MCL-1 promiscuity and the structural resilience of its binding partners

Protein–protein interactions are the fundamental driving force for a majority of cellular processes.^1,2 Understanding the molecular mechanisms behind this protein–protein interplay is of highest scientific interest.³ For numerous protein complexes, it is not clear how they form, how small conformation fluctuations contribute to the complex function and stability, and whether there are intertwined intermediate states of altered conformation.^4,5 Illuminating the nuances of these dynamic processes is particularly essential for complexes formed by intrinsically disordered proteins.⁶ These proteins do not assume an ordered structure in their isolated form but only when they are bound to their complex partner. For intrinsically disordered proteins, the process of complex formation can be explained with models,^7,8 such as the induced fit model,⁹ the conformational selection model,⁴ or a hybrid version of both theories.¹⁰ Intrinsic disorder plays a significant role in promiscuous protein networks, as it enables the complex formation with numerous binding partners.^11,12 Being on “the edge of chaos”¹³ ensures structural and functional flexibility and provides an ideal basis for diverse protein–protein interactions, for instance, in a network of activator, inhibitor, and effector proteins.

The BCL-2 protein family is a paramount example for an intricate protein network, which is driven by promiscuous interactions of several intrinsically disordered protein domains. In this protein family, categorized in subfamilies based on the type and the number of their BCL-2 homology (BH) domains, the disordered binding domains of so-called BH3-only proteins—they solely have a BH3 domain—form complexes with numerous other BCL-2 proteins, thereby controlling apoptosis in a finely balanced manner.^14–17 The BH3-only proteins, abundant when cells suffer cytotoxic stress,^18–20 either directly activate the pro-apoptotic effector proteins (BAK and BAX) or inhibit the anti-apoptotic effector-inhibitors, such as the Myeloid Cell Leukemia 1 (MCL-1) protein, one of the key players in apoptosis regulation¹⁵ (Fig. 1). MCL-1 is overexpressed in various tumor variants, which makes it of high interest in therapeutic application.^18,21–25 In the past, MCL-1 was demonstrated to exhibit allosteric signaling between the canonical binding pocket and remote parts of the protein.²⁶ The time scales of this allosteric communication are putatively in the nanosecond time regime.²⁷ At its canonical binding site, MCL-1 promiscuously binds the α-helical binding domain of the BH3-only proteins PUMA, BIM, and NOXA,^28,29 most likely by induced fit.^30–34 PUMA and BIM inhibit MCL-1 by occupying its binding pocket with affinities in the sub-nanomolar range, but additionally bind other anti-apoptotic factors and the pro-apoptotic effector proteins BAK and BAX.^17,28,35 In contrast, NOXA binds MCL-1 specifically and with a weaker affinity, however, affects the degradation of the whole complex.³⁶ 

In their intriguing review,¹⁵ Kale et al. describe this interplay of different pro- and anti-apoptotic factors very figuratively as a “dance” of various partners—an interesting and vivid depiction of protein promiscuity. In this “apoptotic dance,” the protein affinity and stability determine the complex formation and whether cell death will be initiated or not.

To date, little is known about the protein dynamics behind the MCL-1/BH3-only complex stability and formation, let alone intermediate states of BH3-only folding and unfolding at MCL-1’s binding groove.³⁷ Investigating subtle, dynamical rearrangements inside proteins like the MCL-1/BH3-only complex is connected to a fundamental challenge: the ability to resolve structural flexibility and small conformational fluctuations in a reasonable time frame.^38,39 In the past, the dynamics of protein ensembles—intrinsically disordered or folded—have been experimentally studied via single-molecule FRET spectroscopy,^40,41 NMR spectroscopy,^42,43 as well as transient infrared (IR) spectroscopy.^44,45 IR spectroscopy allows the differentiation of very small conformational differences⁴⁶ and, in its transient form, the sensitive detection of non-equilibrium processes,⁴⁷ making it a highly suitable method to study MCL-1/BH3-only complexes.

