The carboxyl-terminal sequence of PUMA binds to both anti-apoptotic proteins and membranes
Anti-apoptotic proteins such as BCL-XL promote cell survival by sequestering pro-apoptotic BCL-2 family members, an activity that frequently contributes to tumorigenesis. Thus, the development of small-molecule inhibitors for anti-apoptotic proteins, termed BH3-mimetics, is revolutionizing how we treat cancer. BH3 mimetics kill cells by displacing sequestered pro-apoptotic proteins to initiate tumor-cell death. Recent evidence has demonstrated that in live cells the BH3-only proteins PUMA and BIM resist displacement by BH3-mimetics, while others like tBID do not. Analysis of the molecular mechanism by which PUMA resists BH3-mimetic mediated displacement from full-length anti-apoptotic proteins (BCL-XL, BCL-2, BCL-W, and MCL-1) reveals that both the BH3-motif and a novel binding site within the carboxyl-terminal sequence (CTS) of PUMA contribute to binding. Together these sequences bind to anti-apoptotic proteins, which effectively ‘double-bolt locks’ the proteins to resist BH3-mimetic displacement. The pro-apoptotic protein BIM has also been shown to double-bolt lock to anti-apoptotic proteins however, the novel binding sequence in PUMA is unrelated to that in the CTS of BIM and functions independent of PUMA binding to membranes. Moreover, contrary to previous reports, we find that when exogenously expressed, the CTS of PUMA directs the protein primarily to the endoplasmic reticulum (ER) rather than mitochondria and that residues I175 and P180 within the CTS are required for both ER localization and BH3-mimetic resistance. Understanding how PUMA resists BH3-mimetic displacement will be useful in designing more efficacious small-molecule inhibitors of anti-apoptotic BCL-2 proteins.
This is an important study investigating interactions of the pro-apoptotic PUMA with anti-apoptotic BCL-2 proteins. The authors demonstrate convincingly that the PUMA/BCL-2 interactions are mediated not only via BH3-domain interaction, but also depend on a C-terminal sequence of PUMA similar to BIM. Unexpectedly they find PUMA is often localising to the ER. This manuscript is important for researchers focusing on cell death and anti-tumor drug development.https://doi.org/10.7554/eLife.88329.sa0
Apoptosis, a form of programmed cell death, is an essential physiological process responsible for the elimination and disposal of malignant or excessive cells in multicellular organisms (Fuchs and Steller, 2011). The loss of outer-mitochondrial membrane integrity, known as mitochondrial outer-membrane (MOM) permeabilization (MOMP), is regarded as an irreversible event, resulting in the release of cytochrome c and apoptotic factors from the mitochondria that leads to the activation of caspases and cell death (Kale et al., 2018). MOMP is tightly regulated by BCL-2 protein family proteins, including both pro- and anti-apoptotic regulators that share from one to four homology motifs (BH1 to BH4). BCL-2 and its anti-apoptotic homologs possess all four BH motifs and keep the effector pore-forming multi-BH domain pro-apoptotic proteins, BAX and BAK, from inducing MOMP through direct binary protein-protein interactions (Kale et al., 2018; Bogner et al., 2020). Anti-apoptotic proteins can also prevent MOMP by binding BH3-only proteins, which are pro-apoptotic proteins that share only the BH3 region with other BCL-2 family proteins. The interaction of pro-apoptotic proteins with anti-apoptotic proteins, which is often but not always binary, results in a mutual sequestration that inhibits both proteins. BH3-only ‘activator’ proteins, like BIM and BID, bind to and activate BAX and BAK. BH3-only ‘sensitizer’ proteins, like BAD and HRK, bind to and inhibit anti-apoptotic proteins but do not activate BAX or BAK. Unlike the other BH3-proteins where there is a clear distinction, whether the BH3-protein PUMA functions primarily as an inhibitor of anti-apoptotic proteins (sensitizer) or as an activator of BAX and BAK remains unclear and may depend on the cell type analyzed. Although first identified as a p53 transcriptional target (Yu et al., 2001; Nakano and Vousden, 2001), PUMA has been characterized as a mediator of cell death induced by DNA damage, endoplasmic reticulum (ER) stress and oxidative damage. All these stresses are induced by common chemotherapeutics and ionizing radiation (Jeffers et al., 2003; Jiang et al., 2006) suggesting that PUMA may play a role in chemotherapy responses. As the overexpression of pro-survival BCL-2 proteins is not only a hallmark of cancer progression but also critical for tumor cells to sustain a high proliferative rate and survive genomic instability, apoptosis modulation has been proposed as a therapeutic approach to selectively target cancer cells for elimination (Hanahan and Weinberg, 2011; Delbridge et al., 2016).
The development of small molecule inhibitors that mimic the BH3-motif of pro-apoptotic BCL-2 family proteins and function as competitive inhibitors for BH3-protein binding to anti-apoptotic proteins (BH3-mimetics) is an area of active academic and pharmaceutical research. BH3-mimetics result in the death of cancer cells that depend on anti-apoptotic proteins for survival (Delbridge et al., 2016). Among BH3-mimetics, the selective BCL-2 inhibitor, Venetoclax (ABT-199) is leading advancement to the clinic, and is FDA approved for the treatment of relapsed chronic lymphocytic leukemia and acute myeloid leukemia in patients not fit for standard induction chemotherapy (Delbridge et al., 2016; Roberts et al., 2016). Navitoclax (ABT-263), a BH3-mimetic that inhibits BCL-2, BCL-XL, and BCL-W results in undesirable but manageable on-target thrombocytopenia due to the dependence of platelets on BCL-XL (Tse et al., 2008). Unexpectedly complexes of anti-apoptotic proteins bound to PUMA and BIM, are highly resistant to all known BH3-mimetics (Aranovich et al., 2012; Liu et al., 2019). Resistance can be partly attributed to membrane binding by both the BH3-protein and its anti-apoptotic target increasing the local concentrations and therefore binding interactions (Liu et al., 2019; Pécot et al., 2016). However, we recently demonstrated that in addition to the BH3-sequence, there is a second anti-apoptotic protein binding motif in the carboxyl-terminal sequence (CTS) of BIM (Liu et al., 2019). Together the two binding sequences enable BIM to ‘double-bolt lock’ to anti-apoptotic proteins. The increased avidity that results from two independent binding sequences is presumed to confer resistance to displacement by BH3 mimetics.
Similar to BIM, PUMA is an intrinsically disordered protein containing a BH3 motif important for binding other BCL-2 family proteins and a CTS reported to function as a tail-anchor that integrates the protein in the MOM (Rogers et al., 2014; Wilfling et al., 2012). However, unlike conventional tail-anchor sequences and the CTS of BIM, the CTS of PUMA contains multiple prolines and charged residues, and an unusually short span of hydrophobic amino acids (Borgese et al., 2003; Mehlhorn et al., 2021). Therefore, we investigated whether the unusual CTS of PUMA contributes to resistance to BH3-mimetics. Our results suggest that similar to BIM, PUMA ‘double-bolt locks’ the anti-apoptotic proteins BCL-XL, BCL-2 and BCL-W. Unexpectedly, in multiple cell types, exogenously expressed PUMA is primarily localized at the ER and not the mitochondrial outer membrane (MOM). Furthermore, replacing the PUMA CTS with ER-specific tail-anchor sequences from other proteins resulted in PUMA mutants that bound specifically to ER membranes, retained pro-apoptotic function as a sensitizer binding to anti-apoptotic protein(s) but were not resistant to BH3-mimetics, suggesting that the sequence of the PUMA CTS rather than membrane binding is responsible for BH3-mimetic resistance. Mutagenesis and live cell experiments identified PUMA CTS residues I175 and P180 as required for both ER localization and resistance to BH3-mimetics. Our data indicate that inhibition of both BH3 and CTS binding to anti-apoptotic proteins may be required to mobilize PUMA for the treatment of cancer.
The PUMA CTS contributes to BH3-mimetic resistance independent of membranes in vitro
To understand how PUMA resists BH3-mimetic displacement from BCL-XL, full-length recombinant BCL-XL (CAA80661) and full-length PUMA (AAB51243) were purified, and their interaction was measured in vitro using Förster resonance energy transfer (FRET) (Pogmore et al., 2016). Previously, it was shown that PUMA and BCL-XL complexes are highly resistant to BH3-mimetic displacement when bound to membranes (Pécot et al., 2016), and that mutations in the BCL-XL CTS abolished both membrane binding and BH3-mimetic resistance (Pécot et al., 2016). However, PUMA also contains a CTS that binds the protein to membranes (Wilfling et al., 2012; Yee and Vousden, 2008) and it remains unclear whether it contributes to BH3-mimetic resistance (Liu et al., 2019). Therefore, PUMA with a deletion of the last 26 amino acids (PUMA-d26) that encompasses the CTS was also purified. To measure function, purified PUMA protein was added to a 'SMAC-mCherry MOMP assay', as previously described (Chi et al., 2020). In brief, mitochondria were isolated from baby mouse kidney (BMK) cells deficient for Bax and Bak (BMK-dko cells) expressing a fluorescent protein (FP), mCherry, fused to the N-terminal sequence of SMAC (SMAC-mCherry). SMAC-mCherry localizes to the intermembrane space of mitochondria, and is released upon MOMP. Isolated SMAC-mCherry containing mitochondria were reconstituted with recombinant BCL-2 family protein(s), and % SMAC-mCherry release (MOMP) was measured by monitoring SMAC-mCherry fluorescence in the supernatant after pelleting mitochondria, relative to the detergent-induced full permeabilization.
As expected, the addition of each BH3-protein alone was insufficient to induce MOMP as indicated by the low % Smac-mCherry Release (≈20%) (Figure 1A, columns 2,3,9,10). However, the combination of recombinant tBID [4 nM] and BAX [20 nM] was sufficient to induce MOMP at ≈60% release (column 4), which was inhibited by [10 nM] of BCL-XL (column 5). Addition of the BCL-XL inhibitor BAD [50 nM] (column 6), recombinant PUMA (column 7) or PUMA-d26 (column 8) was sufficient to functionally inhibit BCL-XL, thereby inducing SMAC-mCherry release (MOMP) in a BAX-dependent manner (Figure 1A). This confirms that recombinant PUMA and PUMA-d26 function as sensitizers by inhibiting BCL-XL to indirectly activate BAX.
We next tested PUMA pro-apoptotic function in live cells by transient transfection of HEK293, Baby mouse kidney (BMK) and HCT116 cell lines and assessed cell death in the transfected cells by Annexin V staining (Figure 1—figure supplement 1A). Here we use a Venus-fused to the N-terminus of the BH3-only protein BIM (VBim) or PUMA (VPUMA) to identify the transfected cells expressing the BH3-proteins by fluorescence intensity. As we previously published (Chi et al., 2020), HEK293 cell lines are unprimed and can only be killed efficiently by an activator protein such as BIM (60% Annexin V positive), whereas PUMA and the previously established sensitizer-only BIM mutant (BIM-dCTS) did not induce cell death in HEK293. On the other hand, BIM, PUMA and PUMA-d26 killed BMK and HCT116, which are primed cells that can be killed by a sensitizer protein (Chi et al., 2020). Similar to sensitizer proteins, BH3-mimetics are designed to target anti-apoptotic proteins and do not activate BAX or BAK. Thus, we used a combination of two BH3-mimetics S63845 (MCL-1 inhibitor) and ABT-263 (BCL-XL, BCL-2, and BCL-W inhibitor) to validate the apoptotic priming of these stable cell lines by doing a two-way titration of the two drugs (Figure 1—figure supplement 1B). Upon BH3-mimetics treatment, activator proteins and/or active BAX/BAK are displaced from the antiapoptotic proteins in the primed cells resulting in cell death. The combination of these drugs at 10 µM resulted in 47% and 73% Annexin V positivity in HCT116 and BMK, respectively, but did not increase Annexin V positivity in HEK293, confirming that HEK293 is the only unprimed cell line among the three. Taken together, the data show that PUMA acts as a sensitizer and that its CTS is not required for sensitizer function, as truncated PUMA without its CTS can still bind to and inhibit BCL-XL.
