Polyubiquitin-sensor proteins reveal localization and linkage-type dependence of cellular ubiquitin signaling

Avidity-based design for linkage-specific polyUb binding

We previously described how the short linkers between domains in naturally-occurring tUIM proteins can determine polyUb linkage specificity by promoting avid interactions with multiple Ub units in a particular configuration¹⁴. Helical 7-amino acid linkers position 2-UIM peptides for optimum avid binding across a single Lys63-linkage, but not to other linkage-types or monoUb. Whereas 8-amino acid linkers are near-optimum for Lys63 binding, 6-amino acid helical linkers confer low affinities for Lys63-polyUb because they position the Ub-binding sites out of helical phase with each other (Fig. 1a,b). We extended this natural strategy to create “Vx3”, a 3-UIM peptide with the relatively tight-binding Vps27 UIM domains (K_d^monoUb = 100–200 μM¹⁶) and weakly helical 7-amino acid linkers. Remarkably, Vx3 bound Lys63-polyUb with K_d = 4.4 nM, a 70-fold preference over Lys48-polyUb, and a 2,700-fold preference over linear (α-amine-linked) polyUb.

Consistent with the principle that more-structured linkers provide higher affinity and specificity, strongly α-helical hepta-alanine linkers (Supplementary Fig. 1) improved the affinity and specificity of Vx3 (Table 1, Vx3(A7)). This peptide bound to Lys63-Ub₄ more tightly than we could accurately measure, with K_d ≤ 200 pM (Supplementary Fig. 1). In spite of the improved Lys63-linkage specificity, we reasoned that the range of concentrations over which Vx3(A7) was highly linkage-specific (0.01 – 1 nM) was too low to be useful in cells because physiological concentrations of non- Lys63-linked species probably exceed 1 nM in some cases; total cellular Ub may range several-fold to 85 μM, with ~10% distributed among mostly Lys48-, Lys63-, and Lys11-linked polyUb conjugates^2,17.

To shift the high-specificity regime to high-nanomolar concentrations, we paired the weaker UIMs from human Rap80 (K_d^monoUb = 500–1,000 μM¹⁴) with all-alanine linkers to create Rx3(A7) (Fig. 1c). Indeed, Rx3(A7) bound Lys63-polyUb as tightly as Vx3 (K_d^Ub3 = 5 nM) but with additional specificity conferred by the highly-structured linkers. We measured a 1,100-fold preference for Lys63- over Lys48-polyUb, and a 23,000-fold preference over linear-polyUb. Binding to Lys11-polyUb was too weak for us to saturate, but the preference for Lys63- over Lys11-polyUb probably exceeds 3,300-fold (Table 1; Fig. 1d). Rx3(A7) therefore displays high linkage-specificity for Lys63-polyUb over a broad concentration range that likely encompasses typical cellular polyUb levels (shaded region, Fig. 1d). Although this range is narrower for Vx3 (see Table 1), as described below, in many applications Rx3 and Vx3 can be used with qualitatively similar effects. As expected from structural modeling (Fig. 1b), an 8-alanine linker conferred weak Lys63-polyUb binding in the 3-UIM context (Rx3(A8)), and Rx3(A6) had even lower affinity (K_d = 160 nM and 2,900 nM, respectively; Supplementary Fig. 1).

These results indicated that linkage-specific avidity may be applied to produce polyUb-binding reagents of arbitrarily high affinity and with linkage specificities that rival or exceed linkage-specific polyUb antibodies^18,19. Indeed, biotin-labeled Vx3 bound specifically to Lys63-polyUb chains immobilized on a membrane, and thus may be used to identify polyUb types much like an antibody in immunoblot and other biochemical applications (Fig. 1e). Unlike antibodies, however, these reagents may be easily expressed inside living cells; here we focus on these applications.