To selectively trigger a dynamical process inside a protein for transient IR spectroscopy, a fast and precisely induced perturbation of the proteins is required, ideally initiated by short light pulses. In this regard, a plethora of photoreceptor proteins, i.e., light-sensitive or fluorescent, have been investigated in the past.^48–54 Beyond proteins that show natural photo-activity, linking azobenzene photoswitches covalently to selected protein domains—most prominently α-helical structures—defines a potent strategy to introduce photo-sensitivity in molecules, which are otherwise “blind” to light. The light-induced isomerization of the cross-linked azobenzene moiety leads to a fast perturbation of the secondary structure,⁵⁵ and in turn to a slower protein response, both of which can be detected via transient IR spectroscopy.^56–58 With this technique, the allosteric signal propagation in PDZ domains and the unbinding in the RNase S complex were investigated.^44,45,59 Similarly, we recently revealed the signal propagation inside the MCL-1/BIM complex, indicating that the structural adaptations of MCL-1 upon photo-perturbation may be classified as an allosteric communication.³⁴ 

In this study, we apply transient IR spectroscopy to explore the protein dynamics of the intrinsically disordered binding domains of PUMA and NOXA in complex with MCL-1. An azobenzene photoswitch has been covalently linked to the short PUMA and NOXA peptides. The induced isomerization of the photoswitch leads to a subsequent destabilization of the secondary structure of the linked peptides, yet remaining bound to MCL-1 (Fig. 2). Together with data from the previously analyzed MCL-1/BIM complex,³⁴ we classify the kinetic response of these three MCL-1/BH3-only complexes and discuss how the observed protein dynamics integrate into the promiscuous nature of the MCL-1/BH3-only system.

We designed photoswitchable variants of the binding domains of PUMA and NOXA in complex with MCL-1 to compare them to the previously generated photoswitchable MCL-1/BIM variant.³⁴ The domains, from here on pars-pro-toto referred to as PUMA, NOXA, and BIM, are 25 to 29 amino acid long intrinsically disordered peptides that become α-helical when complexing with MCL-1.¹⁶ They interact with a classical heptad pattern with hydrophobic side chains at the a/d positions of the helix (Fig. 2, sequences). On the solvent-exposed side of PUMA and NOXA, we introduced two cysteine residues, which were used to covalently bind the photoswitch 3,3′-bis(sulfonato)-4,4′-bis-(chloroacetamido) azobenzene (BSBCA)⁶¹ to the peptide. In a previous study with BIM, we identified positions 16 and 23 (two consecutive c positions of the heptad pattern) as anchoring points for the azobenzene moiety. For the newly generated complexes, we introduced cysteines at the corresponding positions (Fig. 2, in cyan).

Circular Dichroism (CD) spectroscopy of the isolated BH3-only peptides reveal a random coil structure in both the cis-state and the trans-state of the photoswitch [exemplified for PUMA in Fig. 3(a), red and blue]. The CD spectrum of MCL-1 alone displays the classical response for an α-helical structure [Fig. 3(a), black]. When the BH3-only peptides BIM, PUMA, and NOXA are added in equivalent amounts in the dark, in which case the photoswitch is in the trans-state [Fig. 3(a), purple], the mean residue ellipticity representing the α-helical content is only slightly divergent from that for MCL-1 alone [Fig. 3(b)]. This proves that the photoswitchable BH3-only peptides assumes an α-helical structure in the presence of their natural binding partner MCL-1. By illuminating the MCL-1/BH3-only complexes with 375 nm laser light, we uniformly switched the photoswitchable BH3-only peptides to the cis-state and could detect a small increase in α-helical content for NOXA, PUMA, and BIM [see Fig. 3(b)], as anticipated from the spacing of seven amino acids between the two anchoring points of the photoswitch. Previous results on the photoswitchable MCL-1/BIM complex showed that the dissociation constants in the cis- and trans-state (⁠$<$4 μM) fall far below the concentration that we applied in the spectroscopic experiments (⁠$>$400 μM). For MCL-1/BH3-only complexes, the slight destabilization of the α-helix in the trans-state is not sufficient to result in an unbinding event for the concentration range needed in IR spectroscopy.³⁴ 

For the transient IR experiments, the sample was first prepared in the cis-state with the help of a cw-LED at 375 nm. Upon subsequent switching of the azobenzene moiety from the cis-state to the trans-state by the irradiation with an ultrashort UV/VIS laser pulse at 420 nm, the abrupt isomerization process perturbs the secondary structure of the peptide, and in consequence also that of its binding partner MCL-1. To monitor these structural changes, we set our focus on the C=O stretch vibrations of the protein backbone, i.e., the amide I band in the region around 1650 cm⁻¹. The amide I band serves as an indicator for rearrangements and alterations in the protein structure.⁶² We performed these experiments both in a steady-state manner with the help of a Bruker Tensor 27 FTIR spectrometer [Figs. 4(a) and 4(c)], as well as transiently with the pump-probe delay time ranging from picoseconds to 42 µs [Figs. 4(b) and 4(d)]. Additionally, we recorded equivalent steady-state difference spectra of the isolated peptides (Fig. S1). The late time (42 µs) transient spectra are in essence the same as the steady-state difference spectra [Figs. 4(a) and 4(c) red vs black lines], indicating that most of the structural changes have already found their end at this time point.

