Quantitative Measurement of Secretory Protein Mistargeting by Proximity Labeling and Parallel Reaction Monitoring

Proximity labeling is a powerful approach for characterizing subcellular proteomes. We recently demonstrated that proximity labeling can be used to identify mistrafficking of secretory proteins, such as occurs during pre-emptive quality control (pre-QC) following endoplasmic reticulum (ER) stress. This assay depends on protein quantification by immunoblotting and densitometry, which is only semi-quantitative and suffers from poor sensitivity. Here, we integrate parallel reaction monitoring mass spectrometry to enable a more quantitative platform for ER import. PRM as opposed to densitometry improves quantification of transthyretin mistargeting while also achieving at least a ten-fold gain in sensitivity. The multiplexing of PRM also enabled us to evaluate a series of normalization approaches, revealing that normalization to auto-labeled APEX2 peroxidase is necessary to account for drug treatment-dependent changes in labeling efficiency. We apply this approach to systematically characterize the relationship between chemical ER stressors and ER pre-QC induction in HEK293T cells. Using dual-FLAG-tagged transthyretin (FLAGTTR) as a model secretory protein, we find that Brefeldin A treatment as well as ER calcium depletion cause pre-QC, while tunicamycin and dithiothreitol do not, indicating ER stress alone is not sufficient. This finding contrasts with the canonical model of pre-QC induction, and establishes the utility of our platform.


INTRODUCTION
Eukaryotic cells depend upon the secretory pathway to properly traffic about one-third of their proteome 1 , including nearly all secreted and plasma membrane proteins. As the first compartment of the secretory pathway, the endoplasmic reticulum (ER) maintains a calcium-rich environment for calcium binding 2 , an oxidative environment for disulfide bond formation 3 , and possesses a unique set of enzymes and chaperones for glycoprotein biogenesis and quality control 4 .
Secretory proteins have evolved to rely on this unique folding environment, and hence if mistargeted, these proteins present a threat to the cytosolic proteostasis 5 .
Multiple checkpoints and quality control steps ensure high fidelity of translocation of secretory proteins 6 . In the presence of ER stress, translocation for some secretory proteins is attenuated, leading to their cytosolic mislocalization. These mistargeted proteins are primarily directed towards degradation 7-10 . This process is termed ER pre-emptive quality control 11,12 (sometimes denoted ER pQC; we use ER pre-QC instead, to avoid confusion with generic protein quality control PQC [13][14][15] [20][21][22][23][24] . We recently demonstrated that proximity labeling is an effective method for identifying mistargeting of secretory proteins 25,26 . In this approach, N-terminal FLAG-tagged APEX2 with a nuclear export signal (NES) 27,28 is expressed and localized in the cytosol ( cyt APEX). Upon initiation of labeling reactions with a 1-min H 2 O 2 pulse, cytoplasmic proteins are biotin-phenol (BP)-labeled, and these BP-labeled proteins can be affinity purified.
Secretory proteins that mistarget and accumulate in the cytosol are labeled and purified as well, and the relative amount of mistargeted protein can be determined by immunoblotting (IB). While this assay allows easy measurement of protein mistargeting under stress, the use of IB introduces several limitations. The relatively limited sensitivity of IB necessitates the use of several million cells for each drug treatment condition. Proximity labeling can hide epitopes, particularly on the most popular affinity tags, such as the FLAG tag DYKDDDDK and the hemagglutinin tag YPYDVPDYA. 29,25 In addition to the availability, specificity, and sensitivity of antibodies, good quality of gel electrophoresis and electrotransfer must be assured for reliable quantification. If assayed proteins fall in a wide range of molecular weights, for the sake of resolution, more gels of different acrylamide/bis-acrylamide compositions should be used. If some assayed proteins share similar molecular weights and the same host animal of the available antibody, another gel or stripping must be considered. Only a few antibodies can be employed per gel, limiting the potential for multiplexed detection. Finally, IB is a time-consuming process.
Targeted mass spectrometry methods like parallel reaction monitoring 30 (PRM) avoid these problems and have some advantages over IB 31 . Peptides can be chosen to avoid those sensitive to peroxidase labeling. During data acquisition, these pre-defined peptides are isolated according to their mass-to-charge (m/z) ratios with a pre-set isolation window. Isolated precursor peptide ions are fragmented to generate product ions and all resulting product ions are analyzed in parallel with a mass analyzer that allows MS 2 full scan (e.g. Orbitrap, time-offlight 32 or linear ion trap 33 ). Just as antibodies must be carefully validated for their specificity for the protein of interest, PRM transitions must be carefully validated for specificity 34 . This is particularly easy for PRM, as multiple product ions of each isolated precursor are analyzed. If fragment ions (conventionally at least three 35 ) associated with each precursor peptide maintain similar chromatographic profiles, constant proportion of fragment ion intensity and the same retention time, they can be considered as bona fide product ions for later quantification. In addition to the high sensitivity and specificity of (tandem) mass spectrometry, PRM also benefits from liquid chromatography in that it allows multiplexing of hundreds of peptides and inferably hundreds of proteins with ease 32 . Quantification of more than 1000 peptides from a single run has now been achieved, using internal standardtriggered scheduling with modern instrumentation. 36 Herein, we integrate PRM mass spectrometry with our assay to quantify mistargeting of the model secretory protein transthyretin (TTR) in response to ER stress by distinct mechanisms (Figure 1). Using one-order-of-magnitude-less sample, we obtain the same quantification results as are seen by IB 25 . We compare multiple normalization factors and demonstrate the necessity to have a control for proximity labeling efficiency. For drug treatments that do not change proximity labeling efficiency, most normalization factors yield the same result. For treatment that changes labeling efficiency, normalization to auto-labeled APEX2 peptides may be the most accurate method. With the PRM assay and proper data normalization, we establish that not all ER stressors induce ER pre-QC in HEK293T cells. Rather, only Brefeldin A (BFA) and sarcoplasmic/endoplasmic reticulum calcium ATPase (SERCA) inhibitors thapsigargin (Tg) and cyclopiazonic acids (CPA) induce ER pre-QC. While tunicamycin (Tm) or 1,4dithiothreitol (DTT) induce ER stress, they do not increase relative FLAG TTR mistargeting in the cytosol. Hence, we show that PRM-based quantification of secretory protein mistargeting can be used to determine the factors responsible for pre-QC in living cells.

