeptin (pH 8). The protein of interest (POI) was purified on Ni-NTA resin with subsequent TEV-digest overnight at 4 °C, followed by reverse IMAC purification on Ni-NTA resin. Anion-exchange chromatography was performed in buffer containing 10 mM Tris, 50 mM NaCl (pH 8) using a MonoQ HR 16/10 column on an ÄKTA Pure chromatography system with a gradient from 50 mM to 500 mM NaCl over 40 CV. A final polishing step and buffer exchange to PBS was performed using a Superdex 75 10/300 GL column (Cytiva) on an ÄKTA Micro system. Samples were frozen at a concentration of approximately 200 µM in liquid nitrogen and stored at − 80 °C.

To produce eGFP constructs starting with different amino acids, they were expressed in the E. coli strain BL21 and contained an N-terminal genetic fusion to a polyhistidine-tag, followed by a TEV cleavage site, encoding differing amino acids in the P1′ position (see above). TEV protease was shown to tolerate and process almost every amino acid in the P1′ position of the recognition motif, as observed by others⁵⁴. We were able to produce 18 XS-eGFP constructs, where X stands for any proteinogenic amino acid except Pro and Cys. PS-eGFP was excluded as it was found that Pro in the P1′ position was found to prevent TEV cleavage⁵⁴, while CS-eGFP could not be produced without N-terminal adduct formation of formaldehyde (data not shown). After expression, constructs were extensively purified by IMAC and anion-exchange chromatography, as described above to ensure high purity for live-cell measurements. Yields showed high variation, as was expected due to cleavage preferences of TEV protease⁵⁴.

N-terminal acetylation

Protein was buffer-exchanged to phosphate buffer (0.1 M, pH 6.6) using a PD25 Minitrap column (Cytiva). Acetic anhydride was added at a molar ratio of 20:1 and the reaction incubated for 3 h at RT with shaking at 550 rpm. The acetylated protein was buffer-exchanged to PBS using a Superdex 75 10/300 GL column on an ÄKTA Micro system. Acetylation was confirmed by ESI-MS and LC-MS-MS.

Fluorescence labeling

200 µl protein containing a C-terminal tag with a cysteine was incubated with 10 mM DTT for 30 min at RT. The buffer was exchanged to degassed PBS containing 1 mM EDTA using a PD25 Minitrap (Cytiva) under a nitrogen atmosphere.

TMR5-maleimide (94506, Sigma-Aldrich), DY-647P1-maleimide or DY-6479-maleimide (Dyomics GmbH) at a concentration of 20 mM in DMSO was added in a 1:1 molar ratio and incubated for 3 h at RT with shaking at 550 rpm. The reaction was thereafter quenched with 100 mM DTT for 10 min. The labeled eGFP was separated from excess dye and unlabeled protein using AEX and buffer-exchanged using a SD 75 10/300 GL column (Cytiva) on an ÄKTA Pure system. The labelled DARPins were separated from excess dye and buffer was exchanged to PBS using a SD 75 10/300 GL column (Cytiva) on an ÄKTA micro system. Labeling was confirmed by ESI-MS.

Mass spectrometry

Mass spectrometry analyses were carried out at the Functional Genomics Center Zurich. For ESI-MS analysis samples were 3-fold diluted with 1% TFA and transferred to an autosampler vial for LC/MS. 7 µl of sample were injected into an Acquity UPLC@ BioResolve-RP-mAb 2.7µ 2.1 × 150 450 A (Waters, USA) column. For separation and elution on an Acquity UPLC station, a gradient of buffer A (0.1% difluoroacetic acid (DFA) in water) and buffer B (0.1% DFA in acetonitrile/ 75% 2-PrOH) at a flow rate 200 µl/min at 60 °C over 10 min was applied. The analysis was performed on a Synapt G2 mass spectrometer directly coupled to the UPLC station. Mass spectra were acquired in the positive-ion mode by scanning the m/z range from 400 to 5000 Da with a scan duration of 1 s and an interscan delay of 0.1 s. The spray voltage was set to 3 kV, the cone voltage to 35 V, and the source temperature to 100 °C. The data were recorded with the MassLynx 4.2 Software (both Waters, UK). Where possible, the recorded m/z data of single peaks were deconvoluted into mass spectra by applying the maximum entropy algorithm MaxEnt1 (MaxLynx) with a resolution of the output mass 0.5 Da per channel and Uniform Gaussian Damage Model at the half height of 0.5 Da.