In order to extract the dynamical processes contained in the transient spectra, we performed global multiexponential fitting, assuming interconverting discrete states with time-invariant spectra,^63–65 

Here, we treated the amplitudes a(ω_i, τ_k), as well as a common set of time constants τ_k as the free fitting parameters, with the number n of exponential terms being restrained to a minimum.^49,66 To find this minimum, we compared the residuals for an increasing number of exponential terms. The absolute values of the residuals decrease with increasing number of terms for n ≤ 3 (PUMA) and for n ≤ 2 (NOXA). For n above these thresholds, there were no evident differences anymore. In consequence, the minimal number of terms for PUMA was 3, and for NOXA 2. We fitted the experimental data of PUMA with four states, S₁, S₂, S₃, and a terminal state S_t and three time constants connecting them (τ₁₂ = 1.6 ns, τ₂₃ = 18 ns, and τ_t = 1.6 µs). For NOXA, three states S₁, S₂, and S_t and two time constants were sufficient (τ₁₂ = 85 ns and τ_t = 1.4 µs). The corresponding time constants in BIM are τ₁₂ = 9.7 ns, τ₂₃ = 150 ns, and τ_t = 3.6 µs (see Ref. 34).

The observed timescales are summarized in Fig. 5. While the last timescale (τ_t) is more or less the same in all three samples, the preceding two processes (τ₁₂ and τ₂₃) vary by almost a factor 100, with PUMA being the fastest and NOXA the slowest. In the case of NOXA, τ₂₃ is not resolved, presumably since it coincides with the terminal process τ_t leading to a spectral signature in state S_t that is very similar to those observed for PUMA and BIM.

Assuming a sequential, unidirectional process (see Fig. 6, top), we also calculated evolution associated difference spectra (EADS) according to

where C_k(t_j) is the concentration profile of component k as a function of time t_j, and A_k(ω_i) its spectrum at probe frequency ω_i. The spectra A_k(ω_i) are a linear combination of the fitting amplitudes a(ω_i, τ_k).⁶⁶ 

For all investigated peptides, the first EADS, revealing the response immediately after the pump pulse is over, shows in essence a bleach of the amide I band (Fig. 6, red). The subsequent state (Fig. 6, yellow), populated with time constant τ₁₂, reveals an additional positive band on the higher-frequency side, and hence a blue-shift of the amide I band. This blue shift has been similarly detected for MCL-1/BIM³⁴ and other α-helical peptides^46,67 where it displayed the same asymmetric appearance and was interpreted as a partial α-helical unfolding. The asymmetry in signal strength between the bleach at 1640 cm⁻¹ and the emerging feature at 1660 cm⁻¹ follows the same trend as already seen for the mentioned systems^34,46,67 and only allows a qualitative rather than a quantitative classification. In Ref. 34, we had also investigated the response a localized vibrational mode directly associated with the photoswitch, supporting this interpretation. The blue-shift can be identified in the raw data of Fig. 4 as well, where it is marked with triangles. While the blue-shifted band is observed for all three samples with very similar spectroscopic characteristics, the timescale with which it appears, τ₁₂, varies by almost a factor 100.

Based on isotope labeling experiments,³⁴ the subsequent spectral changes occurring with τ₂₃ and τ_t have been attributed to mostly the protein MCL-1 responding to the structural perturbation of its binding partner together with further rearrangements of BIM, located inside MCL-1’s binding pocket. Based on an analogous spectral signatures, we assume the same for PUMA and NOXA.

MCL-1 promiscuously binds to numerous intrinsically disordered inhibitors, BH3-only peptides, at its binding groove.^28,29 MCL-1’s centrality in the cancer-related apoptotic networks and its promiscuous nature makes it of high interest for biochemical and pharmaceutical research^21,22 In this study, we characterized the promiscuity of MCL-1 on an atomistic level. To that end, we explored the protein response of MCL-1/PUMA and MCL-1/NOXA upon ultrafast photo-perturbation. Together with the recently published data on a photoswitchable MCL-1/BIM variant, we could draw on dynamical information for three of MCL-1’s natural inhibitors in a picosecond to microsecond time window.