Intracellular Lys63-linked polyUb structures and dynamics

We transiently expressed tUIM fusions to mEGFP (monomeric Enhanced Green Fluorescent Protein) in cultured human cells. Both Vx3-mEGFP and Rx3(A7)-mEGFP were somewhat excluded from the nucleus and localized to large perinuclear foci that vary widely in appearance from cell to cell (Fig. 2a; Supplementary Fig. 2). Similar structures have been observed in cells examined by immunofluorescence with anti-Lys63-polyUb antibodies¹⁹. A subset of these foci co-localized with markers of autophagic vesicles (Supplementary Fig. 2), consistent with recent reports of a role for Lys63-polyUb in cargo selection for some autophagy processes¹¹. Consistent with Lys63-polyUb binding, the intermediate Lys63-binder Rx3(A8)-mEGFP was mostly diffuse, with cytosolic punctae appearing only in cells that expressed the very highest levels of the fusion protein. The weaker-binding Rx3(A6)-mEGFP was completely diffuse in all cells, identical to the non-Ub-binding control protein Vx3NB-mEGFP (Figs. 1c and 2b,d–f). Nuclear-localized NLS-Vx3-EGFP appeared in smaller and more uniformly-sized foci (Fig. 2c), a subset of which co-localized with the DNA double strand break (DSB) markerγH2AX (Fig. 4d, and discussed below). We found that some Vx3-EGFP was apparently ubiquitinated in human cells (≤10%, data not shown), an unwanted modification that could be eliminated by lysine-to-arginine substitutions within the tUIM sequence. The thus altered peptide, Vx3K0, was virtually indistinguishable from Vx3 with respect to in vitro polyUb affinity (Table 1), subcellular localization, and inhibitory activity (discussed below). Taken together, these results indicate that the cellular structures revealed by the Vx3 and Rx3(A7) proteins were ubiquitin-specific, but were not the result of a generic Ub-binding capacity. Instead, these peptides appeared to localize to cellular structures through interactions with specific polyUb topologies.

To determine the linkage-type dependence of the cytosolic aggregates that Vx3K0-and Rx3(A7)-mEGFP bind in vivo, we used yeast strains previously engineered to express Ub from a single transgene⁵. We compared Vx3K0-EGFP localization in single-Ub (SU) yeast expressing wild-type Ub (SU-wt) or a Lys63R-mutant Ub (SU-Lys63R) that can produce all polyUb topologies except Lys63-linked⁶. Similar to the pattern in human cells, Vx3K0-EGFP localized to large, mostly perivacuolar structures in SU-wt but not SU-Lys63R yeast (Fig. 2g–i). The binding-deficient Vx3NB-EGFP was diffuse in both strains, as expected (Supplementary Fig. 2). Vx3K0-EGFP expressed in a wild-type yeast (not SU) was localized as in the SU-wt strain, indicating that this pattern was not an artifact of the strain manipulation used to replace Ub.

Next we used Vx3-EGFP to track Lys63-polyUb signals in live cells. The E3 Ub-ligase Parkin translocates to damaged mitochondria where it promotes Lys63-polyubiquitination of multiple substrates^20,21. The damaged organelles are ultimately condensed into perinuclear aggregates for processing by autophagic degradation²². We transfected YFP-Parkin alone or Vx3K0-EGFP with unlabeled Parkin into HeLa cells stably expressing the mitochondrial marker mito-dsRed, and treated the cells to depolarize the mitochondria. Consistent with previous reports, YFP-Parkin rapidly relocalized to damaged mitochondria (Fig. 3a); this was followed by recruitment of Vx3K0-EGFP, presumably by Lys63-polyUb-conjugated mitochondrial proteins (Fig. 3b,c). This activity required the Ub-binding residues of Vx3 and, as HeLa cells do not express the ligase, an active, exogenous Parkin (Supplementary Fig. 3). These results indicate that our designed proteins can localize Lys63-polyUb-modified conjugates dynamically inside cells. Note that clustering of damaged mitochondria was somewhat reduced in cells also expressing Vx3K0-EGFP (Fig. 3a,b). As clustering is thought to be promoted by binding of p62 to Lys63-polyUb at the mitochondrion²¹, this suggests that the Lys63-polyUb was masked by bound Vx3. However, because the extent of clustering can vary with the expression levels of Vx3K0-EGFP or Parkin, this mechanism, though likely, is speculative.

PolyUb-sensor proteins as inhibitors of cellular signaling

Ub recognition by natural receptors usually is limited to a small set of residues³ and thus typically excludes simultaneous binding by other receptors or deubiquitinating enzymes. For this reason, we expected that expression of Lys63-specific polyUb sensors may inhibit Lys63-dependent signaling and specifically prevent deubiquitination of these conjugates (Supplementary Fig. 4). Indeed, expression of Rx3(A7)-mEGFP in HeLa cells for 20 h doubled Lys63-conjugates as measured by quantitative immunoblot (Fig. 4a,b). Total polyUb, Lys48-polyUb, and unconjugated Ub were largely unchanged compared to control-transfected cells, though the polyUb species appeared partially shifted to higher molecular weight forms. These results are consistent with the in vitro linkage specificities of the polyUb sensor proteins, and indicate that these constructs may specifically inhibit the action of endogenous Lys63-polyUb effectors in cells.