Figure 1
Proposed experimental workflow of this assay. cyt APEX and FLAG TTR are transiently transfected into HEK293T cells via calcium phosphate transfection. Cells are reseeded for later drug treatment (16 h). We expect increased differential FLAG TTR mistargeting to be observed under the condition of Sec61 blockade (during mycolactone A/B treatment) or ER pre-QC induction (during ER stress). 30 min before the H 2 O 2 pulse, biotin-phenol (BP) is added. The 1min BP-labeling reaction is quenched by washing cells on ice with quencher solution 3 times.
Cells are then harvested and lysed; cell lysates are brought to the same mass concentration and subjected to affinity purification with avidin agarose beads. Instead of loading eluate samples for SDS-PAGE and IB, we pellet avidin-enriched proteins by MeOH/CHCl 3 precipitation and get the protein pellets washed with MeOH to clean-up, resuspended, reduced, carbamidomethylated and trypsin-digested, followed by acidification and clarification by hard spinning. Eluate digests are then analyzed by parallel reaction monitoring mass spectrometry. Displayed at the right bottom corner are schematic MS 1 precursor ion chromatograms and MS 2 product ion chromatograms of a targeted peptide. Areas under chromatograms are used for quantification.
For electrochemiluminescence, nitro-cellulose membranes were incubated in HRP-conjugated streptavidin (Thermo, 1.25 mg mL -1 , 1:5000 dilution in 1% milk/TBST) for 4 h at room temperature, followed by washing three times with TBST, once with TBS, and once with H 2 O. Membranes were then drained and placed on the image tray. ECL substrate and peroxide (Cytiva) were mixed, applied to the entire membrane and drained before acquisition on the LI-COR Fc Odyssey imager.