Binding of recombinant Alexafluor 647 (A*647) labeled BCL-XL to Alexafluor 568 (A*568) PUMA and PUMA-d26 was measured directly by FRET in the presence of liposomes with a phospholipid composition similar to that of mitochondria (Figure 1B–E). In the absence of BH3-mimetics, (indicated as Concentration, 0 nM) the FRET efficiency was ~40% between the donor PUMA*A568 and the acceptor BCL-XL*A647. The addition of the selective BCL-XL BH3-mimetic, A-1155463, did not result in a significant decrease in FRET efficiency, demonstrating that binding of PUMA*A568 to BCL-XL*A647 remains unchanged (Figure 1B, black line). A FRET efficiency of ~40% was also measured with the same concentrations of PUMA-d26*A568 incubated with BCL-XL*A647, indicating similar protein binding (Figure 1B, grey line). However, the addition of low concentrations of A-1155463 reduced FRET efficiency to less than 10%, demonstrating BH3-mimetic mediated displacement of PUMA-d26*A568 from BCL-XL*A647 (Figure 1B). Thus, removing the PUMA CTS increases PUMA susceptibility to BH3-mimetic displacement from BCL-XL.
When the same experiment was performed in solution, full length PUMA*A568 was more resistant to BH3-mimetic displacement than PUMA-d26*A568 suggesting that the CTS of PUMA contributes to BH3-mimetic resistance by binding to BCL-XL even in the absence of membranes (Figure 1C). To corroborate the result the experiment was repeated using the less potent but better studied BCL-XL inhibitor ABT-263. In the presence of liposomes, both PUMA*A568 and PUMA-d26*A568 resisted displacement from BCL-XL*A647 by ABT-263 (Tao et al., 2014; Figure 1D) but in solution, PUMA-d26*A568 was displaced from BCL-XL*A647 (Figure 1E), consistent with the CTS of PUMA contributing to BH3-mimetic resistance by binding to BCL-XL independent of binding PUMA to membranes.
The PUMA CTS contributes to BH3-mimetic resistance in live cells
To determine if our findings with purified proteins replicate what occurs in live cells, we used quantitative fast fluorescence lifetime imaging microscopy – Förster resonance energy transfer (qF3) to measure PUMA binding to anti-apoptotic proteins (Osterlund et al., 2022). For these measurements, the donor fluorescent protein mCerulean3 was stably expressed as a fusion to the N-terminus of the indicated anti-apoptotic protein in BMK-dko cells in which the Bax and Bak genes are deleted. The acceptor protein of the FRET pair - Venus (V) fused to the N-terminus of PUMA (VPUMA) was expressed in the cells by transient transfection. Four hours later the media was exchanged to add the indicated BH3-mimetic or DMSO as a solvent control. Twenty hr later, the cells were analyzed by qF3. To assess the effect of mutations in PUMA on binding to the anti-apoptotic proteins, BMK-dko cell lines expressing either CBCL-XL, CBCL-2, or CBCL-W were transfected with plasmids encoding variants of VPUMA as previously described (Pemberton et al., 2019). In these assays, binding of both VPUMA and VPUMA-d26 to CBCL-XL, CBCL-2 and CBCL-W was established with approximately equal apparent dissociation constants (6–9 μM) by ensuring that in the cell images analyzed the expression level of the donor exceeded the absolute dissociation constant (Figure 2A, DMSO lanes) (Osterlund et al., 2022). In these experiments, the non-binding mutants with a 4E mutation within the BH3 motif (BH3-4E) of VPUMA and VPUMA-d26 were used to control for collisions as opposed to binding interactions (Figure 2A, bottom panel of heatmap).
Regions of interest (ROIs) in the qF3 images were identified automatically and fluorescence lifetimes were calculated as previously (Osterlund et al., 2022). The decrease in the 3.8ns fluorescence lifetime in CBCL-XL expressing cells (blue) to less than 3.4 ns in cells expressing VPUMA is indicative of a binding interaction (Figure 2B, red cells). However, to clearly differentiate binding from collisions that can occur at high frequency for colocalized proteins it is necessary to demonstrate that the qF3 data can be better fit to Hill equation than to a straight line. For this purpose, values that are directly related to the fraction of donor molecules in the bound state, the fractional change in angular frequency (Δω) obtained from polar plots of the data, were plotted as a function of the concentrations of the unbound donor (Free Venus) and fit to a Hill equation to enable calculating apparent dissociation constants for each binding curve (Osterlund et al., 2022).
Addition of the dual BCL-2/BCL-XL inhibitors ABT-263 [30 μM], AZD-4320 [20 μM] or a newer BCL-XL inhibitor related to A-1155463, A-1331852 [2.5 μM] caused increases in the apparent dissociation constants that were more pronounced for VPUMA-d26 than for VPUMA binding to CBCL-XL (Figure 2A). This result demonstrates that these BH3-mimetics displaced truncated PUMA (PUMA-d26), to a much greater extent than full-length PUMA. The most dramatic change in apparent Kd was for A-1331852 which severely reduced the affinity for VPUMA-d26 binding to CBCL-XL resulting in an apparent Kd increase from ~9 to~37 μM. In contrast, measurements of the effect of the same inhibitor on binding of full-length VPUMA to CBCL-XL revealed much less but detectable displacement (the apparent Kd increased from ~6 to~13 μM) (Figure 2A). Consistent with these results, in live cells, A-1331852 is more selective and more potent than either AZD-4320 or ABT-263 (Osterlund et al., 2022).
The inhibitors ABT-263 and AZD-4320 are BH3-mimetics reported to target BCL-XL, BCL-2 and BCL-W (Tse et al., 2008; Leverson et al., 2015) and BCL-XL and BCL-2, (Leverson et al., 2015) respectively. When these drugs were used to probe PUMA binding to BCL-2, surprisingly ABT-263 had no effect and AZD-4320 only partially reduced the binding of VPUMA-d26 with CBCL-2, indicating that deletion of the PUMA CTS alone is not sufficient to make PUMA susceptible to displacement from BCL-2 (Figure 2—figure supplement 1). None of the inhibitors reduced binding of VPUMA-d26 to BCL-W (Figure 2—figure supplement 1), however that might be due to poor binding of the inhibitors to the anti-apoptotic protein rather than to the binding affinity of PUMA for BCL-W (Osterlund et al., 2022).
Unexpectedly, fusing Venus to the C-terminus of PUMA (PUMAV) resulted in a higher FRET efficiency with anti-apoptotic proteins (indicated by the increase in Δω), likely due to increased proximity between the donor and acceptor fluorescence proteins in the complex and inconsistent with the PUMA CTS spanning a membrane (Figure 2—figure supplement 2). A larger change in Δω results in a greater dynamic range in the assay and therefore the possibility of detecting smaller changes in binding. Both PUMAV and PUMA-d26V bound to anti-apoptotic proteins in a BH3-dependent manner (Figure 2—figure supplement 2). Similar to the results obtained above, when binding to CBCL-XL was measured, PUMAV resisted BH3-mimetic displacement, while PUMA-d26V was displaced (Figure 2C-E, Figure 3). Indeed, binding was reduced to the point where the change in Δω did not return to the level seen for the bound state even at 45 μM free PUMAV (Figure 2C, blue line). Similar to the results with Venus fused to the amino terminus, none of the mimetics displaced PUMA-d26V from BCL-W (Figure 2D) consistent with relatively poor binding of the mimetics to BCL-W. However, for BCL-2 visual inspection of the binding curves revealed that the mimetics, particularly AZD-4320, partially displaced PUMA-d26V (Figure 2E). Taken together our data indicate that the PUMA CTS contributes to PUMA binding to BCL-XL and BCL-2 in live cells.
In cells PUMA resists BH3-mimetic displacement by double-bolt locking to BCL-2 family anti-apoptotic proteins
The resistance of PUMA to displacement from anti-apoptotic proteins by BH3-mimetics is similar to recently published results for the BH3-protein BIM (Aranovich et al., 2012; Liu et al., 2019; Pécot et al., 2016) yet the CTS of PUMA is very different than the CTS of BIM. To determine whether PUMA is also ‘double-bolt locked’ to anti-apoptotic proteins by the BH3-motif and CTS we constructed new mutants of PUMA. These mutants have the BH3-motif mutated (Lock 1), the CTS deleted (Lock 2), or both (Lock 1 and 2) mutated (Figure 3A). To disable the PUMA BH3 region sufficiently to measure the importance of the CTS separately but without abolishing the interaction entirely we replaced PUMA (residues 133–151) with the BH3 region from the protein BID. Unlike PUMA, BID is a pro-apoptotic BH3-protein that is easily displaced from anti-apoptotic proteins by BH3-mimetics (Aranovich et al., 2012; Pécot et al., 2016). The resulting protein, PUMA(BID-BH3)V has a compromised BH3-motif, but an intact CTS (Figure 3A). To disable both Lock 1 and 2 the CTS was deleted from PUMA(BID-BH3)V. The resulting mutant, PUMA(BID-BH3)-d26V, has a compromised BH3-motif and is missing the PUMA CTS (Figure 3A).
Constructs encoding these proteins were transfected into BMK-dko cells expressing one of CBCL-XL, CBCL-2 or CBCL-W, and after ~20 hr, binding interactions at equilibrium in live cells were measured by qF3. In the DMSO controls, PUMAV bound with Kds of ~4–6 μM to all three anti-apoptotic proteins while the mutants bound with Kds of ~5–10 μM (Figure 3B). As expected, addition of BH3-mimetic did not increase the dissociation constants for PUMAV substantially, although there was detectable inhibition of BCL-XL by A-1331852 and BCL-2 by AZD-4320 (Figure 3B). Also as anticipated, deletion of the PUMA CTS (PUMA-d26V) resulted in increased BH3-mimetic mediated displacement from BCL-XL, but not for BCL-2 and BCL-W as reported above (compare Figure 2A and Figure 3B). As a positive control for BH3-mimetic mediated displacement, we included Venus fused to the N-terminus of truncated BID (VtBID). This protein was displaced from CBCL-XL, CBCL-2 and CBCL-W by the cognate BH3-mimetic to a similar extent as PUMA-d26V (Figure 3B, far-right column). Replacing the BH3 region of PUMA with that of BID (PUMA(BID-BH3)V) was sufficient to result displacement similar to that of VtBID from CBCL-XL with the inhibitor A-1331852 and from BCL-2 with AZD-4320. However, in some cases, PUMA(BID-BH3)V was more resistant than VtBID to displacement by the cognate BH3-mimetics (e.g. CBCL-2 with ABT-263 and CBCL-W with AZD-4320). In contrast, resistance to displacement from CBCL-XL and CBCL-2 was abolished by mutation of Lock 1 and Lock 2 (PUMA(BID-BH3)-d26V). While resistance to AZD-4320 was reduced for PUMA(BID-BH3)-d26V for binding to CBCL-W the apparent Kd remained substantially lower than for VtBID. Overall, these data suggest that both the PUMA BH3 and CTS regions contribute to the high affinity interaction of PUMA with anti-apoptotic proteins, and that the primary mechanism of resistance to BH3-mimetics is a ‘double-bolt lock’, similar to BIM (Liu et al., 2019).