To test this, we expressed Vx3 fused to maltose binding protein (MBP-Vx3) in wild-type yeast on high- or low-copy plasmids (Fig. 4c). We compared these strains for their resistance to the arginine analogue canavanine or the genotoxic alkylating agent methyl methanesulfonate (MMS), previously-described phenotypes attributed to the capacity to make Lys63-polyUb^{6, 23}. Indeed, yeast expressing MBP-Vx3 variants from high-copy plasmids were nearly as sensitive to canavanine as the SU-Lys63R strain (Fig. 4c). We did not identify the impaired function that results in canavanine sensitivity in Vx3-expressing cells, although prominent possibilities include defective arginine permease downregulation and impaired turnover of misfolded canavanine-containing proteins. Next we examined yeast growth in the presence of MMS, where Lys63-polyUb-modified proliferating cell nuclear antigen (PCNA) at sites of MMS-caused DNA damage is required to recruit specialized damage-tolerance machinery⁹. Remarkably, MBP-Vx3 expression restricted growth on MMS-containing plates in a dose- and compartment-specific manner. Yeast expressing NLS-MBP-Vx3 from a low-copy CEN plasmid showed inhibited growth in the presence of MMS, and yeast expressing NLS-MBP-Vx3 from the high-copy 2μ plasmid were as sensitive to MMS as the SU-Lys63R yeast (Fig. 4c). Constructs lacking the NLS did not accumulate in the nucleus (as assessed using an EGFP fusion) and had no effect.

We also examined the effect of NLS-Vx3-EGFP expression on DNA double-strand break (DSB) repair in mammalian cells, where Lys63-polyUb similarly recruits repair factors to damage sites. Expression of NLS-Vx3-EGFP in ionizing radiation (IR)-treated cells, while not affecting formation of γH2AX-containing foci, severely delayed foci disappearance (Fig. 4d,e). This is consistent with the proposed role for Lys63-polyUb downstream of the γH2AX modification at DSBs¹⁰, but upstream of repair factors important for clearance like Rap80 and BRCA1¹⁰, and suggests that NLS-Vx3-EGFP inhibits DSB processing at these sites by blocking access of Rap80 and other repair factors to Lys63-modified substrates. Indeed, in response to IR, Rap80-positive foci were reduced in cells expressing NLS-Vx3-EGFP as compared to cells expressing the Vx3NB-EGFP negative control (median mCherry-Rap80-foci per cell were 5 and 19, respectively; p-value < 0.0001, n = 60) (Fig. 4f). These results establish that our designed proteins can inhibit Lys63-polyUb signaling in cellulo, and suggest that they act by displacing endogenous receptors of polyUb signals.

Resolving roles for linkage-specific polyUb in NF-κB activation

To address linkage specificity and inhibitory activity in cells, we sought a functional assay in which inhibitor expression levels could be linked to related outcomes with known Lys63-and non-Lys63-polyUb dependences. Ligand-mediated NF-κB activation is an ideal test case, since recent reports have illuminated the divergent polyUb landscape downstream of two important cell-surface receptors, interleukin-1 receptor (IL-1R) and tumor necrosis factor receptor (TNFR).

Both IL-1R and TNFR signaling follow the canonical NF-κB activation pathway (namely, ligand binding to a receptor, inhibitor of kappa B kinase (IKK) activation, and nuclear translocation of the NF-κB transcription factor)^13,24. Non-degradative polyUb chains formed on one or more substrates (for example, kinase receptor interacting protein-1, RIP-1) near the receptor serve as a signaling platform to recruit a polyUb-binding subunit of IKK for activation. Originally, these chains were thought to be Lys63-linked for both TNFR and IL-1R signaling. However, a Ub-replacement strategy⁷ in the human U2OS osteosarcoma cell line showed that, when the sole source of Ub was Ub(Lys63Arg), IL-1β-mediated IKK activation was completely abrogated, whereas activation in the TNF-α-stimulated pathway was unaltered. More recently, mass spectrometry and linkage-specific immunoblot analyses indicated that Lys63- as well as Lys11-linked chains are present at the TNFR in roughly equal proportions after treatment with TNF-α²⁵. The apparent resolution of these two studies is that IL-1β signaling is strictly dependent on Lys63-polyUb whereas signaling through the TNFR can function with Lys11-polyUb alone, although the typical polyUb complement is a poorly-defined mix of Lys11 and Lys63 linkages. Thus, a stringent test of our inhibitor proteins is the ability to discriminate these distinct polyUb dependencies in otherwise highly similar pathways.