After the abrupt photo-isomerization of the azobenzene moiety, the three complexes reveal very similar spectroscopic responses, both regarding the late time (steady-state) response [Figs. 4(a) and 4(b)], as well as that of the transient intermediates (Fig. 6). There is, however, one crucial difference between the three investigated MCL-1/BH3-only complexes: Partial α-helical unfolding, occurring with τ₁₂, happens at strongly divergent time points for every BH3-only peptide; see blue lines in Fig. 5. The fastest response was detected for PUMA (τ₁₂ = 1.6 ns), succeeded by BIM (τ₁₂ = 9.7 ns), and finally by NOXA (τ₁₂ = 85 ns). Interestingly, the time point of partial α-helical unfolding does not correlate with the binding affinity of the peptide,¹⁷ but significantly with computed scores for its polarity⁶⁸ and hydropathicity,⁶⁹ as well as with the decline of α-helical content due to photoswitching (Fig. 7). The latter was determined by measuring triplicates of photoswitchable MCL-1/BH3-only complexes with CD spectroscopy. The stronger the decrease in α-helicity is upon cis-to-trans isomerization, the faster is its response. The same is true for increasing hydrophilicity as well as polarity.

Disrupting PUMA, which reveals the fastest response, with the trans-state azobenzene moiety may result in unfavorable contacts between hydrophilic side chains of PUMA and the hydrophobic interface of the binding groove of MCL-1. This strong tension could be dissolved by a rapid partial α-helical unfolding. Interestingly, a computational study solely on PUMA highlighted the increased α-helical propensity at the C-terminal relative to the rest of the peptide,⁷⁰ the same region that we destabilize in our experiments. Apparently, the perturbation of this region, which is innately more likely to form ordered structures, leads to conformational tension and thus to the fast protein response. In contrast, the slow opponent NOXA—inherently more hydrophobic and less polar—is not confronted with comparable tensions; thus, it is not forced to rearrange quickly and, thus, unfolds later. The discrepancies between the peptides may also arise from breaking peptide-exclusive electrostatic interactions at the putative contact interface (Fig. 2, yellow residues) upon photoswitching. Electrostatic interactions at the contact interface are known to support the complex formation of MCL-1 and BH3-only peptides, namely, those formed by MCL-1 Lys234 and Glu9 of NOXA or PUMA (numbering according to the alignment in Fig. 2),^28,71 MCL-1 Asp256/peptide Arg15,^71,72 and MCL-1 Arg263/peptide Asp19.^71,72

Beyond this atomistic view on the investigated MCL-1/BH3-only complex, we tried to infer the consequences that our observations have for the biological understanding of the system.

First, we note that the observed time constants for the α-helix destabilization match the time scales of putative allosteric communication in MCL-1.²⁷ This implies that the destabilization of BH3-only domains in the binding pocket of MCL-1 could temporally interfere with synchronous signaling pathways inside MCL-1, making our observation relevant to understand the simultaneity of processes in MCL-1’s “apoptotic dance”¹⁵ with its pro-apoptotic inhibitors.

Second, our observations pose the question of how the peptide-specific differences correlate with the promiscuous nature of the MCL-1/BH3-only complexes on a cellular level. Pro-apoptotic BH3-only proteins are “damage sensors” and abundantly expressed when cells suffer cytotoxic stress.^18–20 Although their binding domains share the same α-helical structure of similar length and identical hydrophobic heptad pattern, as well as the same binding site at MCL-1, BH3-only proteins differ substantially in their relationship inside the BCL-2 family and their interaction pattern (Fig. 1). BIM and PUMA bind the full spectrum of the anti-apoptotic BCL-2 family, and also activate the effector proteins BAK and BAX, whereas NOXA selectively inhibits MCL-1 and BCL-2, another member of the same protein family.¹⁵ BIM and PUMA stabilize MCL-1, while NOXA promotes its degradation.^28,36 At this point, we cannot decide whether these functional differences result from structural alterations at the binding groove or from a more complex involvement of other domains that were not investigated in the present study.