The CTS of PUMA localizes the protein to the endoplasmic reticulum
The data above demonstrate a previously unrecognized function of the PUMA CTS in double-bolt locking to BCL-XL and BCL-2 and that PUMA-BCL-XL complexes can form in the absence of membranes. Although, BCL-XL is found in both the cytoplasm and bound to intracellular membranes, previous reports suggest that PUMA is localized in a CTS dependent manner primarily to mitochondria (Yu et al., 2001; Nakano and Vousden, 2001; Wilfling et al., 2012; Yee and Vousden, 2008). However, in these reports, localization was interpreted by visual inspection and was not rigorously analyzed. To our surprise, although the spatial resolution of FLIM-FRET images is limited, VPUMA-CBCL-XL complexes in FLIM-FRET images were not obviously localized only at mitochondria (Figure 2B). This could be due to complexes forming and remaining cytoplasmic; however, previous reports suggest that binding to membranes increases the stability of the complexes and contributes to the resistance we observed to displacement by BH3 mimetics (Pécot et al., 2016). Therefore, we re-examined PUMA localization using confocal microscopy and created mutants to identify the CTS residues that mediate binding to subcellular membranes.
To assign the localization of PUMA in live cells using an unbiased quantitative approach we made use of a previously described random forests classifier built from a reference library of 789,011 optically validated landmark-based localization images (Schormann et al., 2020). Briefly, NMuMG (normal murine mammary gland) cells were infected with lentivirus to express a fusion protein consisting of EGFP fused to the N-terminus of a BH3-4E mutant PUMA to visualize the fusion protein (EGFPPUMA-4E) without it binding to anti-apoptotic proteins or killing the cells. Using automated-confocal microscopy, 2225 images of individual cells expressing EGFPPUMA-4E were classified as the landmark from the reference library they most resemble. To our surprise, 55% of EGFPPUMA-4E cell micrographs were classified as most similar to one of the resident endoplasmic reticulum (ER) markers (blue bars), indicating ER localization (Figure 4A). In contrast, smaller fractions of the cells were classified as patterns resembling that of a protein that recycles between the ER and Golgi (Calr-KDEL) or that is resident at mitochondria (MAO, monoamine oxidase). For images of other cells classification was to one of a variety of other locations particularly transport vesicles, a phenomenon not uncommon for over-expressed proteins (Schormann et al., 2020). As a positive control for correct classification of an ER localized EGFP-fused BH3-protein, we infected NMuMG cells with lentivirus to express EGFPBIK-L61G a mutant of the tail-anchored ER protein BIK containing the BH3 mutation L61G previously shown to abrogate apoptotic activity (Mathai et al., 2002). As expected, 79% of the images of EGFPBIK-L61G expressing cells were classified as showing ER localization (yellow bars) compared to 55% for EGFPPUMA-4E (Figure 4A).
As an orthogonal approach to examine localization in the BMK-dko cell line used for FLIM-FRET experiments, co-localization with ER and mitochondrial markers was examined by calculating Pearson’s correlation coefficients. Because it lacks both BAX and BAK, the BMK-dko cell line will not undergo apoptosis in response to expression of PUMA or BIK. Therefore, inactivating mutations that might affect localization are not required. To calculate correlation values between PUMA and landmarks for two locations in the same cell, the fluorescent protein mCerulean3 was fused to the N-terminus of PUMA (CPUMA). The ER membrane and mitochondria were visualized using the dyes BODIPY FL thapsigargin (green) and MitoTracker Red, respectively. Individual cells were identified using far red nuclear stain DRAQ5. Images of cells were obtained by automated confocal microscopy (Opera Phenix, PerkinElmer) and analyzed with Harmony software (V4.9). To assess localization to the ER membrane, Pearson’s correlation coefficients were calculated between BODIPY FL thapsigargin and mCerulean3 fluorescence intensity for each cell. As seen in Figure 4B, the expression of the negative control mCerulean3 alone resulted in a median coefficient close to zero, indicating no correlation. The positive control for ER localization, mCerulean3 fused to the N-terminus of BIK (CBIK), resulted in a median coefficient close to 0.75, suggesting this value represents localization at the ER membrane. Both CPUMA and CPUMA-4E have Pearson’s coefficient values with BODIPY FL thapsigargin of ~0.6 similar to what was seen for CBIK. In contrast, the Pearson’s correlation coefficient for CPUMA-d26 with BODIPY FL thapsigargin was similar to the negative control (Figure 4B) suggesting CPUMA-d26 is located in the cytoplasm. Overall, this data indicates that exogenously expressed PUMA localizes to the ER in a BH3-independent, CTS-dependent manner.
The same approach was used to assess protein localization at the mitochondria. As expected, the Pearson’s correlation coefficients calculated between mCerulean3 alone and MitoTracker Red resulted in a median coefficient close to zero, indicating no correlation. As a positive control for perfect mitochondrial localization, the calculated median coefficient between the dyes MitoTracker Green and MitoTracker Red was close to 0.85 (Figure 4B). As expected, the median Pearson’s coefficient for the ER localized control CBIK with MitoTracker Red was only slightly higher than the cytoplasmic control mCerulean3 alone. The Pearson’s correlation coefficients for both CPUMA and CPUMA-4E with MitoTracker Red were less than 0.2, much lower than those calculated for these proteins with the ER marker BIK (Figure 3B, compare the bottom and top panels), but slightly higher than values obtained for the cytoplasmic protein mCerulean3. Indeed, the Pearson’s correlation coefficients for CPUMA and CPUMA-4E with MitoTracker Red were similar to the values obtained for the ER marker CBIK and MitoTracker Red. Consistent with these observations, the Pearson’s correlation coefficients for CPUMA-d26 (lacking the CTS) and mCerulean3 with mitochondria were both close to zero (Figure 4B, lower panel). Finally, images of CBCL-XL, a protein known to be both cytoplasmic and localized to multiple membranes resulted in intermediate Pearson’s correlation coefficients. Together, this data suggests that when exogenously expressed the PUMA CTS localizes the protein primarily to the ER.
To further dissect the PUMA CTS and identify the region responsible for directing PUMA localization additional mutants were generated. Incremental deletions of the PUMA CTS decreased the Pearson’s correlation coefficient of the mutant proteins with ER localization suggesting that deleting as few as the last 11 amino acids (VPUMA-d11) impacts ER localization (Figure 4D). As progressive deletion of the CTS correlated with decreasing ER localization, this suggests that the entire CTS contributes to PUMA localization. Unexpectedly, the fusion protein with Venus at the C-terminus of PUMA (PUMAV) retained ER localization (Figure 4C and D), suggesting that unlike CTS of BIK which fully inserts into the ER (Wilfling et al., 2012) the CTS of PUMA is not a conventional tail-anchor that spans the membrane bilayer. This conclusion is also congruent with the data indicating that the PUMA CTS sequence is too short to span the ER membrane.
Our data showed that exogenously expressed PUMA mostly localizes to the ER while BCL-XL localized to both mitochondria and ER (Figure 4B; Osterlund et al., 2022; Kaufmann et al., 2003; Osterlund et al., 2023). Despite this, in BMK cells, PUMA bound to BCL-XL with an apparent Kd of 5.5 µM (Figure 2). To further investigate the subcellular localization of PUMA:BCL-XL heterodimers, FLIM-FRET experiments with VPUMA were done using MCF-7 cells stably expressing CBCL-XL and as a localization marker, mCherry fused to the tail anchor sequence of ActA (ChActA) or of Cytochrome-b5 (ChCb5). This enables the differentiation of the mitochondrial (ChActA) and ER (ChCb5) subcellular regions within the cells where VPUMA could interact with CBCL-XL facilitating localization of complexes. The mCherry channel was used to generate a mitochondrial or ER ‘mask’ and select ROIs for which the mCerulean3 fluorescence lifetime and % FRET were calculated. The plateau observed for the data indicate that both PUMA and PUMA-d26 bind to BCL-XL at both the ER and the mitochondria. In contrast, the FLIM data for the BH3-4E negative controls (VPUMA-4E and VPUMA-d26-4E) can be fit to a straight-line, indicating collisions (Figure 4—figure supplement 2B). The simplest explanation for exogenously expressed VPUMA binding to BCL-XL at mitochondria is that BH3-dependent binding of PUMA to mitochondrial localized BCL-XL results in localization of VPUMA at or close to mitochondria. Supporting this hypothesis, immunofluorescence staining in MCF-7 cells for endogenous PUMA which is expressed at levels more similar to that of BCL-XL and BCL-2 in these cells (Antony et al., 2012; Mukherjee et al., 2015) revealed that endogenous PUMA co-localized more with the mitochondrial landmark (ChActA) (Pearson correlation coefficient ≈ 0.6) than the ER landmark (ChCb5) (Pearson correlation coefficient ≈ 0.4; Figure 4—figure supplement 3C, column 4, 8, and 12).
Given the relatively lower co-localization of endogenous PUMA to the ER marker in MCF-7 cells, we examined localization of endogenous PUMA after the induction of PUMA expression by ER stress (Reimertz et al., 2003; Yu and Zhang, 2008) or genotoxic stress (Meyerkord et al., 2008; Jamil et al., 2015). Previous results suggest that a P53-dependent response to genotoxic stress results in the extension of peripheral tubular ER and promotes the formation of ER-mitochondrial contact sites (Wang et al., 2007) which may be binding sites for PUMA. Therefore, to test this hypothesis, immunofluorescence staining for endogenous PUMA was done in ChActA (mitochondria marker) or ChCb5 expressing (ER marker) MCF-7 cells treated with Thapsigargin (TG) or Tunicamycin (TN) to induce ER stress and Etoposide (ETOP) to induce genotoxic stress (Figure 4—figure supplement 3A). As expected, we observed an increase in the intensity of the Alexa-488 immunofluorescence signal for PUMA in cells treated with the drugs compared to the DMSO-treated cells (Figure 4—figure supplement 3B). Intriguingly, we saw an increase in Pearson correlation coefficients for PUMA to both the ER and the mitochondria, most notably at the highest concentrations of the 3 drugs (Figure 4—figure supplement 3C). As genotoxic stressors can also induce expression of the anti-apoptotic BCL-XL (Jamil et al., 2015) which is both ER and mitochondria-localized (Figure 4B), we speculate that binding to BCL-XL contributes to increased localization of PUMA with the mitochondrial marker ChActA. Moreover, stress induced changes in ER structure and increased MAMs would also result in apparent localization at mitochondria. Taken together, these data suggest that PUMA subcellular localization in living cells is dynamic and dependent not only on the inherent specificity of the PUMA CTS but also on the abundance and localization of its binding partners. This phenomenon has been observed with another BH3-only protein, BIK whereby the predominantly-ER localized BIK can localized to the mitochondria in BMK-dko upon expression of a mitochondria-localized BCL-XL mutant (Osterlund et al., 2023).
Restoring ER localization to PUMA-d26 does not result in BH3-mimetic resistance
Together the data in Figures 1—3 demonstrate that the CTS of PUMA is required for the protein to resist BH3-mimetic-mediated displacement from BCL-XL in vitro and in live cells. For purified proteins resistance to BH3-mimetic displacement is independent of binding of the proteins to membranes. However, it remains possible that in live cells, PUMA-d26 no longer resists BH3-mimetic displacement because the protein no longer binds subcellular membranes. To test this hypothesis, we constructed two mutants in which the CTS of PUMA was replaced by the tail-anchor sequences of the two proteins used as ER localized controls: BIK (VPUMA-d26-ER1) or cytochrome b5 (CB5) (VPUMA-d26-ER2) (Figure 5A, Figure 5—figure supplement 1A).