We examined IL-1β- and TNF-α-stimulated NF-κB activation in U2OS cells in the presence of retrovirally-transduced Rx3(A7)-mEGFP or, as a negative-control, Vx3NB-mEGFP. We tracked activation status at the single-cell level by immunofluorescence detection of the endogenous NF-κB subunit p65, which translocates from the cytosol to the nucleus upon activation by either ligand (see Online Methods). As expected, p65 was cytosolic (translocation score 0) in unstimulated cells, but nuclear (translocation score 1) in IL-1β– or TNF-αtreated cells after 30 min (Fig. 5b). Remarkably, Rx3(A7)-mEGFP inhibited NF-κB activation in IL-1β-stimulated cells at all expression levels (Fig. 5a,b). At the lowest measurable mEGFP concentrations, inhibition was incomplete, but rapidly reached saturation upon higher Rx3(A7)-mEGFP expression, consistent with high affinity of the Rx3(A7)· Lys63-polyUb interaction (Table 1). In contrast, activation in TNF-α-treated cells was unaffected by low doses of Rx3(A7)-mEGFP (Fig. 5a,b) and, compared with stimulation by IL-1β, required ~10-fold more inhibitor (as measured by mEGFP intensity) to prevent p65 translocation. At low expression levels, where we expect the inhibitor proteins to be most specific, Rx3(A7)-mEGFP clearly distinguishes the linkage-type dependences of these two ligand-mediated NF-κB pathways and thus functions as a linkage-specific inhibitor of cellular polyUb signaling. We also found that inhibition by Vx3K0-mEGFP could distinguish IL-1β and TNF-α signaling in U20S cells, though with a somewhat smaller window of inhibitor concentrations over which linkage-specificity is observed (Supplementary Fig. 5).

To explore the relationship between in vitro linkage specificities and inhibitor activity in cells, we evaluated the effects of each of our tUIM-mEGFP fusions on TNF-α signaling in HeLa cells. Consistent with its higher in vitro Lys63-linkage specificity, higher cellular concentrations of Rx3(A7) were required to inhibit activation as compared to Vx3K0-mEGFP (Fig. 5c; note also the similarity to the Rx3(A7) profile in 5b); we suspect that, consistent with its lower linkage specificity in vitro, Vx3K0 is a slightly better binder for the non-Lys63-polyUb species involved in TNFR signaling. Rx3(A8)-mEGFP showed greatly reduced inhibitory activity, and Rx3(A6)-mEGFP was virtually indistinguishable from the Vx3NB non-binding control (Fig. 5c). These data taken together further suggest that the inhibition observed is due to binding to specific polyUb topologies.

These inhibitor profiles highlight the requirement to account for polyUb sensor expression levels in order to distinguish Lys63-dependent from non-Lys63-dependent signaling; with very high Rx3(A7)-mEGFP expression, both IL-1β and TNF-α signaling were completely inhibited, presumably because of weak “off-target” interactions with non-Lys63-polyUb chains such as homogeneous Lys11-polyUb, or perhaps a mixed Lys63/Lys11-polyUb species. To establish an upper limit of cellular inhibitor expression levels for linkage-specific action, we determined the IC50 of TNF-α inhibition by Rx3(A7)-mEGFP in HeLa cells by calibrating image data to quantitative immunoblots. With TNF-α signaling inhibition as a marker for non-Lys63 binding, we determined that Rx3(A7)-mEGFP above 16 μM no longer showed strong Lys63-linkage preference (Supplementary Fig. 6). This concentration agrees closely with our in vitro polyUb binding constants for non-Lys63-linked chains (Table 1), particularly when considering that effective intracellular inhibitor concentrations must be slightly lower than this because some Rx3(A7)-mEGFP typically is in perinuclear aggregates. By our estimate, the level required for linkage-specific inhibition is well below expression levels achievable from typical transient transfection of cultured cells (Fig. 5d); therefore, care must taken when applying these reagents to either account for cellular inhibitor levels explicitly at the single-cell level or, by adjusting average inhibitor expression levels (for example, through optimized transfection or a tunable induction strategy), to be within the linkage-specific concentration regime of the inhibitor.