Mass spectrometry
Only MS grade organic solvents were used during sample preparation, except chloroform (CHCl 3 , certified ACS). Buffer A is 0.1% formic acid in 5% was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ;https://doi.org/10.1101https://doi.org/10. /2023 μ L 100 mM HEPES, pH 8.0. Proteins were then reduced by 10 mM tris(2carboxyethyl)phosphine (TCEP, Millipore Sigma) for 30 min at 37°C, alkylated by 5 mM iodoacetamide (Millipore Sigma) for 30 min in dark at ambient temperature and digested by trypsin (Thermo Fisher Scientific, final concentration 0.01 μ g μ L -1 ) overnight (16-24 hours) at 37°C with 600-rpm agitation. Tryptic digestion was quenched by adding formic acid (Acros) to pH 2.0. Acidified samples were heated at 37°C for 1 hour and hard spun for 30 min to precipitate Rapigest decomposition products. Clarified samples were transferred to new low-protein-binding tubes.
This process of heating and hard spinning was repeated twice. Samples were stored in freezer ≤ -50°C until analysis.

Sample preparation from protein resuspension using pH-buffered 9 M urea:
Dried protein pellets were resuspended in 9 M urea in 25 mM NH 4 HCO 3 , pH 7.8 (or 50 mM Tris, pH 8.0). Proteins were then reduced by 10 mM TCEP in 200 mM NH 4 HCO 3 , pH 7.8 (or 50 mM Tris pH 8.0) for 30 min at 37°C, alkylated by 10 mM iodoacetamide in 25 mM NH 4 HCO 3 , pH 7.8 (or 50 mM Tris pH 8.0) for 30 min in dark at room temperature. Samples were diluted with 25 mM NH 4 HCO 3 , pH 7.8 (or 50 mM Tris pH 8.0) to ≤ 2 M urea and brought to 1 mM CaCl 2 , before digested by trypsin (Thermo Fisher Scientific, final concentration 0.01 μ g μ L -1 ) overnight (16-24 hours) at 37°C with agitation. For sample volume greater than 60 μ L, we used an orbital shaker at 37°C and 600 rpm. For samples of 20 μ L, we placed those samples inside a 37°C incubator, with 300 rpm agitation, to avoid water evaporation and condensation at the EP tube cap. Tryptic digestion was quenched by adding formic acid (Acros) to pH 2.0 and digests stored in freezer (≤ -50°C) until analysis.
Eluted peptides were ionized by electrospray (3.0 kV) and scanned from 110 to 2000 m/z in the Orbitrap with resolution 30000 in MS 1 at scheduled 10-min-long window. Targeted precursors were isolated (isolation window 2.0 m/z) and fragmented by collision-induced dissociation (CID, normalized collision energy NCE 35%, activation time 10 ms) in the ion trap, and detected in the orbitrap with a resolution of 7500. Raw data were imported into and analyzed with Skyline 37 .
Peak boundaries for integration were manually inspected and adjusted if necessary to include the entire peak. Where indicated, normalization was performed by dividing raw TTR peptide peak areas by either the TIC or raw peptide peak areas of the indicated normalization factor. The MS proteomics data and associated results files have been deposited to the Panorama Repository 38 and are available at https://panoramaweb.org/GenereuxLab_MistargetingAssay.url Data Dependent Acquisition: 15 μ L avidin-purification digest from HEK293T cell expressing eGFP.N2 was analyzed using the same interface and LC gradient as in PRM method. Eluted peptides were ionized by electrospray (3.0 kV) and scanned from 110 to 2000 m/z in the Orbitrap with resolution 30000 in data dependent acquisition mode. The top ten peaks from each full scan were fragmented by higher energy C-trap dissociation (HCD) using a normalized collision energy of 38%, a 100 ms activation time, and a resolution of 7500.
Dynamic exclusion parameters were 1 repeat count, 30 ms repeat duration, 500 exclusion list size, 60 s exclusion duration, and 1.50 Da exclusion width. MS 1 and MS 2 spectra were searched with MSFragger (with FragPipe 39,40 ) against a combined database of Uniprot human proteome database (downloaded with FragPipe, 2021-07-09), cyt APEX, ER HRP and chicken avidin, and reverse sequences for each entry as the decoy set, with common contaminants (e.g. keratin, porcine trypsin, etc.). Closed searches were allowed for static modification of cysteine residues (57.02146 Da, carbamidomethylation), variable modification of methionine (15.9949 Da, oxidation), N-terminal free amino group (42.0106 Da, acetylation) and tyrosine residues (361.14601 Da, +BP), half tryptic peptidolysis specificity, and mass tolerance of 20 ppm for precursor mass and 20 ppm for product ion masses. Spectral matches were assembled and filtered with a false discovery rate (FDR) of 0.01.