To determine if these fusion proteins correctly localize to the ER, they were expressed by transient transfection of the corresponding constructs into BMK-dko cells stably expressing CBIK as an ER localization marker (labelled as ER Marker in Figure 5C). The cells were stained with the nuclear dye DRAQ5, imaged by automated confocal microscopy and co-localization was assessed using Pearson’s correlation coefficients, as described above. Both VPUMA-d26-ER1 and VPUMA-d26-ER2 generated Pearson’s correlation coefficients with the ER localization marker that were substantially greater than those for either VPUMA-d26 or Venus alone with the same ER marker (Figure 5B and C). In addition to correct localization, exogenous expression of both fusion proteins induced apoptosis in BMK-wt cells, in a BH3-dependent manner, as measured by Annexin V positivity using confocal microscopy (Figure 5—figure supplement 1B). Given that the fusion proteins are both functional and correctly localized at ER membranes, we used qF3 to measure binding with CBCL-XL in live cells. As seen in Figure 5C, both VPUMA-d26-ER1 and VPUMA-d26-ER2 bound to CBCL-XL (DMSO column) in a BH3-dependent manner in BMK-dko cells. However, neither complex was resistant to displacement by the addition of ABT-263, A-1331852 or AZD-4320. Addition of any of these drugs resulted in significantly higher dissociation constants for binding to CBCL-XL for VPUMA-d26-ER1 and VPUMA-d26-ER2 compared to VPUMA. In addition to binding to CBCL-XL, VPUMA, VPUMA-d26-ER1 and VPUMA-d26-ER2 also bound to CMCL-1 in live BMK-dko cells. Similar to the results obtained for binding to CBCL-XL, the addition of MCL-1 specific BH3-mimetics; S63845 and S64315, resulted in higher dissociation constants for binding to MCL-1 by both mutant proteins compared to VPUMA (Figure 5C). However, these data also demonstrate that the MCL-1 inhibitors only partially displaced VPUMA-d26-ER1 and VPUMA-d26-ER2 compared to mutation of the PUMA BH3 sequence. Thus, similar to BCL-2 and BCL-W the BH3 sequence of PUMA is sufficient to confer partial resistance of binding to MCL-1 to the drugs.
Thus, these data indicate that reintroducing ER membrane localization to VPUMA-d26 does not restore resistance to BH3-mimetic displacement, suggesting that specific residues or sequences within the PUMA CTS contribute to BH3-mimetic resistance.
Residues I175 and P180 in the PUMA CTS contribute to both ER localization and BH3-mimetic resistance
Data with the CTS substitutions above indicated that localization at the ER is not sufficient to confer resistance to BH3 mimetics. Thus, we sought to identify which regions and residues within the CTS of PUMA are required to confer resistance to BH3 mimetics and if such regions/residues are also required for membrane binding. To this end, mutants were generated containing serial deletions of the PUMA CTS sequence. These mutants were then used to measure binding to membranes and to CBCL-XL in live cells using qF3. Deletion of the last 11 residues (VPUMA-d11) had no effect on localization, binding to BCL-XL or resistance of the complex to BH3-mimetic inhibition (apparent Kd’s: DMSO = 6 μM, plus 2.5 mM A-1331852=14 μM) (Figure 6A–C). Deletion of the last 20 residues (VPUMA-d20) negatively impacted localization (Figure 4D) and substantially altered binding (∆ω). However, the data still resemble a binding curve and although there is substantial noise at low free Venus concentrations where there was no clear change in resistance to addition of BH3-mimetic (apparent Kd’s: DMSO = 2 μM, plus 2.5 mM A-1331852=13 μM) (Figure 6B). The observed change in ∆ω could be due to a conformational change in the protein that increased the distance or altered the dipole orientations between the FRET donor (CBCL-XL) and acceptor (VPUMA-d20) rather than a change in binding affinity. In contrast, deletion of the last 26 amino acids of PUMA (VPUMA-d26) eliminated both ER localization and BH3 mimetic resistant binding to CBCL-XL (apparent Kd’s: DMSO = 8 μM, plus 2.5 mM A-1331852=38 μM). Therefore, the residues between VPUMA-d11 and VPUMA-d20 affect localization and ∆ω, but the effect on BH3 mimetic resistance is less certain. Moreover, this region of the PUMA CTS is the most hydrophobic and contains two proline residues which are unusual for a membrane binding region. Therefore, to probe this region of the protein, nine mutants of PUMAV were generated in which individual residues were replaced with glutamic acid (E). Although replacement with E is a non-conservative mutation, for the DMSO controls similar dissociation constants were measured for all the mutants of PUMAV for binding to CBCL-XL (Figure 6C). Therefore, none of the mutations abolished binding of PUMA to BCL-XL. Addition of the potent BCL-XL inhibitor A-1331852 resulted in an increase in the apparent Kd’s for all the mutants. However, for I175E and P180E, the Kd’s increased substantially, indicating displacement from BCL-XL equivalent to that observed for PUMA-d26V (Figure 6C).
Visual inspection supported by Pearson’s correlation coefficients revealed that PUMAV mutants I175E and P180E were not localized to membranes but were located primarily in the cytoplasm (Figure 6D and Figure 6—figure supplement 3). In contrast, and consistent with the effects seen on binding to CBCL-XL, the other point mutations reduced PUMA localization at ER to a similar minor extent (Figure 6D). Therefore, in live cells residues I175 and P180 in the PUMA CTS appear to be required for both binding to membranes and resisting BH3-mimetic displacement. For the mutations with minor effects, there is also a correlation between membrane binding and BH3-mimetic resistance.
Substitution of amino acids in the PUMA CTS with negatively charged E residues are more likely to influence PUMA binding to membranes than more conservative substitutions such as alanine. Therefore, to determine if more conservative mutations would separate membrane binding from resistance to BH3-mimetic, additional PUMAV mutants I175A and P180A were analyzed for binding to CBCL-XL by FLIM-FRET and for localization by Pearson’s correlation with an ER marker in live BMK-dko cells. PUMAV mutant P180A appeared cytoplasmic, and resulted in a Pearson’s coefficient with the ER-localized control that was similar to that of the Venus cytoplasmic control (Figure 6E and Figure 6—figure supplement 3). This indicates residue P180 is required for PUMA binding to membranes in live cells. In contrast, mutant I175A appears membrane localized upon visual inspection, and generated a calculated Pearson’s coefficient with the ER-localized control that was higher than that of either PUMAV I175E or Venus (Figure 6E and Figure 6—figure supplement 3), but that was lower than the correlation between the ER-localized control and wildtype PUMAV (Figure 6E).
To compare the extent to which these mutants resist BH3-mimetic displacement from BCL-XL in live cells, FLIM-FRET measurements were made and % Resistance was calculated as previously described Liu et al., 2019; Pemberton et al., 2019. Briefly, the FLIM-FRET efficiency at the acceptor/donor ratio of two was calculated from the fitted Hill-slope equation for expression of the proteins alone and in the presence of the BH3-mimetic, and the change in these values was used to calculate the % Resistance (ie the % of complexes that remain intact). Therefore, a low % Resistance is indicative of protein displacement by BH3-mimetic treatment. In the presence of 2.5 μM A-1331852, PUMAV has a high calculated % Resistance (~90%), while PUMA-d26V has a lower % Resistance (~50%) (Figure 6F). PUMAV mutants I175E, I175A, P180E and P180A all have low % Resistance values similar to PUMA-d26V, suggesting that all of the mutant PUMAV proteins were displaced from BCL-XL similarly to PUMA-d26V. Even PUMAV I175A that localized partially to the ER (P=0.04) did not resist displacement (Figure 6E–F). Overall, this data suggests that PUMA CTS residues I175 and P180 are required for both PUMAV binding to membranes and for resisting BH3-mimetic induced displacement from CBCL-XL.
The availability of small-molecule inhibitors for anti-apoptotic proteins has generated new insights into the mechanisms underlying programmed cell death in a number of cell types, and has positively impacted the treatment of cancer (Delbridge et al., 2016). The development of the first true BH3-mimetic, ABT-737, was based on the structure of the BH3 sequence of BAD bound to BCL-XL and efficacy defined by displacement of BAD BH3-peptides from truncated BCL-XL in solution (Oltersdorf et al., 2005). The other BH3-mimetics currently under pharmaceutical development were all derived based on the structures of BH3-peptides and their corresponding binding sites in the different anti-apoptotic proteins. However, other factors that contribute to BCL-2 family interactions have been discovered. For instance, the MOM and the ER membrane act as both the platform and an active participant in BCL-2 family interactions (Pécot et al., 2016; Lovell et al., 2008; Bleicken et al., 2017). Not surprisingly, full-length proteins have different affinities than truncated proteins binding to BH3 peptides (Kale et al., 2018). This is most dramatically shown for the BH3 protein BIM binding to BCL-XL (Liu et al., 2019; Chi et al., 2020). Studies with truncated protein and BH3 peptides showed binding affinities in the micromolar to millimolar range while studies with full length proteins in the presence of membranes revealed a nanomolar affinity between BIM and BCL-XL (see also Figure 1). In the latter studies, both BH3 regions and the CTS previously thought exclusively involved in binding of BIM to membranes were discovered to contribute to BIM function. Moreover, both the BH3 sequence and the CTS are required for BIM to bind BCL-XL, forming a “double-bolt lock”, in a manner that is resistant to inhibition by BH3-mimetics (Liu et al., 2019).
Collectively the data presented here demonstrate that like BIM, the PUMA CTS functions in binding the protein to membranes and together with the BH3 sequence is required for BH3-mimetic resistant binding to anti-apoptotic proteins. However, unlike BIM, the CTS of PUMA localizes primarily at ER instead of mitochondria (Figure 4) and together with the PUMA BH3 region results in almost complete resistance to all currently known BH3 mimetics (Figure 5D). Endogenous PUMA was observed predominantly at mitochondria in MCF-7 cells (Figure 4—figure supplement 3), suggesting other factors besides the CTS may influence PUMA localization, such as the presence of PUMA-binding partners including the anti-apoptotic protein BCL-XL at both the ER and mitochondria (Figure 4B and Figure 4—figure supplement 2).
Unexpectedly, when analyzed using purified proteins the CTS of PUMA conferred resistance to BH3 mimetics in the absence of membranes, demonstrating that inhibition of anti-apoptotic proteins by PUMA does not require binding to membranes (Figure 1) and suggesting the PUMA CTS binds to BCL-XL. Furthermore, when the PUMA CTS sequence was replaced with classical tail-anchor sequences from each of two other proteins, the mutant proteins were localized at the ER in live cells but remained sensitive to displacement from BCL-XL by BH3 mimetics (Figure 5). Thus, we conclude that membrane binding is not sufficient for the PUMA CTS to confer resistance to displacement from BCL-XL by BH3 mimetics. Instead binding of both the BH3 region and CTS of PUMA to BCL-XL increases both the affinity and avidity of the interaction.
To our surprise delineating the specific residues involved in resistance to BH3 mimetics and localization of PUMA at membranes by mutagenesis revealed that residues I175 and P180 are required for both localization and resistance to BH3-mimetics. This suggests that, at least in live cells, binding to membranes and resistance to BH3-mimetics may not be separable. This is very different from the CTS of BIM in which separate regions of the CTS are involved in binding BIM to mitochondria and binding it to BCL-XL (Liu et al., 2019). It remains to be determined how residues I175 and P180 are involved in PUMA binding to both membranes and anti-apoptotic proteins. We speculate that for P180E or P180A mutation, the diminished membrane binding and resistance to BH3-mimetic displacement might be due to disruption of the structure of the PUMA CTS. As the CTS sequence contains 4 prolines in the last 14 amino acids and there are 6 prolines in the last 30 amino acids of PUMA the structure of this region is expected to be very unlike other tail-anchor sequences that localize proteins at the ER or mitochondria (Figure 5—figure supplement 1).