For MS-MS analysis, proteins were digested by mixing 4 µl of each sample with 46 µl of 10 mM Tris, 2 mM CaCl₂, pH 8.2 buffer and 5 µl chymotrypsin (100 ng/µl in 10 mM HCl). If needed, the pH was adjusted to pH 8. Digestion was performed for 4 h at 37 °C and samples were dried subsequently. They were then dissolved in 20 µl ddH₂O containing 0.1% formic acid and transferred to the autosampler vials for liquid chromatography-mass spectrometry analysis (LCMS/MS). 5 µl were injected on a nanoAcquity UPLC coupled to a Q-Exactive mass spectrometer (Thermo). All MS/MS samples were analyzed using PEAKS Studio (Bioinformatics Solutions, Waterloo, ON Canada; version 10.5, 2020-02-19). PEAKS Studio was searched with a fragment ion mass tolerance of 0.020 Da and a parent ion tolerance of 10.0 ppm. Peptide identifications were accepted if they could be established at greater than 78.0% probability to achieve a false discovery rate (FDR) of less than 0.1% by the Percolator posterior error probability calculation⁵⁵. Protein identifications were accepted if they could be established at greater than 11.0% probability to achieve an FDR less than 1.0% and contained at least 3 identified peptides. Protein probabilities were assigned by the Protein Prophet algorithm⁵⁶.

Transient transfection of HEK293 cells

The vector pcDNA 3.1 containing a CMV promotor was used for transient expression in HEK293 cells. Full-plasmid PCR was performed to remove the PstI site from the pcDNA3.1 vector in the HygR gene. Inserted DNA fragments were ordered from Integrated DNA Technologies (IDT). A construct starting with a NheI cleavage-site followed by a Kozak sequence, a methionine, the DARPin sequence starting with a BamHI cleavage site and ending with a PstI cleavage site, and a stop codon followed by a HindIII cleavage-site was inserted between NheI and HindIII cleavage sites. A construct containing an IRES sequence, followed by a sequence for mRuby3 expression and a stop-codon, was inserted between HindIII and XbaI cleavage-sites.

5 million cells per 75 cm² plate were seeded one day before transfection. HEK293 cells were transfected with pcDNA3.1 constructs to produce the proteins of interest for further analysis. For western blot analysis, cells were prepared in 6-well plates. Reverse transfection was performed as recommended by the manufacturer (Mirus Bio) in Opti-MEM by scaling up reactant volumes from a 96-well plate to a 6-well plate by a conversion factor of 27.43. The DNA-TransIT (Mirus Bio) mixture was added to the cell suspension at 4 × 10⁵ cells/ml. 2 mL of cell mixture per well was added to a 6-well plate. For luciferase assays, cells were prepared in 96-well plates. Reverse transfection with two separate vectors (see below) was performed according to the manufacturer’s recommendations (Mirus Bio) with 4 × 10⁵ cells/ml and 100 µl of cell mixture added per well. Cells were incubated for 48 h before assays.

Cycloheximide chase assay

100 µg/ml cycloheximide (C4859, Sigma-Aldrich) was added to the cells. For western blotting, one well of the transfected 6-well plate was harvested for each timepoint in two-hour intervals. Cells were lysed by adding 150 µl of NP40 in PBS with protease inhibitors after washing with 1 ml of ice-cold PBS. Lysates were vortexed and incubated on ice for 15 min and subsequently cleared by centrifugation at 20,000×g for 20 min. Protein concentrations were determined by BCA assay (A55865, Thermo Scientific). Samples were flash-frozen at − 80 °C.

Western blots

25 µg of protein of each timepoint were loaded onto an 8–16% a gel for SDS-PAGE. As a molecular weight reference, marker Chameleon Duo (#928-60000, LI-COR) was used. Gel electrophoresis was performed for 90 min at 100 V. Western blotting was performed using a Trans Blot Turbo (BioRad). Nitrocellulose membranes (#1704271, BioRad) were rinsed in ddH₂O and dried for 30 min. Blocking was performed for 1 h using 1× casein blocking buffer (Sigma). Primary antibodies (polyclonal anti-DARPin serum [produced in-house] and monoclonal mouse anti-β-actin antibody [#ab8224 Abcam]) were diluted in PBS-T-casein solution and incubated with the membrane for 1 h at RT. Three washing steps of 5 min with PBS-T were performed before incubation of the secondary antibodies (affinity-purified, AlexaFluor680-coupled, goat anti-rabbit IgG [#A21076, Invitrogen] and polyclonal DyLight800-coupled, mouse anti-sheep IgG [#610-645-002, Rockland Immunochemicals Inc.]) in PBS-T-casein. Three washing steps of 5 min with PBS-T were carried out before imaging using an Odyssey^® CLx imaging system (LI-COR). Protein levels were determined using Image Studio Lite version 5.2. Blot image analysis is described further in Supplementary Fig. S24, where raw data are shown.