Next, we examined signaling through the TNF-related weak inducer of apoptosis (TWEAK) receptor FN14. TWEAK ligand is a physiological activator of NF-κB that acts through signaling mechanisms that are largely unknown with respect to polyUb²⁶. Interestingly, the ubiquitination enzymes UbcH5 and c-IAP1 are important downstream components of both TWEAK- and TNF-α-mediated NF-κB activation^{25, 27}. To determine whether these pathways share a common polyUb-linkage dependence, we retrovirally-transduced inhibitor and control constructs into mouse embryo fibroblasts (MEFs), a cell-type that gives robust and reproducible activation of NF-κB by TWEAK. TWEAK-receptor signaling was inhibited at very low Vx3K0-mEGFP concentrations, identical to the inhibition of the strictly Lys63-dependent IL-1β signaling in both MEF and U2OS cells (Fig. 5e; compare to the IL-1β inhibition profile in 5b). We infer from this assay that TWEAK-mediated NF-κB activation shows the same strict requirement for Lys63-polyUb as the IL-1β pathway. These results show that our linkage-specific proteins may be applied across cell types to diagnose polyUb linkage-type dependences of cellular processes.

Plasmids and proteins

We created designed tUIM genes by overlapping primer-extension PCR with 20–40 base oligonucleotides. Cloning and sequence details are summarized in Supplementary Tables 1 and 2. mEGFP used in fusion proteins was described previously³¹. For binding assays, we bacterially expressed His₆-tagged tUIMs and purified them using Ni-NTA agarose (Qiagen) according to the manufacturer’s directions. We labeled recombinant tUIMs on a single cysteine residue with fluorescein maleimide or biotin maleimide (Pierce), and further purified them by gel filtration or anion-exchange chromatography as needed. We produced polyUb chains of defined length and linkage as previously described^{32, 33}.

Polyubiquitin binding assays

We performed direct fluorescence anisotropy binding titrations as described¹⁴, with fluorescently-labeled peptide present at 1, 4, 10, or 100 nM in the cuvette, depending on the anticipated affinity of the peptide being tested. We made binding measurements at 25 °C in fluorescence buffer (25 mM Na phosphate pH 7.4,150 mM NaCl, 5 mM β-mercaptoethanol, 1 mM EDTA, and 0.05% Brij35). To reduce the amount of polyUb needed to determine K_d values for weakly-binding proteins, we performed some titrations in a competition assay format (Supplementary Fig. 1). Typically, we combined fluorescein-labeled binding peptide (e.g., 4 nM fluorescein-Vx3) with polyUb (e.g., 50 nM Lys63-Ub₃) to give ~80% saturation; we determined fluorescence anisotropy values with various amounts of unlabelled competitor protein and fit K_d values with a model of competitive binding to a single site; we derrived baseline anisotropy values from samples with no competitor protein, and the plateau value from fluorescent binding protein without polyUb ligand. Where relatively weak binding was measured, high protein concentrations (> 10 μM) caused anisotropy values to increase slightly due to changes in the solution viscosity. In those cases, we determined corrections for the viscosity effects by titrations of fluorescent binding peptide with competitor protein in the absence of polyUb. For the far-western dot blot, we spotted the indicated amounts of purified ubiquitin species in binding buffer plus 1 mg/ml BSA directly onto dry PVDF membrane and allowed them to dry completely. We blocked the methanol-wetted membranes in 3% BSA, probed the membranes with 30 nM biotinylated His₆-Vx3 peptide, washed in TST, probed with streptavidin-HRP (Pierce), and washed again with TST; detection was by chemiluminesence. Antibodies with specificity to total polyUb were from Millipore (clone FK2) and antibodies to unanchored Ub were from Epitomics (Ub C-terminus antibody; clone RPS27A).