Unexpectedly, we find using two different quantitative approaches that the PUMA CTS localizes the protein primarily to the ER, and not the MOM as previously claimed (Wilfling et al., 2012). Random forest image classification is an unbiased approach to assess the localization of EGFPPUMA-4E in live NMuMG cells (Figure 4A). A caveat of this approach is that it assumes that the 4E mutation that disrupts the BH3 region and abrogated pro-apoptotic function has no effect on PUMA subcellular localization. Therefore, since binding to anti-apoptotic proteins such as BCL-XL and MCL-1 requires an intact BH3 region, localization due to binding these proteins would not be observed using EGFPPUMA-4E. For this reason, we also calculated Pearson’s Correlation values for the overlap between exogenously expressed CPUMA or VPUMA, with ER membrane and mitochondrial markers in live BMK-dko cells (Figure 4B, C and D). Moreover, fusion of the PUMA CTS (residues 167–193 of PUMA) to the carboxyl-terminus of the fluorescent protein Venus (VPUMA-CTS) and expression of this mutant in BMK-dko cells results primarily in ER localization (Figure 4—figure supplement 1), indicating that the CTS of PUMA alone is sufficient to direct ER localization. Although we have shown that PUMA with an N-terminal fluorescence protein for visualization is functional as a sensitizer and can activate BAX or BAK indirectly to kill cells, it remains possible that the absence of available pro-apoptotic protein binding partners (BAX and BAK) may alter PUMA localization. Nevertheless, taken together the two approaches provide strong evidence that in epithelial cells, in the absence of cellular stress, the majority of exogenously expressed PUMA is located at the ER. In contrast, endogenous PUMA was located primarily at mitochondria and when expression increased in response to stress agonists, PUMA localization was increased at both ER and mitochondria. We speculate that the localization of PUMA may be determined by the availability and localization of anti-apoptotic binding partners such as BCL-XL. It is also possible that the protein localizes at ER-mitochondrial contact sites similar to the BH3-protein BIK (Osterlund et al., 2023) or that localization at the ER may be required for a non-apoptotic function of the protein, sometimes referred to as a ‘day job’.
Our data demonstrate that PUMA residues I175 and P180 are required for BH3-mimetic resistance of PUMAV binding to CBCL-XL. However, in the absence of a BH3-mimetic, the entire CTS region is dispensable for binding of the two proteins as shown by binding of PUMA-d26V to CBCL-XL (Figure 6C). In contrast, disabling the BH3 region of PUMAV (BH3-4E mutation) resulted in no binding to anti-apoptotic proteins (Figure 6—figure supplement 1). This may suggest that a conformational change in either PUMA or the anti-apoptotic proteins that results from binding of the PUMA BH3 sequence increases the affinity of the interaction with the PUMA CTS. Further delineation of the molecular mechanism may require high resolution structure determination. Nevertheless, our data using purified proteins clearly demonstrate that high affinity binding of PUMA conferred by the CTS occurs in solution and strongly suggests direct binding of the PUMA CTS to BCL-XL is distinct from the BH3-binding site. Thus, PUMA CTS binding would increase both the affinity and avidity of the interaction via a mechanism that does not require PUMA binding to membranes (Figure 1). That BH3-mimetic resistant binding requires only PUMA and CBCL-XL in solution suggests that future structural studies may be possible.
Finally, our demonstration that PUMA-d26 binding to BCL-2 is more resistant to inhibition by BCL-2 specific BH3-mimetics suggests that the BH3 sequence of PUMA makes unique contacts with BCL-2 not seen with other BH3-proteins. Only when the BH3-motif is changed to a susceptible one (BID-BH3) and the CTS is deleted (PUMA(BID-BH3)-d26V) can BH3-mimetic treatment fully inhibit PUMA binding to BCL-2 (Figure 3B). Thus, it may be possible to design BH3-mimetics selective for inhibition of PUMA binding to specific anti-apoptotic proteins. PUMA(BID-BH3)V that contains the BH3-mimetic displaceable BH3 sequence and the PUMA CTS was more resistant than VtBID, to displacement by the cognate BH3-mimetics (e.g. CBCL-2 with ABT-263 and CBCL-W with AZD-4320). Together these results suggest a role for the PUMA CTS in binding to anti-apoptotic proteins that depends on BH3 sequence binding but is independent of the specific BH3 sequence. We speculate that binding of a BH3 sequence may lead to a conformational change in the anti-apoptotic protein that facilitates binding of the PUMA CTS. Finally, the unique amino acid sequence of the PUMA CTS argues that generating an efficient specific inhibitor of PUMA binding to anti-apoptotic proteins may require inhibiting binding of both the PUMA BH3-motif and its CTS.
Materials and methods
Purification of proteinsRequest a detailed protocol
Full length human BCL-XL and human BAX were purified as previously described (Pogmore et al., 2016). Full length PUMA(α isoform) was expressed in Bl21-AI Escherichia coli purchased from New England Biolabs, pelleted using centrifugation, then lysed with the following buffer: PUMA Lysis Buffer – pH 8.5, 20 mM HEPES, 50 mM NaCl, 0.2% w/v CHAPS, 20% w/v Glycerol, 50 mM Imidazole, 10 μg/mL DNase, 1 mM PMSF. The lysate was then centrifuged to pellet debris. The supernatant was loaded onto a chitin resin column, 2 mL total volume (New England Biolabs) pre-equilibrated with PUMA Lysis Buffer. The column was then washed with PUMA Chitin Wash Buffer - pH 8.5, 20 mM HEPES, 50 mM NaCl, 0.6% w/v CHAPS, 20% w/v Glycerol, 50 mM Imidazole. After washing, 10 mL of PUMA Chitin Cleavage Buffer was poured on top the resin allowing ~8 mL to flow through. PUMA Chitin Cleavage Buffer - pH 8.5, 20 mM HEPES, 10 mM NaCl, 1% w/v Triton-X-100, 20% w/v Glycerol, 50 mM Imidazole, 300 mM Hydroxylamine. The column was capped retaining ~2 mL of the cleavage buffer on top of the chitin resin and left at 4 °C for 48–72 hours.
For labelling PUMA with a fluorescent dye,~2 mL was eluted from the chitin column, and the hydroxylamine was removed by gel filtration on a 10 mL Sephadex G25 medium column (GE Healthcare life science) equilibrated with PUMA Chitin Cleavage Buffer without hydroxylamine. The concentration of the eluted PUMA (~0.5 mL) was determined by spectrophometry and then labeled by adding urea to a final concentration of 3 M, 4 x molar excess relative to total protein of TCEP and 20 x molar excess of AlexaFluor 568 C5 Maleimide (ThermoFisher Scientific). The labelling reaction was adjusted to pH 7.2 using 12 M HCl, and rotated at 37 °C for 3 hours. The ~0.5 mL labelling reaction or chitin column elution (for unlabeled protein) was diluted in 30 mL of PUMA Ni-Binding Buffer - pH 7.2, 20 mM HEPES, 100 mM NaCl, 1% w/v CHAPS, 20% w/v Glycerol, 15 mM Imidazole and loaded onto a 0.5 mL HisPur Ni-NTA resin column (ThermoFisher Scientific). After washing the Ni-NTA resin with PUMA Ni-Wash Buffer 1 pH 7.2, 20 mM HEPES, 50 mM NaCl, 0.4% w/v CHAPS, 20% w/v Glycerol, 20 mM Imidazole and then with PUMA Ni-Wash Buffer 2 pH 7.2, 20 mM HEPES, 50 mM NaCl, 20% w/v Glycerol, 20 mM Imidazole, 0.5 M guanidine hydrochloride (GdnHCl) PUMA protein was eluted from the Ni-NTA resin with PUMA Ni-Elute Buffer - pH 7.2, 20 mM HEPES, 10 mM NaCl, 20% w/v Glycerol, 300 mM Imidazole, 1 M GdnHCl. Protein was quantified using a Bradford assay, flash frozen and stored at minus 80 °C until use.
SMAC-mCherry Mitochondria Release AssayRequest a detailed protocol
SMAC-mCherry release assays were carried out as previously described (Niu et al., 2017). Briefly, BAX-/-/BAK-/- baby mouse kidney (BMK) cells stably expressing a fusion protein comprised of the mitochondrial import peptide of SMAC (amino acids 1–58) fused to the amino-terminus of the fluorescent protein mCherry were lysed by nitrogen cavitation at 150 psi for 10 min on ice in buffer containing 20 mM HEPES (pH 7.2), 250 mM sucrose, 150 mM potassium chloride, 1 mM EDTA, 1 X protease inhibitor cocktail. Nuclei and cell debris were removed by centrifugation of the lystate at 2000 g for 4 min at 4 °C. Heavy membranes containing mitochondria were obtained by centrifuging the supernatant at 13,000 g for 10 min at 4 °C and washing the pellet once in lysis buffer. Membrane fractions (0.2 mg/ml protein) were incubated with desired BCL-2 family proteins in 250 μL volume and incubated for 30 min at 37 °C. Mitochondria were then pelleted by centrifugation at 13,000 g for 10 min. The release of SMAC-mCherry was measured as fluorescence intensity using a Tecan M1000 microplate reader and comparing fluorescence intensity between the supernatant and pellet fractions. The percentage release of SMAC-mCherry was calculated as [Fsupernatant /(Fsupernatant +Fpellet)] x 100.
In vitro FRET AssaysRequest a detailed protocol
FRET experiments were carried out as previously described either in the presence or absence of liposomes (Pogmore et al., 2016; Kale et al., 2014). Briefly, using the same assay buffer that was used to make the liposomes, 100 uL reaction volumes were prepared in a 96-well half-area non-binding surface plate (Corning). Fluorescence intensity of the background was recorded for 30 minutes at 37 °C (back). Next, donor labelled protein (e.g., PUMA*A568) was added to each well, and the fluorescence intensity measured for 30 minutes at 37 °C (FD). Finally, acceptor labelled protein, (e.g. BCL-XL*A647) was added at the desired concentrations to each well, and the fluorescence intensity of the donor was measured for 60 minutes at 37 °C (FDA). The FRET efficiency expressed as a percentage was calculated using the following formula:
As a control the fluorescence intensity of the donor (FD) was recorded in the presence of unlabeled acceptor. Therefore, for each concentration of labelled acceptor protein tested (FDA), a separate reaction containing the same concentration of unlabeled acceptor protein was measured (FD).
Cell Lines and CultureRequest a detailed protocol
Baby mouse kidney (BMK) cells (wildtype and dko) were cultured in DMEM supplemented with 10% fetal bovine serum (Gibco) and 5% non-essential amino acids (NEAA, ThermoFisher). The BMK-dko cell line was a kind gift from the originator Eileen White, Rutgers Cancer Institute of New Jersey. NMuMG cells (a generous gift of J. Wrana, Lunenfeld-Tanenbaum Research Institute, Toronto, Canada) were cultured in DMEM, containing 10 µg/ml bovine insulin (Sigma), 10% fetal bovine serum (Gibco), and penicillin/streptomycin (Wisent). NMuMG cells expressing fluorescent landmark proteins were generated as previously described (Schormann et al., 2020). The originating cell lines used in the studies reported here (BMK-wt, BMK-dko, NMuMG) and all stably transfected clones were shown to be free of mycoplasma using a PCR based test (Hopert et al., 1993). All cell lines were maintained in a 5% CO2 atmosphere at 37 °C.
ImmunofluorescenceRequest a detailed protocol
For immunofluorescence experiments to visualize the localization of endogenous PUMA, MCF-7 cells expressing either mCherry-ActA (mitochondrial marker) or mCherry-CB5 (ER marker) were seeded into 384-well plates, at a density of 7000 cells/well. The next day, cells were treated with indicated drugs and concentrations for 24 hours. On the third day, the cells were stained with DNA stain (DRAQ5) plus MitoTracker Green, or DRAQ5 plus Bodipy-thapsigargin for a total of 30 minutes. Subsequently, all cells were fixed using 4% formaldehyde diluted in Dulbecco’s Phosphate Buffered Saline (DPBS). Blocking solution (normal goat serum 5% v/v, Tiriton-X-100 0.3% v/v, diluted in DPBS) was then applied for 1 hour, followed by overnight incubation with primary anti-PUMA antibody. The next day, cells were washed with DPBS, then incubated with secondary anti-rabbit (Alexa Fluor(R) 488) antibody for 2 hours.