Statistics
For quantification of IB or PRM experiments of same types of conditions, we normalized individual densitometric signal or MS 1 TIC-normalized peak area by the sum of all conditions. For the comparison across different PRM experiments with distinct drug treatment conditions, we divided individual raw peak area or global standard-normalized peak area datum to that of (MG132 and Veh.) sample (fold change) across 10 experiments. To be conservative, these fold changes were subjected to two-tailed heteroscedastic t-test in Excel, with Bonferroni correction (6 comparisons).

RESULTS AND DISCUSSION
ER pre-QC has been described as a general protective mechanism of the ER in the presence of ER stress, and it is presumably regulated by activation of the ER unfolded protein response (UPR). 11,18,41 This model suggests that all ER stressors should induce ER pre-QC to a similar extent. We used proximity labeling combined with immunoblotting to determine whether ER stress, independently of the mechanism by which it is activated, always mistargets FLAG TTR (a known pre-QC substrate 12,25 ) into the cytosol. In addition to Tg, which induces ER stress through depletion of ER Ca 2+ , we considered the well-studied small molecule ER stressors tunicamycin (Tm), 2-deoxy-D-glucose (2-DG), Brefeldin A (BFA) and 1,4-dithiothreitol (DTT). Tm inhibits GlcNAc-1-phosphate transferase, blocking the first step of N-glycosylation. 42 Tm treatment leads to glycoprotein misfolding inside the ER and activation of ER UPR, and it is also a reported ER pre-QC inducer in HepG2 cells 12 . Different from Tm, 2-DG inhibits N-glycosylation due to its aberrant incorporation into the N-glycan, in place of mannose. 43 BFA leads to cis-Golgi cisternae collapse into the ER and a complete loss of ER-to-Golgi transport and canonical protein secretion. 44-46 DTT is a cell-penetrable reductant that triggers ER stress by preventing disulfide bond formation inside the ER.

Immunoblotting provides inadequate sensitivity for quantifying mistargeted protein
We treated HEK293T cells co-expressing cyt APEX and FLAG TTR with chemical ER stressors (Tg, Tm, BFA, DTT, or 2-DG) in the presence of MG132 and determined the relative mistargeted (cytosolic) FLAG TTR under each condition using proximity labeling and IB. Only Tg increases FLAG TTR mistargeting relative to vehicle treatment ( Figure S1, avidin-purification IB: TTR). BFA and 2-DG did not induce as much BiP expression (a UPR target) under these conditions as Tg, Tm, or DTT ( Figure S1, whole cell lysate IB: KDEL, lanes 3-5 vs 2). Hence, we performed titrations to determine optimized conditions for UPR induction ( Figure   S2). 2-DG did not effectively induce BiP upregulation at any concentration in these cells, leading us to exclude this stressor in future experiments. We also observed that DTT, 2-DG, and higher concentrations of BFA inhibited total peroxidase labeling yield. For BFA, we chose the minimum concentration that still yields maximum BiP expression. For DTT-treated cells, cyt APEX labeling can be rescued by aspirating DTT-containing media and replacement of fresh media containing 1 mM H 2 O 2 (Figures S3, whole cell lysate, ECL: biotin). We repeated the treatments with the optimized conditions for each stressor, but still found that only Tg-treated cells display increased FLAG TTR mistargeting in the cytosol ( Figure S3, avidin purification, IB: TTR). It is difficult to evaluate the extent to which conditions affect TTR mistargeting, however, because mistargeted populations are small and IB bands after proximity labeling are often faint and difficult to quantify ( Figure   S1, S3). Firm conclusions would require substantial material scale-up, and there are many chemical and genetic ER stressors worth considering, especially if pre-QC activation is dependent on how stress is induced. This limitation made us consider using a more sensitive platform for quantifying mistargeted proteins.