Studies with truncated or mutated versions of NOXA demonstrated that the C-terminal region of the binding domain regulates the stability of the MCL-1/NOXA complex and therefore is important to control MCL-1/NOXA degradation^28,36,73 (Fig. 1, left). In in vivo experiments with the full size NOXA protein and MCL-1, Willis et al. proved that mutating only three residues of the binding domain (alignment position 14, 17, and 21 in Fig. 2) was enough to drastically reduce the biological activity of the full size NOXA protein.³⁶ The perturbation in our experiments aims precisely at this region of NOXA’s binding domain, where apparently subtle differences can entail the dissociation of the protein complex. To maintain its biological function, this region of NOXA has to bind to MCL-1, which requires a certain “structural resilience” for this region, even when the whole complex is destabilized in the proteolytic process. We delimit this “structural resilience” from the already established terms “stability”⁷⁴ and “structural plasticity”.⁷⁵ The former is classically used in a thermodynamic context to describe how partners in a protein complex form and maintain folded conformations.^74,76–78 The latter is frequently used to characterize the ability of promiscuous proteins, e.g., BCL-2-type,⁷⁹ chaperones,^80,81 trypsin,⁸² or protein kinases,⁸³ to flexibly bind various, different binding partners at the same interface. With “structural resilience,” we have in mind a kinetic stabilization, to contrast to a thermodynamics one.

Our results show that even after photo-induced destabilization, the photo-perturbed C-terminus of the NOXA peptide remains folded more than 10 times longer than the equivalent region in other peptides (Fig. 5). While that is definitely speculative, the high structural resilience of NOXA’s C-terminus inside the binding pocket of MCL-1 could help to remain in place, even when the whole complex is confronted with major rearrangements leading to the potent proteolysis of MCL-1. Future experiments could test this hypothesis by connecting a mutational analysis of NOXA and in vivo screening of MCL-1 proteolysis with monitoring the peptide’s structural resilience by IR spectroscopy.

Different to NOXA, PUMA and BIM increase the stability of MCL-1^28,35 and block the binding groove of MCL-1 (Fig. 1, right). This in turn limits MCL-1’s ability to bind the effector proteins BAK and BAX. The competition between BIM, PUMA, and the effector proteins for MCL-1 manifests in the extremely high affinities of BIM (K_D = 25 pM) and PUMA (K_D = 180 pM) in comparison to the already high affinities for the effector proteins BAK (K_D = 1.4 nM) and BAX (K_D = 22 nM).¹⁷ Transient IR spectroscopy demonstrated that the structural resilience of PUMA and BIM is smaller than that of NOXA, presumably because their natural role is different, binding the partner in a highly stable complex with little structural flexibility. Any disturbance seems to result in a fast adaptation, i.e., the partial unfolding that we observed in our experiments.

In summary, our study reveals insights into the promiscuity of the anti-apoptotic, allosteric MCL-1 for the intrinsically disordered binding domains of BH3-only proteins. By using transient IR spectroscopy in combination with photo-switchable protein complexes, we quantified the dynamic response of BIM, PUMA, and NOXA in the binding pocket of MCL-1 upon photo-perturbation. All peptides show partial α-helical unfolding, however, on very different time-scales. The correlations in Fig. 7 are indicating that the speed of protein response in our system is coupled to the extent of structural rearrangements [Fig. 7(d)] and possibly linked to favorable or unfavorable interactions, caused by the perturbation. Computational studies could test this hypothesis.

NOXA is structurally more resilient than BIM and PUMA. This finding reveals a new viewpoint on the nature of BH3-only peptides, which could help to better understand promiscuous protein–protein interactions, in general, as well as the design of novel molecules and peptides to therapeutically manipulate the oncologically relevant MCL-1/BH3-only complex.

The BH3 domains of PUMA (EEQWAREIGAQLRCMADDLNCQYER) and NOXA (RAELPPEFAAQLRCIGDKVYCTWSAP), both containing cysteine mutations with a spacing of seven amino acids, were synthesized using solid state peptide synthesis on a Liberty 1 peptide synthesizer (CEM Corporation, Matthews, NC, USA). These peptides were purified analogously to the BIM variant (GGSGRPEIWIAQELRCIGDEFNCYYARRV), which was investigated in a preceding study.³⁴ The water-soluble photoswitch 3,3′-bis(sulfonato)-4,4′-bis(chloroacetamido)azobenzene (BSBCA) was subsequently covalently linked to the cysteine residues, as described earlier.^61,84 The successful linkage, purity, as well as the integrity of the peptide, was controlled via mass spectrometry.

Human MCL-1 (hMCL-1ΔN-ΔC, 171-327, C286S⁸⁵) was expressed in Escherichia coli BL21. The cells were grown until they reached OD₆₀₀ = 0.6 and induced with 0.75 mM isopropyl β-D-1-thiogalactopyranoside. After the induction, the cells were incubated for 20 h at 30 °C until they were harvested end lysed using sonication. The proteins were purified under native conditions via Ni-affinity chromatography and a His₆-Tag at the N-terminus of the protein. The N-terminal His₆-Tag was removed by 3C protease cleavage. All experiments were performed in 50 mM Tris (pH 8) and 125 mM NaCl. The integrity, as well as the purity of the protein sample, was controlled via mass spectrometry. For IR spectroscopy, H₂O containing sample buffer was exchanged against D₂O based buffer via spin column centrifugation. To circumvent the contamination with atmospheric H₂O, the sample was kept in a water-vapor free nitrogen environment.