Quantitative Fast FLIM-FRET (qF3)Request a detailed protocol
Detecting protein interactions in live cells was carried out as previously described (Osterlund et al., 2022). Briefly, BMK-dko cells stably expressing a mCerulean3-fused anti-apoptotic protein were seeded at 4000 cells/well into a CellCarrier-384 Ultra Microplate (PerkinElmer). 24 hours later, individual wells were transfected with plasmids encoding Venus fused BH3- proteins using TransIT-X2 reagent (Mirus). Non-transfected wells were treated with transfection reagent alone (no DNA added). Media was changed after 3–5 hours. At this time, selected wells were treated with BH3 mimetic or DMSO. Sample plates were incubated 12–18 hours prior to imaging. Immediately before imaging, fluorescence protein standards were added to empty wells. Fluorescein (10 nM in 0.1 M NaOH) and quenched fluorescein (30 μM Fluorescein in 8.3 M NaI and 100 mM Na2HPO4 (pH 10)) were also added to the plate for instrument calibration. Data acquisition was performed on the INO-FHS microscope as previously described (Osterlund et al., 2019).
Data from each replicate was analyzed to generate binned binding curves. All biological replicates (3 or more) were combined and fit to a Hill equation to generate binding curves and calculate dissociation constants. A 1:1 binding between donor (mCerulean3) fused protein and acceptor (Venus) fused protein was assumed for all protein pairs. Regions of interest (ROIs) were automatically identified within images via a watershed algorithm applied to the TCSPC channel. Pixels within each ROI were combined and used to calculate the fluorescence lifetime. Change in angular frequency is then calculated from corresponding phasor analysis as described in Osterlund et al., 2022. The intensity of mCerulean3 and Venus per ROI was measured and then converted to concentration values based on measured standard curves prepared using recombinant fluorescent proteins and imaged on the same plate. We applied a narrow physiologically relevant range of mCerulean3 expression (1–3 µM) for our binding curves, and 0–50 µM range for Venus. In the 10–20 µM VenusFree range, the difference in angular frequency between positive control (Venus-BH3) and negative (Venus-BH3-4E mutant) must be greater than 0.05 to be considered as sufficient dynamic range in the assay to reliably detect changes. For each curve, the Bmax (saturation parameter) was estimated from the median of the points in the far right of each binding curve (30–50 µM VenusFree). The minimum sRatio, a threshold value representing the curvature of the binding data used to differentiate binding from collisions, was set to 1.5 based on previous analyses and theoretical modelling (Osterlund et al., 2022). For each protein pair data were combined from independent replicates (n≥3), and the average Kd reported. In heatmaps, ‘binding’ was represented in blue and ‘no binding’ in red. However, the colors are strictly to enable easy visual comparison of the data. In addition, the numerical value of the apparent Kd values are provided in each cell of the heatmaps.
FLIM–FRET with ER or mitochondrial segmentationRequest a detailed protocol
In all other experiments in this paper, to extract FLIM-FRET binding curves ROIs were selected by applying a watershed segmentation algorithm to the TCSPC image of mCerulean3 (e.g. CBCL-XL). Our standard segmentation approach was run on all data as a positive control for detection of binding throughout the cell (see Figure 4—figure supplement 2 “mC3 total cell”). BCL-XL is found at the ER, mitochondria and in the cytoplasm (Kaufmann et al., 2003; Osterlund et al., 2023).
The purpose of this experiment was to express mCherry-tagged landmarks for ER or mitochondria in the same cells as PUMA and BCL-XL, then use the expression of the mCherry landmark to define the boundaries of ROI segmentation. This modification in ROI selection allows us to examine interactions in the ER verses mitochondria. MCF-7 cells stably expressing both CBCL-XL and mCherry (red channel)- tagged markers for ER (mCherry-Cb5) or mitochondria (mCherry-ActA) (Osterlund et al., 2023) were transiently transfected with constructs to express VPUMA, VPUMA-4E, VPUMAd26 orVPUMAd26-4E. Data were collected using two imaging configurations as we recently described (Osterlund et al., 2023). The two imaging configurations are rapidly exchanged to acquire signal from mCerulean3 and Venus (fluorescent proteins) by time correlated single photon counting (TCSPC) FLIM, and Venus and mCherry by using a 64-channel hyperspectral detector. Altogether, the Venus protein expression was captured in both configurations and may be used to confirm alignment images acquired in configurations 1 and 2. In contrast to our previous publication (Osterlund et al., 2022), we observed low transient expression of Venus. As a result, we modified our analysis pipeline to calculate relative Venus expression as a measure of intensity (arbitrary units) from the TCSPC FLIM image, rather than use the hyperspectral data, which requires background subtraction.
In detail, the intensity of signal from mCherry was determined by summing the hyperspectral counts at wavelengths 592–660 nm. The same parameters were used to segment all images of mCherry expression (Red Channel): background threshold (1000), Laplacian of a Gaussian kernel size (Delbridge et al., 2016) with sigma (0.13), structural element size (Wilfling et al., 2012), minimum ROI size (Chi et al., 2020), erode factor (0). Nevertheless, the resulting ROIs selected by the Watershed Algorithm were distinct (example images given in Figure 4—figure supplement 2A compare ChActA Mitochondria to ChCb5 ER). As expected, ROIs selected based on the mCherry-Cb5 expression appear more elongated and mesh-like compared to the more punctate ROIs selected based on mCherry-ActA (mitochondria). We provide the MATLAB analysis package on DataVerse (https://doi.org/10.5683/SP3/ZKXQW8).
The ROIs selected from mCherry-marker expression, were directly applied to extract for each ROI the average intensity and lifetime of mCerulean3 (donor) and the average intensity of Venus (acceptor). These data were used to generate FLIM-FRET binding curves in Figure 4—figure supplement 2B for ER- versus mitochondria-segmented ROIs as described (Osterlund et al., 2023). The % FRET efficiency was calculated using the phasor approach (Osterlund et al., 2022; Ranjit et al., 2018) and “Acceptor:Donor intensity ratio” represents the ratio of the average intensity per ROI of Venus to mCerulean3 both recorded by TCSPC. During acquisition, a 10 nM fluorescein sample in 0.1 M NaOH was used to standardize the instrument response for the three biological replicates to enable combining the binding curve data.
Image ColocalizationRequest a detailed protocol
Analyses of colocalization by image classification were carried out as previously described (Schormann et al., 2020). Conventional colocalization studies to measure Pearson’s correlation coefficients were carried out in BMK-dko cells. We utilized multiple fluorescent proteins and dyes to validate protein localization. To identify the endoplasmic reticulum, we either used a dye (BODIPY FL thapsigargin, ThermoFisher Scientific) or an overexpressed ER resident protein composed of EGFP fused to the ER specific tail anchor of BIK (CBIK). To identify mitochondria, the dyes MitoTracker Red and MitoTracker Green (ThermoFisher Scientific) were used. The cells were also stained with the nuclear stain, DRAQ5 to permit segmentation and quality control (Oltersdorf et al., 2005). Query proteins fused to the fluorescent proteins mCerulean3 or Venus were expressed by transfection of BMK-dko cells with gene encoding plasmids. Images were acquired using a 63 X water immersion objective, on the Opera Phenix microscope (PerkinElmer) and colocalization analyzed using an analysis pipeline created in Harmony software. In this pipeline, the DRAQ5 fluorescence signal was used to identify images of individual cells and create a mask for the nuclear and cytoplasmic areas. A low-intensity fluorescence threshold was applied to all cells in order to select for those successfully transfected and expressing the query protein (i.e. Venus or mCerulean3 fused BH3 only protein). Background was subtracted for each image and a Pearson’s correlation coefficient between fluorescent protein-fused query protein and fluorescently labelled organelle was calculated for all remaining objects within the cytoplasmic area. The median Pearson’s correlation coefficient for each replicate is plotted in Graphpad Prism software with a line indicating the average between all three replicates.
Cell death assayRequest a detailed protocol
Cell death in response to exogenous expression of vBH3-only proteins in HEK293, BMK, and HCT116 cells, was measured as described (Chi et al., 2020). Briefly, cells were trypsinized and seeded in CellCarrier-Ultra 384-well plates. One day later, cells were transfected with plasmids encoding vBimL or vPuma (or the mutant proteins) using Mirus TransIT-X2 transfecting reagent. The media was aspirated prior to the addition of the transfection mix into each well. Four hr after transfection, the media was exchanged to remove the transfection mix. To measure cell death in response to BH3 mimetic treatments, cells were seeded as described above and the BH3 mimetics were added the next day as described (Chi et al., 2020). After 24 hr of transfection or drug treatment, cells nuclei were stained with Hoechst 33342 and Alexa 647 labelled Annexin V was added to detect externalized phosphatidylserine on the outer leaflet of the plasma membrane in dead or dying cells. Stained cells were imaged by automated confocal microscopy (Opera Phenix, PerkinElmer) and analyzed with Harmony software (V4.9) to obtain %Annexin V positivity for the population of cells in each treatment. Cell death due to vBimL or vPuma expression was quantified as percentage of Venus positive, Annexin V positive cells out of all Venus positive cells. Background Venus intensities in the un-transfected cells were used to determine the threshold for Venus positivity (2 standard deviations above the mean Venus intensity in the un-transfected wells).
Image-based analysis of subcellular localization by machine learningRequest a detailed protocol
NMuMG cells expressing either EGFP-BIK-L61G or EGFP-PUMA4E were seeded (3000 cells per well) in multiple wells (at least 3) of a 384-well microplate (CellCarrier-384 ultra, B128 SRI/160; Perkin Elmer) and allowed to grow for 24 hr before staining with the nuclear dye DRAQ5 (5 nM; Biostatus). Cells were imaged (>15 field of views) on a spinning disk automated confocal microscope (PerkinElmer) with a 40 x water objective (NA = 0.9) in a defined temperature (37 °C) and CO2 (5%) environment. Images were collected using 3-Peltier cooled 12-bit CCD cameras (Type sensiCam, camera resolution 1.3 megapixels; PCO.imaging), unbinned. Segmentation and feature extraction were carried out as described in Schormann et al., 2020. Subcellular localization of cell images was determined by using a Random Forests classifier. Landmarks were expressed as EGFP fusion proteins in NMuMG cells for cell organelles shown in Figure 4: ER, (Cytochrome b5, BIK) and Calr-KDEL (ER targeting sequence of calreticulin fused to the N-terminus of EGFP and the ER retention sequence KDEL at the C-terminus of EGFP); ERGIC,ERGIC53; Golgi, GalT (N-terminal 81 amino acids of human beta 1,4-galactosyltransferase) and Golgin84; Mitochondria, MAO (sequence encoding the TA region of Monoamine oxidase A) and CCO (Cytochrome c oxidase, subunit VIII); MAM, Phosphatidylserine synthase 1; RAB5A and RAB7A vesicular compartments, Ras-related protein Rab5A and Rab7A; secretory pathway, VAMP2 and VAMP5 (Vesicle-associated membrane protein 2 and Vesicle-associated membrane protein 5); cytoplasm, Vesicle-associated membrane protein 1 A without TMD (delta-TMD-VAMP1); nuclear membrane, Lamin A; lysosomes, LAMP1 and peroxisomes, PTS1 (Peroxisome targeting signal 1).
We provide the MATLAB data analysis package on DataVerse (https://doi.org/10.5683/SP3/ZKXQW8).