Development of the PRM assay
Peroxidase proximity labeling has been previously integrated with mass spectrometry, with quantification by SILAC 28,47-50 , TMT 51-53 , or label-free methods including MRM and PRM 54-59 . We decided to replace IB with PRM as our detection methods, as shown in Figure 1. In addition to FLAG TTR, we included peptides from the labeling peroxidase cyt APEX. Immunodetection of cyt APEX in avidin-enriched samples is difficult, since common epitopes including the FLAG . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10. 1101/2023 tag are disrupted by proximity labeling, presumably due to direct modification. We considered endogenously biotin-binding proteins in mitochondrial matrix 52,60-62 as potential global standards for data normalization. These include PC, PCCA, MCCC1 and ACACB. We also targeted common loading control proteins β -actin, α -tubulin and GAPDH. Stress-inducible chaperones HSPA1A (nucleocytosolic) and BiP (alias HSPA5/GRP78, primarily ER luminal) were also included.

Figure 2
Workflow of choosing peptides for PRM assay. For the generation of the initial target list, protein sequences were subjected to in silico tryptic digestion (C-term to K/R, not before P) and filtered in Skyline. For proteins that are at low abundance in lysate, namely mitochondrial biotin carboxylases, avidin-enriched samples were prepared for liquid chromatography -tandem mass spectrometry (LC-MS/MS) data dependent acquisition (DDA) mode. Raw data was searched with MSFragger 39,40 . Fully tryptic precursor ions with decent intensity, proper retention time, and no sub-stochiometric modification sites and ragged ends are kept. Filtered candidates . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10. 1101/2023 were confirmed with unscheduled PRM-MS first, retention times of those were confirmed with scheduled PRM-MS. Peak shapes and other properties of selected precursors were evaluated, before or during application. The workflow of seeking target peptides is summarized in Figure 2. In general, we required that peptides (i) be 8-25 amino acids long, (ii) do not contain labile modification sites, (iii) do not contain ragged ends of tryptic digestion, 63 and (iv) be available in NIST peptide tandem mass spectra library 64 . For cyt APEX peptides and some of chicken avidin peptides, we turned to Prosit 65 to predict their CID fragmentation patterns at NCE 35%. Uniqueness was examined by either using a background proteome in the Skyline software 37 , or by uploading the candidate precursor list into Nextprot 66 . For actin and tubulin, uniqueness was required at the class-level and not the family-level. Retention times for candidate peptides from FLAG TTR, cyt APEX, actin, tubulin, GAPDH, HSPA1A and BiP were determined by unscheduled analysis of lysate digest, followed by scheduled analysis to evaluate peak shapes. For cyt APEX, actin and tubulin peptides, precursor ion chromatograms are used for quantification because little interference is observed. Mitochondrial matrix biotin carboxylases are at low levels in HEK293T lysate and chicken avidin should not be present in HEK293T lysate.
Hence, their peptides were first evaluated from an avidin-purification sample digest in data dependent acquisition mode with the same LC gradient. Only PC peptides, as opposed to those from PCCA, MCCC1 and ACACB, were used because they are the most abundant among the mitochondrial biotin carboxylases.
Peptides with proper retention time, high intensity, good peak shape, little transition interference were confirmed with scheduled PRM. We also removed peptides that were confirmed to be deamidation-prone.
The targeted peptides in this study are summarized in Table 1, with coefficients of variance (CVs) of 8 technical replicates listed as well. CVs CV (TIC) is calculated based on peak areas normalized by MS 1 total ion current (TIC) chromatogram.
. CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint

Validation of proximity-labeling PRM Assay
We tested whether IB and PRM yield comparable results with our assay. To perform the comparison, we used drug treatment conditions that provide a large dynamic range of FLAG TTR mistargeting yields. Mycolactone A/B (ML) is an inhibitor of Sec61-mediated co-translational translocation for secreted and type-I and type-II transmembrane proteins 67-69 . It should completely arrest FLAG TTR translocation during the time-course of the experiment. When used in combination with the proteasome inhibitor MG132, ML is expected to give us the most FLAG TTR accumulation inside the cytosol. Tg is a noncompetitive SERCA inhibitor that rapidly induces severe ER stress 70 , and is a known inducer of ER pre-QC 11,12,9 . From our previous study by IB, combined Tg and MG132 treatment triggers around a 3-fold increase in FLAG TTR mistargeting compared to MG132 treatment alone, and 6-fold increase compared to the basal condition. 25 HEK293T cells co-expressing FLAG TTR and cyt APEX were treated with vehicle, ML or Tg for 16 hours in the absence or presence of 16-h MG132 treatment, before BP-labeling and quenching. Eluate samples (avidinpurifications) were split in half. One half was separated by SDS-PAGE followed by IB (Figure 3a), while the other half was prepared for bottom-up proteomics and analyzed by PRM. The amount of eluate digest we injected is equivalent to one tenth the amount we recovered from avidin beads. IB is quantified by densitometry, and PRM by raw peak areas of the TTR peptide AADDTWEGFASGK 2+ . Each sample was normalized to the total intensity across conditions for a given replicate. 71 Tg induces a 2.5-fold increase over vehicle in FLAG TTR mistargeting when co-treated with MG132 (Figure 3b,  conditions; for PRM, it is raw peak area from each condition normalized to the sum across all conditions. See Figure S6 for full blots of both replicates. c) Plot of PRM quantification against IB is displayed to show correlation between the two methods, within the same scale. Linear regression equation and Pearson's R 2 displayed as well.