For pump–probe measurements, we used two electronically synchronized 2.5 kHz Ti:sapphire oscillator/regenerative amplifier femtosecond laser systems (Spectra Physics), allowing a delay of maximally 45 µs.⁸⁶ For the pump pulses, one laser system was tuned to 840 nm and then brought to 420 nm pulses via second harmonic generation in a β-BaBO₄ crystal, later needed for cis-to-trans-isomerization of the photoswitch. After light amplification, the compressor has been bypassed, resulting in ≈60 ps stretched pulses, in order to reduce sample deposition on the sample cell windows. At the sample cell, the power was 3 μJ per pulse, focused to a ≈140 μm beam diameter. The second laser system was used to generate mid-IR probe pulses in an optical parametric amplifier (100 fs, spot size 110 μm, center wavenumber 1625 cm⁻¹).⁸⁷ To prepare the investigated samples for the spectroscopic experiments, the crosslinked peptides and MCL-1 were mixed in a 1:1 ratio in D₂O with a total complex concentration of 1 mM. The protein samples were constantly circulated in a closed-cycle flow cell system, comprising a CaF₂ measurement cell (50 μm optical path length) and a reservoir. The measurement took place for 2 to 3 h at room temperature. The sample was visually controlled for degradation on the CaF₂ windows. To exclude precipitated protein from the measurement, the sample was centrifuged before every experiment and the supernatant was used. To prepare the investigated sample for the measurement in the cis-state, we irradiated it with a 375 nm continuous wave diode laser (90 mW, CrystaLaser) before entering the measurement cell. This guaranteed that the sample was $>$85% in cis-state.⁸⁴ In the past, it was shown that azobenzene photoswitches had photoisomerization quantum yields that were substantially higher for cis-to-trans switching than for trans-to-cis. Furthermore, this was demonstrated to be independent of the bound peptide.⁸⁸ Thus, trans-switched peptides are in a minority and additionally exhibit a lower isomerization quantum yield than their cis-counterparts, resulting in a negligibly small effect on the effective signal upon cis-to-trans photoswitching.

To better understand the differences of the various BH3-only peptides, we calculated parameters for the polarity according to Grantham,⁶⁸ and the hydrophobicity/hydropathicity according to Kyte and Doolittle⁶⁹ from the sequences of PUMA, BIM, and NOXA using the Expasy/ProtScale server (http://web.expasy.org/protscale/). The calculations were executed with a window size of seven residues (interval length for the computation) and the same weight for every residue in that interval (relative weight = 100%).⁸⁹ The Kyte–Doolittle method was used here explicitly with a window size of seven that allows the scan for hydrophobic/hydrophilic regions.⁸⁹ On the resulting hydropathicity scale, positive values indicate hydrophobic and negative values hydrophilic regions.

See the supplementary material for steady-state IR difference spectra of isolated PUMA and NOXA.

We thank Markus B. Glutz for the synthesis of the peptides and Functional Genomics Center Zurich for their work on the mass spectrometry analysis. We thank Roland Zehnder for the technical support and Kerstin T. Oppelt, Gerhard Stock, and Steffen Wolf for fruitful discussions. This work was supported by the Swiss National Science Foundation (SNF) through Grant No. 200020B_188694/1.

The authors have no conflicts to disclose.

Philipp J. Heckmeier: Conceptualization (lead); Investigation (lead); Software (supporting); Visualization (lead); Writing – original draft (equal). Jeannette Ruf: Investigation (supporting); Software (supporting). Brankica G. Janković: Conceptualization (supporting). Peter Hamm: Funding acquisition (lead); Resources (lead); Software (lead); Supervision (lead); Writing – original draft (equal).

The data that support the findings of this study are openly available in Zenodo at https://doi.org/10.5281/zenodo.7561735, Ref. 90.

BH

BCL-2 homology

BSBCA

3,3′-bis(sulfonato)-4,4′-bis-(chloroacetamido) azobenzene

CD

Circular dichroism

EADS

evolution associated difference spectra

IR

Infrared

MCL-1

Myeloid cell leukemia 1.