The tale of tail-anchored proteins: coming from the cytosol and looking for a membraneThe Journal of Cell Biology 161:1013–1019.https://doi.org/10.1083/jcb.200303069
Thirty years of Bcl-2: Translating cell death discoveries into novel cancer therapiesNature Reviews. Cancer 16:99–109.https://doi.org/10.1038/nrc.2015.17
Characteristics of a human cell line transformed by DNA from human adenovirus type 5The Journal of General Virology 36:59–74.https://doi.org/10.1099/0022-1317-36-1-59
Etoposide induces cell death via mitochondrial-dependent actions of p53Cancer Cell International 15:79.https://doi.org/10.1186/s12935-015-0231-z
Bcl-2 family proteins: changing partners in the dance towards deathCell Death and Differentiation 25:65–80.https://doi.org/10.1038/cdd.2017.186
Characterization of the signal that directs Bcl-x (L), but not Bcl-2, to the mitochondrial outer membraneThe Journal of Cell Biology 160:53–64.https://doi.org/10.1083/jcb.200210084
Apoptosis induction in human breast cancer (MCF-7) cells by a novel venom L-amino acid oxidase (rusvinoxidase) is independent of its enzymatic activity and is accompanied by caspase-7 activation and reactive oxygen species productionApoptosis 20:1358–1372.https://doi.org/10.1007/s10495-015-1157-6
Puma, a novel proapoptotic gene, is induced by p53Molecular Cell 7:683–694.https://doi.org/10.1016/s1097-2765(01)00214-3
Rapid imaging of Bcl-2 family interactions in live cells using FLIM-FRETMethods Mol Bio 1877:305–335.https://doi.org/10.1007/978-1-4939-8861-7
Endoplasmic reticulum protein BIK binds to and inhibits mitochondria-localized antiapoptotic proteinsThe Journal of Biological Chemistry 299:102863.https://doi.org/10.1016/j.jbc.2022.102863
Using FRET to measure protein interactions between Bcl-2 family proteins on mitochondrial membranesProgr. Cell Death Methods Protoc. Mol. Biol 1419:197–212.https://doi.org/10.1007/978-1-4939-3581-9
Genetic determinants of p53-induced apoptosis and growth arrestGenes & Development 10:1945–1952.https://doi.org/10.1101/gad.10.15.1945
Gene expression during ER stress–induced apoptosis in neuronsJournal of Cell Biology 162:587–597.https://doi.org/10.1083/jcb.200305149
Targeting BCL2 with venetoclax in relapsed chronic lymphocytic leukemiaThe New England Journal of Medicine 374:311–322.https://doi.org/10.1056/NEJMoa1513257
Coupled folding and binding of the disordered protein PUMA does not require particular residual structureJournal of the American Chemical Society 136:5197–5200.https://doi.org/10.1021/ja4125065
A reference library for assigning protein subcellular localizations by image-based machine learningThe Journal of Cell Biology 219:e201904090.https://doi.org/10.1083/jcb.201904090
Discovery of a potent and selective Bcl-xL inhibitor with in vivo activityACS Medicinal Chemistry Letters 5:1088–1093.https://doi.org/10.1021/ml5001867
Abt-263: a potent and orally bioavailable Bcl-2 family inhibitorCancer Research 68:3421–3428.https://doi.org/10.1158/0008-5472.CAN-07-5836
Bh3-Only proteins are tail-anchored in the outer mitochondrial membrane and can initiate the activation of BaxCell Death and Differentiation 19:1328–1336.https://doi.org/10.1038/cdd.2012.9
Puma, a potent killer with or without p53Oncogene 27 Suppl 1:S71–S83.https://doi.org/10.1038/onc.2009.45
Volker DötschSenior and Reviewing Editor; Goethe University, Germany
In the interests of transparency, eLife publishes the most substantive revision requests and the accompanying author responses.
[Editors' note: this paper was reviewed by Review Commons.]https://doi.org/10.7554/eLife.88329.sa1
Reviewer #1 (Evidence, reproducibility and clarity (Required)):
In this paper, the authors present convincing experimental proof on why the BH3-only protein PUMA resists displacement by BH3-mimetics, while others such as tBID do not. Using a SMACmCherry based MOMP assay on isolated mitochondria, FRET in the presence of liposomes with a phospholipid composition similar to that of mitochondria as well as quantitative fast fluorescence lifetime imaging microscopy (Frster resonance energy transfer - qF3) they show that the C-terminal region of PUMA (CTS), together with its BH3-domain, effectively "doublebolt" locks its interaction with BCL-XL and BCL-2 to resist displacement by the BCL-XL-specific BH3-mimetic A-1155463 or the BCL-2/BCL-XL inhibitor ABT-263 and AZD-4320. Although a similar mechanism has previously been published for BIM, the novel C-terminal binding sequence in PUMA is unrelated to that in the CTS of BIM and functions independent of PUMA binding to membranes. First, in contrast to BIM, PUMA contains multiple prolines and charged residues, and an unusually short span of hydrophobic amino acids, secondly, full length PUMA was more resistant to BH3-mimetic displacement than a PUMA mutant lacking the CTS (PUMAd26) even in solution suggesting that the CTS of PUMA contributes to BH3-mimetic resistance even in the absence of membranes.
The second, quite unexpected finding of this paper is that, in contrast to previous publications, the CTS of PUMA does not target the protein to mitochondria but to the ER. The authors show this by FLIM-FRET imaging and confocal microscopy, and they created mutants to identify the CTS residues (I175 and P180) that mediate binding to ER membranes.
The authors did an excellent job to show the mechanism of displacement resistance of PUMA from BCL-2 survival factors from different angles (in vitro, on isolated mitochondria, liposomes and inside living cells), generating respective BH3 and CTS mutants and also domain swaps with other BH3-only proteins such as tBID. Also, the unexpected finding that PUMA primarily localizes to the ER has been extensively scrutinized and the data presented are convincing.
We appreciate the favourable comments and that the reviewer found the data presented convincing.
I have only three questions which I like the authors to address before this MS can be published.
1) How can PUMA perform its pro-apoptotic action on MOMP from its site on the ER? Does
PUMA eventually localize to MAMs (mitochondrial/ER contact sites)? Is it possible to co-IP PUMA with BCL-XL or BCL-2 from ER membranes or show such an interaction inside cells with PLA?
The reviewer raises an important point. One of the main conclusions from this paper is that the primary localization of exogenously expressed PUMA is at the ER. Our intent was to highlight the inherent specificity of the PUMA CTS sequence. However, we agree that identifying the localization of PUMA-BCL-XL complexes would add significantly to the manuscript. We carefully considered using co-IP or a proximity ligation assay (PLA) in order to investigate the localization of PUMA-BCL-XL complexes. In our experience the use of co-IP is very difficult to interpret due to the well characterized detergent-induced artifacts previously shown for BCL-2 family protein interactions (PMID: 9553144, PMID: 33794146). Moreover, PLAs are a proximity assay with a detection range of ~>20nm, and are difficult to quantify beyond enumerating frequency (ie counting spots). In contrast, the detection of FRET by fluorescence lifetime imaging microscopy (FLIM) is very sensitive to distance with a maximum that is <10nm, and the results can be interpreted quantitatively as apparent dissociation constants (manuscript Figures 2-3). Therefore we elected to use FLIM-FRET to address this question. We examined PUMA-specific interactions with BCL-XL at the ER and mitochondria by differentially segmenting the FLIM-FRET image data based on the signal from a mCherry-fused landmark expressed at the ER (mCherry-Cb5) or mitochondria (mCherry-ActA). This approach has similar spatial resolution to PLA yet retains more rigorous requirement for proximity and the quantitative interpretability of FLIM-FRET.
For these experiments we used a recently described the method of mitochondrial image segmentation using hyperspectral image data collected during FLIM-FRET imaging (Osterlund et al., 2023). In this approach, a watershed segmentation algorithm was used to identify mitochondria areas from mCherry-ActA images collected simultaneously with the FLIM data. The ER was identified in separate samples using the same approach with mCherry-Cb5 image data. Simultaneous collection of the images ensures that the data are not affected by movement within the cells. Example images showing the segmentation results for each organelle have been added to the manuscript as Figure 4 - Figure Supplement 2A.
The results of this FLIM-FRET experiment described in the text lines 581-598, revealed that VPUMA interacts with CBCL-XL within both ER and mitochondria-segmented ROIs (new Figure 4 - Figure Supplement 2B). These results can be explained by the fact that VPUMA is targeted to the ER, and BCL-XL is known to localize to the ER and mitochondria when bound to BH3 proteins in cells (Kale et al., 2018, PMID: 29149100). This result is similar to what we reported for BIK, another ER-localized BH3 protein that exerts its pro-apoptotic function from ER membranes (PMID: 11884414 and PMID: 15809295). Our recent data for ER localized BIK binding to mitochondria-targeted BCL-XL (Osterlund et al., 2023), suggests that, as the referee suggested, binding to occurs via a membrane-spanning interaction at MAMs (ER-mitochondia contact sites) and/or via relocalization of BIK and/or BCL-XL in response to their co-expression (Osterlund et al., 2023). Consistent with these interpretations, when expression of endogenous PUMA was upregulated in response to stress (Figure 4- figure supplement 3A-B), the amount of PUMA increased at both ER and mitochondria (Figure 4- figure supplement 3C). We have presented this data and interpretation on lines 599-621 and discussed the localization results and the similarity to BIK in the manuscript discussion, lines 1029-1035.
2) Since PUMA seems to be "double-bolt" locked to BCL-2 or BCL-XL via its BH3-domain and CTS, how can it act as a pro-apoptotic inducer? Is its main function to act as an inhibitor of BCL-2 and BCL-XL rather than a direct BAX/BAK activator? And if it acts as a BAX/BAK activator, how can it be released from BCL-2/ BCL-XL, for example by another BH3-only protein which is induced by apoptosis stimulation? Or would in this case PUMA remain bound to BCL-2/ BCL-XL in order to activate BAX/BAK (which would be a kind of new activation mechanism)?
We appreciate the reviewers queries and have clarified the text to indicate that our interpretation is that by binding to BCL-XL, PUMA releases active BAX that is sequestered by BCL-XL (as shown in Figure 1A for purified proteins). Double bolt locking increases both affinity and avidity of PUMA for BCL-XL enabling competition to favor PUMA binding and displacement of sequestered BAX. To further address the reviewers point we added two additional experiments now shown in figure supplements to Figure 1. The data shown in new Figure 1 – figure supplement 1A (described on lines 182-191 of the revised manuscript) demonstrates that PUMA kills HCT116 and BMK cells but not HEK293 cells. New Figure 1 – figure supplement 1B shows that inhibition of BCL-XL and MCL-1 using BH3 mimetics is sufficient to kill HCT116 and BMK cells while HEK293 cells are not killed by even high concentrations of these BH3 mimetics. To kill HEK293 cells requires activation of BAX (described on lines 191-201). Together this data indicates that the primary pro-apoptotic function of PUMA is inhibiting BCL-XL and MCL-1 rather than by activating BAX. This data fits very well with PUMA double-bolt locking resulting in very tight binding of PUMA to BCL-XL and likely MCL-1 as the primary mode of PUMA mediated induction of cell death, at least in the three cell lines investigated here. The importance and role of PUMA mediated BAX activation is an interesting area of active investigation that is beyond the purview of the current paper.
3) Is PUMA still bound to the ER when it is transcriptionally induced by genotoxic stress. In this case, the extra amount of PUMA produced is supposed to directly activate BAX/BAK. Does it do this on the ER or on mitochondria?