Luciferase assay

HEK293 cells were transfected with two separate pcDNA3.1 vectors for expression of LgBit and DARPin-HiBit⁵⁷ fusion two days before measurements. The 11-amino-acid HiBit was genetically fused to the C-terminus of each of the DARPins. HEK293 cells were co-transfected with vectors expressing DARPin-HiBit and LgBit. The medium was exchanged to HEPES-buffered DMEM without phenol red (Thermo Fisher Scientific) and Nano-Glo Endurazine Live Cell Substrate (N2570, Promega) was added 2 h before measurement. 100 µg/ml cycloheximide (C4859, Sigma-Aldrich) was added to the cells directly before measurement and luminescence was measured once per hour using a Tecan Spark plate reader.

Generation of eGFP-expressing cell line

For the generation of a stable cell line expressing eGFP under an inducible promoter, the corresponding gene, containing flanking attB sites, was ordered at Twist Bioscience and cloned into pLenti CMV/TO Zeo DEST (Addgene #17294) via gateway cloning. The plasmid pLenti CMV TetR Blast (Addgene #17492) was left unmodified.

To generate lentiviral particles, HEK293T cells were seeded at 4.5 × 10⁶ cells per 10 cm dish (4 total) and transfected 24 h later. Transfection was carried out using 3.5 µg packaging plasmid (pSPAX2), 0.6 µg envelop plasmid (pMD2.G) and 5.9 µg expression plasmid with TransIT293 (Mirus Bio) according to the manufacturer’s instructions. Medium was exchanged to 3.5 ml Opti-MEM + 10% FCS 14 h post-transfection and the supernatant was harvested 38 h and 62 h post-transfection. The supernatant was filtered through a 0.45 μm PES filter and lentiviral particles were concentrated 50-fold by precipitation via Lenti-X Concentrator (Takara Bio) according to the manufacturer’s instructions.

For stable cell line generation, HEK293 cells were first transduced with lentiviral particles carrying the tetracycline repressor and selected with blasticidin (5 µg/ml, 14 days) and subsequently transduced with lentiviral particles carrying the inducible eGFP cassette, followed by Zeocin selection (300 µg/ml, 10 days). In a last step, fluorescence-activated cell sorting (FACS) was used to sort a pool of cells that showed high expression levels of eGFP upon doxycycline induction. In both cases, 60,000 cells per well were seeded in 6-well plates in complete medium containing 8 µg/ml Polybrene and 100 µl of the respective lentivirus. After 30–48 h of transduction, the cells were exposed to selection medium.

CHIP ubiquitination assay

The procedure was adapted from that described previously⁵⁸. Ubiquitination was performed in 10 µl reactions. The reaction mixture consisted of 50 mM Tris (pH 7.5), 50 mM KCl, 0.2 mM DTT, 0.1 µM Ube1 (#SRP6147, Sigma), 2.5 mM ATP, 2.5 mM MgCl₂, 100 µM ubiquitin (#U-100 H, Bio-Techne), 1 µM UbcH5c (#662098, Sigma) and 0.03 nmol POI or POI/eGFP complex. Reactions were incubated for 1 h at 37 °C in a PCR cycler, and afterwards 5 µl of 4× reducing Laemmli buffer was added. Samples were boiled for 10 min at 95 °C, then centrifuged using a tabletop centrifuge. SDS-PAGE was performed using a Mini Gel Tank (Thermo) at 120 V for 70 min with 12% Bolt Bis-Tris Plus gels (#NW00125BOX, Invitrogen) with 1× MOPS running buffer. 5 µl of each sample were loaded. The proteins were transferred to a PVDF membrane (#1704275, BioRad) using the Bio-Rad Trans-Blot Turbo system according to the manufacturer’s manual. The blotting protocol “Mixed-MW” was selected. After blotting, the membrane was blocked for 1 h in 1× casein blocking buffer (Sigma). Incubation with primary antibody was performed in PBS-T (0.05% Tween) and 1× casein blocking buffer at 4 °C on a roller shaker overnight. The membrane was incubated with polyclonal rabbit anti-GFP antibody (#600-401-215, Rockland Immunochemicals) at a dilution of 1:3,000. The blot was subsequently washed 3 times for 5 min with PBS-T. Secondary HRP-coupled goat anti-rabbit IgG (#7074, Cell Signaling Technology) was diluted 1:2,000 in PBS-T (0.05% Tween) and 1× casein blocking buffer, and incubated with the membrane for 1 h at RT on a roller shaker. The blot was washed 3 times for 5 min with PBS-T. HRP substrate (051730, Merck) was added to the membrane and chemiluminescence imaging was performed after 5 min on Viber Fusion FX7 imager.