ER pre-QC induction is indeed ER stressor-dependent
With this PRM-coupled mistargeting assay, we revisited the small molecule ER stressors Tg, Tm, BFA and DTT (raw peak areas in Table S2). Having observed that not all ER stressors increase mistargeting of FLAG TTR in HEK293T cells, we also included two other molecules that impact ER calcium homeostasis.
Cyclopiazonic acid (CPA) is another SERCA inhibitor, but differs from thapsigargin in that it is a competitive inhibitor, less potent, and inhibits SERCA reversibly 72 . Diltiazem (Dil.) is a calcium channel blocker that is used to maintain ER calcium level by preventing Ca 2+ leakage. Dil. does not induce ER stress, but does influence the ER protein homeostasis through elevated activity of ER calcium binding proteins 73,74 . We confirmed that 100 µM CPA induces ER stress similarly to 50 nM Tg on the basis of BiP upregulation following a 16-h treatment ( Figure   S7a, IB:KDEL, lanes 3,4 vs 1,2 and 7,8 vs 5,6).
We also took advantage of the inherent multiplexing of PRM to consider normalization. Appropriate normalization to control for loading, sample handling, and ionization efficiency is necessary for most biological mass spectrometry techniques. However, biased normalization methods can introduce artifacts into . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint interpretation of results. A straightforward normalization factor is the area under the entire MS 1 TIC chromatogram 75 . This factor should control for injection efficiency, loss of material during sample preparation, differences in the recovery of cells or protein, errors in protein quantification prior to avidin purification, and to the extent that the signal is dominated by cyt APEX-labeled proteins, the labeling activity of cyt APEX in a given experiment. Mitochondrial carboxylases such as PC, which are endogenously biotinylated, have been used for normalization in proximity labeling experiments when the peroxidases is localized elsewhere than the mitochondria 52,60-62 , controlling for the total amount protein loaded onto (strept)avidin beads. These proteins as normalization factors are valid if the assumptions of consistent expression level and consistent proximity labeling activity are maintained across conditions. Unlike normalization against TIC and PC, normalization against abundant proteins that share a compartment with the peroxidase can control against changes in BP-labeling efficiency. 76 We considered β -actin, α -tubulin and GAPDH. We found that α -Tubulin 1B levels are affected by cellular stress (e.g. Figure S6a,b, Lysate IB: α -tubulin) and poor chromatographic performance in our gradient for GAPDH peptides, and hence focused on β /γ-actin as a proxy for protein load and BP-labeling yield by cyt APEX. We also considered normalization against the heme peroxidase cyt APEX itself, under the expectation that cyt APEX auto-labeling is a proxy for total biotinylation yield.
We compared relative FLAG TTR mistargeting across drug treatments and different normalization schemes (Figure 4). Similar results are seen for most treatments. The SERCA inhibitors Tg and CPA induce pre-QC to similar extents.
Tm and Dil do not induce pre-QC. The observed relative mistargeting of FLAG TTR following ML treatment varies between normalization methods. While DTT lowers the apparent mistargeted TTR load with each normalization, the extent of this . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint decrease varies from 86% with PC normalization (Figure 4b) to 40% with β -actin or cyt APEX normalization (Figure 4c,d). BFA shows the largest disparity, doubling FLAG TTR mistargeting with β -actin or cyt APEX normalization, moderately increasing (14% increase) with TIC normalization, or having no effect with PC normalization. was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023.

Determination of Appropriate Normalization
To explain the disagreement among normalization methods for ML-, DTTand BFA-treated cells, we considered how each of these will be affected by changes to labeling efficiency. PC recovery from avidin purification will solely reflect the total amount of lysate added to the beads. It will be insensitive to changes in peroxidase labeling. Recovery of biotinylated cyt APEX or β -actin, by contrast, will reflect both the amount of cells harvested as well as the peroxidase labeling efficiency. Normalization against MS 1 TIC will also partially account for differences in labeling efficiency, however several other factors will affect the TIC signal (Figure 5a). These include carboxylases such as PC, which reflect total protein inputs, but also common contaminants (keratin, trypsin, etc.) and avidin that can be leached from the beads in a strongly condition-and lot-dependent manner 77 .
Prior to the affinity purification step, we bring clarified lysates to the same protein concentration, based on Bradford assay, and load the same mass of protein to avidin-agarose beads. If the drug treatments, as compared to vehicle, do not change the profile of labeling-independent biotin carboxylases, background . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; cytoplasmic proteome and the labeling activity of cyt APEX, normalization to different global standards should yield similar results (Figure 5b left, "comparable labeling activity"). If cyt APEX labeling activity somehow decreases under some drug treatment conditions, but the total amount of biotin carboxylases and cytoplasmic background remain consistent, the portion of enriched cytoplasmic protein (e.g. cyt APEX and β -actin) is expected to decrease accordingly with the labeling activity decrease. This may result in a relatively higher proportion of PC and other labeling-independent components. Eventually, if data are divided by the level of cyt APEX or actin, the normalized value will be higher than that normalized by PC. On the contrary, data divided by PC level will be smaller than that by cyt APEX or actin. MS 1 TIC can represent labeling activity to some extent, but it is also convoluted by labeling-independent components, thus mistargeting normalized by TIC is expected to be in the middle of the two extremes (Figure 5b center, "decreased labeling activity"). And vice versa, if cyt APEX labeling activity increases upon some treatments compared to the control condition, an opposite trend will be expected (Figure 5b right, "increased labeling activity"): the proportion of avidin-purified cytoplasmic proteins (e.g. cyt APEX or actin) will be relatively higher, resulting in a lower apparent mistargeting after normalization, while the proportion of labeling-independent PC will be lower, leading to a higher apparent mistargeting after normalization. Again, TIC normalization should yield an intermediate result. We find that our data nicely matches this model ( Figure   5C), indicating that the divergence between TIC-and PC-normalization and cyt APEX-normalization can be entirely ascribed to drug treatment dependent variation in peroxidase labeling efficiency. We saw in our immunoblotting experiments ( Figure S1) that BFA and DTT decrease labeling efficiency, and altered our protocol to mitigate this interference. Nevertheless, it is clear that these treatments even under optimized conditions affect peroxidase labeling enough to . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10. 1101/2023 influence the quantitative accuracy of the data. Given that DTT is a potent reductant, it is not surprising that it inhibits oxidative labeling. The cause of inhibition during BFA treatment is unclear. It could be that changes in glutathione redox state also mediate the increased labeling following ML treatment, as a recent study shows that ML depletes cellular glutathione in myeloid leukemia cells  Whatever the basis of the change in labeling efficiency, by using PRM, we can normalize against cyt APEX auto-labeling and remove this confounding factor to find the most accurate quantification of mistargeting ( Figure 4C) . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint Prediction of how relative amount of BP-labeling yield and PC level impacts normalized FLAG TTR mistargeting. For treatments that neither alter cytoplasmic proteome too much, nor change the relative levels of cyt APEX, actin and PC, normalization to cyt APEX, actin, TIC and PC gives similar results. For conditions that decrease the labeling activity, fold change normalized to cyt APEX or actin will be higher than that by PC. For conditions that increase the relative level of labeling, fold change normalized to cyt APEX or actin will be lower than that by PC. c) Head-tohead comparison between apparent FLAG TTR mistargeting normalized by different factors, per drug treatment. These drug treatment conditions are categorized based on the pattern in panel b. . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; More broadly, we have demonstrated that the multiplexing capacity of PRM can be leveraged to ensure appropriate normalization when using in situ peroxidase labeling ASSOCIATED CONTENT