Yeast and mammalian cell culture

We cloned MBP-Vx3 and NLS-MBP-Vx3 into pRS415-ADH (low-copy CEN plasmid) or pRS425-ADH (high-copy 2μ plasmid), each containing the constitutive alcohol dehydrogenase (ADH) promoter. We transformed these into wild-type yeast strain BY4741 (MATa leu2Δ0 his3Δ1 ura3Δ0 met15Δ0) and transformants were selected on SC-Leu medium. In the original (SUB280-derived) single-Ub yeast strains, LEU2 was used to replace UBI4. To recover the leu2 marker in these strains, we replaced via homologous recombination the LEU2 gene at the UBI4 locus with a cassette that conferred hygromycin resistance. We selected hygromycin-resistant leucine auxotrophs, and confirmed the proper integration by PCR. For yeast growth assays, we grew two independent transformants from each transformation to mid-log phase in SC-Leu medium and spotted 0.25 OD₆₀₀ equivalents (~3×10⁶ cells) in serial 10-fold dilutions on the following plates: SC-Leu (control), SC-Leu containing 0.025% methyl methanesulfonate (MMS), or SC-Leu-Arg containing 0.0001% canavanine.

We maintained HeLa and MEF cell lines in DMEM, and U2OS cells were grown in McCoy’s medium, both media supplemented with 10% fetal bovine serum, 100 I.U. penicillin, and 100 μg/ml streptomycin. We introduced mCherry-Rap80 into Rap80^−/− MEFs (a kind gift from J. Chen, University of Texas M.D. Anderson Cancer Center, Houston, TX) by retroviral transduction, followed by IL-2Rα-mediated bead-sorting to select a population of cells that stably express the construct. We introduced polyUb sensor constructs by lipid transfection using Lipofectamine 2000 (Invitrogen) or Fugene6 (Roche), or by retroviral transduction using the RetroX packaging cell line (Clontech). Confluent HeLa cells in 6-well dishes transfected with 1 μg of inhibitor or control plasmids were used for ubiquitin immunoblot analysis. We lysed cells on ice for 30 min in PBS with 1% Triton X-100, 20 mM NEM, 10 mM EDTA, and mini-Complete protease inhibitor cocktail (Roche). Antibodies in Fig. 4a were used according to the manufacturer’s instructions, and immunoblots were quantified using IR800 secondary antibodies (LI-COR).

Microscopy

For imaging Vx3-EGFP in the single-Ub yeast strains, we cloned Vx3K0-EGFP and NLS-Vx3K0-EGFP into pRS425-ADH, transformed them into yeast strains BY4741 or SUB280⁶ (expressing wild-type or Lys63Arg Ub), and selected transformants on SC-Leu. We fixed mid-log phase cells for 1 h with 3.7% formaldehyde, washed with PBS, and then examined using a 100x objective. For HeLa and MEF imaging, we fixed cells 18 h after transfection with 2% paraformaldehyde for 10 min, mounted them using ProLong Gold antifade medium (Invitrogen) with DAPI, and imaged the cells with a 60x objective and deconvolution. We transiently transfected Rap80^−/− MEF cells expressing mCherry-Rap80 with NLS-Vx3K0-EGFP or Vx3NB-EGFP by using Fugene6 reagent. After 24 h, we exposed cells to 3 Gy of γ-irradiation and allowed them to recover for 30 min at 37 °C before fixation with 2.5% paraformaldheyde. We mounted cells on slides using ProLong Gold medium and aqcuired fluorescence images with an Olympus IX81 spinning-disk confocal microscope (CSU22 head) using a 60x/NA 1.42 oil objective. For the γH2AX foci counts, we irradiated HeLa cells (1 Gy) 24 h after transient transfection. After the indicated recovery times, we fixed cells in 2.5% paraformaldehyde for 10 min, permeabilized them in 0.1% Triton X-100, stained them with anti-phospho-histone H2A.X (Ser139) (clone JBW301, Upstate) for 2 h at 1:100 dilution, incubated them with Alexa Fluor 594 anti-mouse IgG (1:300 dilution), counterstained nuclei with DAPI, and mounted the coverslips on slides using ProLong Gold. Statistical analyses of foci counts employed the non-parametric Wilcoxon rank-sum test. For mitophagy experiments, we imaged live cells in CO₂-independent media (Invitrogen) using an UltraView LCI confocal microscope (PerkinElmer) at 37 °C with a 63x/NA1.4 Apochrome objective. We used Volocity software (PerkinElmer) for analysis. We calculated the fluorescence intensity of the YFP or GFP-tagged proteins in the area of the mitochondria, normalized this to the total cell YFP or GFP, and set the start value to one. For fixed-cell mitophagy controls (Supplementary Fig. 3), we plated HeLa cells onto chambered coverglasses where they were cultured overnight. We then transiently transfected the cells with Vx3K0-EGFP or Vx3NB-EGFP plasmids. The next day, we treated cells with 10 μM CCCP for 3 h, then fixed and stained them for the mitochondrial marker Tom20 (anti-Tom20 polyclonal antibody; Santa Cruz, sc-11415). We collected fixed-cell images on a Zeiss LSM 510 confocal microscope using a 63x/NA 1.40 Apochrome objective. We adapted the staining protocol for LC3 (Supplementary Fig. 2) from a method kindly provided by J. Gutierrez (Millennium – Takeda Pharmaceuticals). 24 h after transfection (Lipofectamine 2000) we fixed cells with 2.5% paraformaldehyde for 10 min, permeabilized them with 100 μg/ml digitonin for 15 min, and blocked for 1 h with 0.5% blocking reagent (Roche #11096176001) in 100 mM Tris pH 7.5, 150 mM NaCl. Subsequently, we incubated the cells with anti-LC3 antibody (clone 4E12, MBL International; 1:300 dilution) overnight at 4 °C, stained the sample with Alexa Fluor 594 anti-mouse IgG, and mounted the coverslips on slides using ProLong Gold medium. We obtained fluorescence images with the Olympus IX81 spinning-disk confocal microscope using a 100x/NA1.4 oil objective.