The referee raises a very interesting point. Interestingly, Zheng et al., 2022 highlighted a P53dependent death response to genotoxic stress, which results in the extension of peripheral, tubular ER and promotes the formation of ER-mitochondria contact sites (PMID: 30030520). Furthermore, PUMA is transcriptionally activated by P53 (PMID: 17360476). Therefore, we hypothesized the induction of PUMA would increase the fraction of PUMA at ER membranes and MAMs. As the latter resemble mitochondria in micrographs of cells we anticipated an increase in apparent mitochondrial localization. To address this question experimentally, we treated MCF-7 cells with genotoxic stress and ER stressors and tracked the expression of endogenous PUMA by immunofluorescence. The results are described in the manuscript (line 603-613, page 28) and shown in Figure 4 figure supplement 3. The immunofluorescence data confirmed that PUMA protein levels increase after genotoxic stress, as expected (Reference 39, 40 in the manuscript) and to a lessor but still significant extent after ER stress (Figure 4 figure supplements 3A and B). In response to stress the amount of PUMA increased at both ER and mitochondria, however, in unstressed cells the endogenous Puma co-localized more to the mitochondria than to the ER (Figure 4- figure supplement 3C). This suggests that similar to BIK localization of PUMA is dynamic. In particular, the abundance and localization of PUMA binding partners such as BCL-XL also affects PUMA localization (the new data are described on pages 2728, Lines 591-621). As described above, the extra PUMA induced by genotoxic stress can indirectly activate BAX by binding BCL-XL and displacing sequestered activated BAX. Our FLIMFRET data suggest PUMA can bind BCL-XL at both the mitochondria and the ER. Moreover, given the expansion of ER-mitochondrial contact sites that occurs during stress we cannot rule out the possibility that ER-localized PUMA can inhibit mitochondria-localized anti-apoptotic proteins (both BCL-XL and MCL-1) at the ER (for BCL-XL) and MAMs for both proteins.
Reviewer #1 (Significance (Required)):
Very significant contribution to the field. Quite novel
Reviewer #2 (Evidence, reproducibility and clarity (Required)):
This study by Pemberton and colleagues investigates interactions of pro-apoptotic PUMA with anti-apoptotic BCL-2 proteins, employing a variety of BH3-mimetics. The authors demonstrate that the PUMA/aa BCL-2 interactions are mediated not only via BH3-domain/groove interactions, but also dependent on a C-terminal sequence of PUMA. This mirrors (with distinct differences) what the authors have previously reported for BIM. They then, reveal that unexpectedly PUMA is often localising to the ER (as opposed to mitochondria), though this localisation is not important for the resistance of PUMA/BCL-2 complexes to BH3-mimetic treatment, authors speculate that ER localised PUMA may have a day job.
In my opinion, the study is important for several reasons, not least it strongly argues that BH3mimetics are not optimal (in themselves) to promote apoptosis dependent on PUMA, and that approaches to disrupt the "double-lock" mechanisms should be sought - this has clear clinical importance, but equally important is it adds a new layer of complexity to how BCL-2 family members "work", how the double-lock mechanism is overcome in physiological apoptosis remains an open question, for instance. The data support the authors' conclusions, I have a few points that could be addressed.
The positive comments from the reviewer are greatly appreciated.
1. The authors data in cells is consistent with a membrane recruitment effect of the PUMA CTS making a contribution to the resistance of PUMA/aa BCL-2 complexes to BH3-mimetics. What I found really intriguing, is that the CTS also influences affinity in the absence of membranes (Figure 1) - could the authors speculate why they think CTS may be affecting PUMA/aaBCL-2 binding in the absence of membranes?
We agree with the reviewer that membrane binding contributes to BH3 mimetic resistant binding of PUMA to BCL-XL consistent with elegant data presented previously (Pécot et al., 2016; PMID: 28009301). However, we show in Figure 5D that mutants of VPUMA-d26 with restored membrane binding (VPUMA-d26-ER1 and VPUMA-d26-ER2) remain sensitive to BH3-mimetic displacement, indicating that membrane binding alone is not sufficient to confer resistance to BH3-mimetics. Furthermore, as the reviewer pointed out BH3 mimetic resistant binding is observed in the absence of membranes (Figure 1).
The data using purified proteins strongly suggests that the CTS of PUMA binds to BCL-XL and is directly involved in the protein-protein interaction. The fact that PUMA with the C-terminal fusion to the fluorescent protein Venus (PUMAV) still localizes to membranes in live cells (Figure 4 D,E) suggests that the C-terminus of PUMA does not span the membrane bilayer. Instead, we hypothesize that the C-terminus of PUMA binds peripherally to the membrane making it available to physically contribute to a protein interaction with anti-apoptotic proteins. This interpretation is consistent with the low hydrophobicity and high proline content (6 of 28 residues) of the amino acid sequence of the PUMA CTS as shown in Figure 6 and compared to the transmembrane tail anchor sequences of other proteins, including the BH3-protein BIK, in Figure 5 supplement 1. Binding of Bcl-XL by both the BH3 region and CTS of PUMA would increase both the affinity and avidity of the interaction. The presentation of this data has been revised to add clarity on pages Page 8, lines 215-223 and in the discussion (Lines 988-997 and 1044-1050).
2. A minor point for clarification, are the mitochondria used in Fig 1A from BAX/BAK DKO cells?
- I had presumed so given exogenous BAX was added, but didn't note this in the text.
We indeed use mitochondria from BAX/BAK DKO cells and exogenous recombinant BAX in Figure 1A. This has now been added to the text on lines 166-180.
Reviewer #2 (Significance (Required)):
Detailed in report above.
Reviewer #3 (Evidence, reproducibility and clarity (Required)):
In this paper, Pemberton et al show that PUMA resists BH3-mimetic mediated displacement from BCL-XL via a novel binding site within its C-terminus of PUMA termed CTS (the last 26aa). Interestingly, the CTS of PUMA directs the protein to the ER membrane and residues I175 and P180 within the CTS are required for both ER localization and BH3-mimetic resistance.
1. BH3-mimetics kill cells by displacing sequestered pro-apoptotic proteins to initiate apoptosis.
However, PUMA resists BH3-mimetic mediated displacement, and PUMA-d26 and PUMA
I175A/P180A (CTS) do not. Thus, are these mutants sensitive to BH3-mimetics cell killing? In other words, do BH3-mimetics kill PUMA-/- cells that express either PUMA-d26 or PUMA I175A/P180A but not PUMA-/- cells that express wild type PUMA?
The reviewer raises a very interesting question that unfortunately we have been unable to address unambiguously. To answer this question requires separating the effects of PUMA on anti-apoptosis proteins and on activation of BAX and BAK as exogenous expression of either PUMA-d26 or PUMA I175A/P180A is sufficient to kill PUMA-/- cells without the addition of a BH3 mimetic. To date we have been unable to identify mutants that inhibit anti-apoptotic proteins but that do not activate BAX and BAK as both PUMA-d26 and PUMA I175A/P180A have impaired BAX-activation function. This is additionally complicated by PUMA mediated inhibition of MCL-1, BCL-2 and BCL-W. Further, it isn’t possible to separate the function(s) using BAX/BAK knock-out cells because then PUMA induced cell death is completely abrogated. Understanding the direct activation of BAX by PUMA is an area of current investigation that is out of the scope of this paper as here we are focused on the interaction(s) of PUMA with anti-apoptotic proteins.
2. The authors elegantly demonstrate using microscopic analysis that over expressed PUMA mostly localizes to the ER membrane. Since this is a major conclusion in the paper which is different than previously reported, the authors should confirm these findings using sub-cellular fractions followed by Western blot analysis. They should demonstrate that endogenous and over-expressed PUMA are mainly localized to the ER membrane and that the PUMA-d26 and PUMA I175A/P180A are mainly localized to the cytoplasm.
We appreciate that the reviewer found the microscopic analysis convincing. We also tested the idea of sub-cellular fractionation proposed by the reviewer. However, we have found it to be very difficult to separate mitochondria and MAMs. To address the question raised we instead performed new co-localization experiments, in addition to those reported for PUMAd26 and the point mutants in Figure 6 (images in Figure 6 - figure supplement 3). The new experiments are for endogenous PUMA at steady state and with increased expressed in response to stress. These immunofluorescence experiments are reported in Figure 4 -figure supplements 3. We also added FLIM-FRET experiments in which ROIs were derived from areas of the cell enriched in either ER or mitochondria (Figure 4 - figure supplement 2). The results of these experiments indicate that PUMA localization is dynamic and are described in detail above in response to reviewer 1 question 3 and in the manuscript from line 579 to 621 and discussed on lines 1029-1036.
Reviewer #3 (Significance (Required)):
The advance in this paper is significant and the paper should be published once the specific comments are adequately addressed.
Article and author information
Canadian Institutes of Health Research (FDN143312)
- David W Andrews
Canada Research Chairs (Tier 1)
- David W Andrews
Canada Foundation for Innovation
- David W Andrews
Ontario Ministry of Research and Innovation
- David W Andrews
- David W Andrews
The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.
The authors wish to thank Dr. Justin Kale, Dr. Qian Liu, and Dr. Philipp Mergenthaler for their thoughtful critique and support of this project. This work was supported by Canadian Institutes of Health Research (CIHR) grant FDN143312 (DWA), Canada Foundation for Innovation (DWA), Ontario Ministry of Research and Innovation (DWA), CQDM (DWA), and Canada Research Chairs Program (DWA).
Senior and Reviewing Editor
- Volker Dötsch, Goethe University, Germany
- Preprint posted: April 2, 2023 (view preprint)
- Received: April 2, 2023
- Accepted: April 3, 2023
- Accepted Manuscript published: April 20, 2023 (version 1)
- Version of Record published: May 15, 2023 (version 2)
© 2023, Pemberton et al.
This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.
- Page views
Article citation count generated by polling the highest count across the following sources: Crossref, PubMed Central, Scopus.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
- Cancer Biology
- Cell Biology
Tumor initiation, progression and resistance to chemotherapy rely on cancer cells bypassing programmed cell death by apoptosis. We report that unlike other pro-apoptotic proteins, Bim contains two distinct binding sites for the anti-apoptotic proteins Bcl-XL and Bcl-2. These include the BH3 sequence shared with other pro-apoptotic proteins and an unexpected sequence located near the Bim carboxyl-terminus (residues 181–192). Using automated Fluorescence Lifetime Imaging Microscopy - Fluorescence Resonance Energy Transfer (FLIM-FRET) we show that the two binding interfaces enable Bim to double-bolt lock Bcl-XL and Bcl-2 in complexes resistant to displacement by BH3-mimetic drugs currently in use or being evaluated for cancer therapy. Quantifying in live cells the contributions of individual amino acids revealed that residue L185 previously thought involved in binding Bim to membranes, instead contributes to binding to anti-apoptotic proteins. This double-bolt lock mechanism has profound implications for the utility of BH3-mimetics as drugs.
- Biochemistry and Chemical Biology
- Cell Biology
Autophagy is an essential catabolic pathway which sequesters and engulfs cytosolic substrates via autophagosomes, unique double-membraned structures. ATG8 proteins are ubiquitin-like proteins recruited to autophagosome membranes by lipidation at the C-terminus. ATG8s recruit substrates, such as p62, and play an important role in mediating autophagosome membrane expansion. However, the precise function of lipidated ATG8 in expansion remains obscure. Using a real-time in vitro lipidation assay, we revealed that the N-termini of lipidated human ATG8s (LC3B and GABARAP) are highly dynamic and interact with the membrane. Moreover, atomistic MD simulation and FRET assays indicate that N-termini of LC3B and GABARAP associate in cis on the membrane. By using non-tagged GABARAPs, we show that GABARAP N-terminus and its cis-membrane insertion are crucial to regulate the size of autophagosomes in cells irrespectively of p62 degradation. Our study provides fundamental molecular insights into autophagosome membrane expansion, revealing the critical and unique function of lipidated ATG8.