Supporting Information
Supporting information (pdf) includes Table S1, description of drug treatment conditions, Figures including whole blots, extracted ion chromatograms, and mass spectra.

Author Contributions
The manuscript was written through contributions of all authors. All authors have given approval to the final version of the manuscript.
. CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ;

Notes
The authors declare no competing financial interest.
. CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made

REFERENCE
The copyright holder for this preprint (which this version posted July 20, 2023.  . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. (  3  9  )  K  o  n  g  ,  A  .  T  .  ;  L  e  p  r  e  v  o  s  t  ,  F  .  V  .  ;  A  v  t  o  n  o  m  o  v  ,  D  .  M  .  ;  M  e  l  l  a  c  h  e  r  u  v  u  ,  D  .  ;  N  e  s  v  i  z  h  s  k  i  i  ,  A  .  I  .   M  S  F  r  a  g  g  e  r  :  U  l  t  r  a  f  a  s  t  a  n  d  C  o  m  p  r  e  h  e  n  s  i  v  e  P  e  p  t  i  d  e  I  d  e  n  t  i  f  i  c  a  t  i  o  n  i  n  M  a  s  s  S  p  e  c  t  r  o  m  e  t  r  y  -B  a  s  e  . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023.  I  S  T  L  i  b  r  a  r  i  e  s  o  f  P  e  p  t  i  d  e  F  r  a  g  m  e  n  t  a  t  i  o  n  M  a  s  s  S  p  e  c  t  r  a  ,  N  I  S  T  S  t  a  n  d  a  r  d  R  e  f  e  r  e  n  c  e   D  a  t  a  b  a  s  e  1  C  ,  2  0  . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10.1101/2023.07.19.549095 doi: bioRxiv preprint TOC graph . CC-BY 4.0 International license available under a was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprint (which this version posted July 20, 2023. ; https://doi.org/10. 1101/2023