NF-κB activation experiments

We measured NF-κB activation at the single-cell level by fluorescence microscopy³⁴. We plated U2OS (5,000 cells), HeLa (9,000 cells), or MEF (5,000 cells) cells in clear-bottom 96-well imaging plates. After overnight growth, we transfected or retrovirally transduced the cells as described above; we changed the medium 18–24 h after transfection or transduction and 1 h prior to the ligand stimulation. We added TNF-α (PeproTech) or IL-1β (Pierce) to the growth medium at 10 ng/ml, and TWEAK (PeproTech) 100 ng/ml. All doses were the minimum that gave reproducible p65 translocation after 30 min. We chose the 30 min time point to avoid the potential complications of apoptosis and autocrine signaling from NF-κB transcriptional targets, both of which have been shown to occur at longer times^35,36. After stimulation, we fixed cells as above, permeabilized them in PBS with 0.2% Triton X-100 for 10 min, blocked with LI-COR Blocking Buffer, incubated with anti-p65 monoclonal antibody (F-6, Santa Cruz) diluted 1:300 in blocking buffer, washed, and stained the plates with Alexa Fluor 647-conjugated anti-mouse IgG (Invitrogen) diluted 1:2000 in blocking buffer. We stained DNA and total protein with a combination of Hoechst 33342 (Invitrogen) and Whole Cell Stain Blue (Pierce). Images were collected at 10x magnification by automated microscopy (ArrayWorx, API).

We used the custom image analysis software Imagerail³⁴ to define cell and nuclear boundaries from empirical threshold values determined on the blue channel. After segmentation, we gated out untransfected cells from the analysis based on the EGFP intensity. We removed the small number of deformed, dead, or incorrectly segmented cells from the analysis by gating on features of cell size, nuclear size, and the ratio of nuclear-to-cytoplasmic intensities on the blue and green channels. Finally, we measured p65 translocation as the ratio of nuclear-to-cytoplasmic staining for each cell on the Cy5 (i.e., Alexa Fluor 647) channel. Measuring the disappearance of IκBα at earlier time points (15 and 20 min), while less robust, gave very similar results with respect to TWEAK and IL-1β inhibition (data not shown). We assigned a p65 translocation score of 0 to the average nuclear/cytoplasmic p65 ratio for unstimulated control cells (Vx3NB-mEGFP), and 1 for the control cells treated with ligand. Untransfected and control-transfected cells gave very similar responses in terms of p65 translocation. We then scaled the nuclear/cytoplasmic ratio for cells transfected with inhibitor proteins between 0 and 1 based on the control-cell analysis.

Control experiments established the equivalence of Vx3NB-EGFP and EGFP alone as negative controls and of Vx3-EGFP and Vx3K0-EGFP as inhibitors (Supplementary Fig. 5). At most transfection levels, the negative control Vx3NB-mEGFP had little effect on the NF-κB-activation status of unstimulated cells; however, we note that some constitutive NF-κB activation was apparent at very high transfection levels of all plasmids. This constitutive activation seemed worse for transient transfection than for retroviral transduction, and worse in U20S cells than in HeLa and MEF cells, but the effect was somewhat variable from experiment to experiment depending on transfection conditions, the precise time between transfection and fixation, and the inhibitor expression level.