A billion years of evolution manifest in nanosecond protein dynamics

Significance The influence of evolution on ultrafast processes inside proteins is an uncharted territory. Here, we use time-resolved infrared spectroscopy to track nanosecond protein dynamics, which form a vital link between protein structure and function. We compare fast dynamics within a protein family, whose role in apoptosis is evolutionarily conserved over nearly a billion years. Our findings reveal a remarkable degree of conservation, subtly tuned by whether a process is intricately linked to protein function or not. Introducing species-specific kinetic footprints, our research enables a comprehensive comparative analysis of species based on their distinct ultrafast protein dynamics. The approach complements established methods of molecular paleontology.

Proteins exist as dynamic ensembles, rather than being rigid and static entities.They constantly undergo rearrangements, folding-, and unfolding processes on a nanosecond time scale 2,6,7,10 .Understanding this dynamic nature is essential to comprehending their function.As protein dynamics serve as the crucial link between structure and function 1 , their experimental investigation has predominantly focused on individual protein examples, providing insights into specific [11][12][13] , often intrinsically disordered cases [14][15][16] .Surprisingly, protein dynamics within a group of closely related proteins, such as a family of homologs, have rarely been experimentally explored, and if so, in the slow-paced millisecondsecond regime [16][17][18] where rapid fluctuations of conformational adaptations are not resolved.Consequently, little is known about whether structural homologs display conserved ultrafast protein dynamics throughout evolution.How may nano-scale protein dynamics evolve over hundreds of million years within a protein family?
Revealing the rapid dynamic processes within proteins requires the use of an appropriate toolkit.Thus far, the conservation of protein structures has been primarily observed through structure comparison using Xray crystallography [19][20][21][22] .X-ray crystallography provides valuable insights with a predominantly static view of proteins, but lacks the mechanistic intricacies that define their dynamics.As an alternative approach, NMR spectroscopy excels at resolving small conformational differences and dynamics in equilibrium 11,16,17,23 , yet it falls short in recording non-equilibrium processes.
In contrast, infrared spectroscopy is sensitive to subtle differences in protein conformations and is a power-ful tool to temporally resolve fast dynamical processes within proteins 8,9 .In combination with an phototrigger, this technique enables the initiation and monitoring of sequential destabilization within a protein complex, with a temporal resolution as fast as a picoseconds 6,7,24,25 .The key challenge lies in investigating the specific time points at which certain processes occur, in order to resolve the influence of evolution on molecules that are inherently dynamic and exhibit fluent transitions between conformational states.

MCL-1: A PRIME EXAMPLE OF CONSERVATION
This study is concerned with the protein myeloid cell leukemia 1 (MCL-1), a member of the BCL-2 protein family, which plays a crucial role as a key regulator of apoptosis, the programmed cell death 3,29 .It is found not only in humans, but also in a diverse range of metazoan organisms 4,30 .Functioning as an anti-apoptotic protein, MCL-1 interacts promiscuously with pro-apoptotic factors through α-helical domains known as BCL-2 homology domain 3 (BH3) 4,29,31,32 , e.g. the BH3 domain of the pro-apoptotic protein PUMA 26,33 (Fig. 1a).Homologs of this protein family have been identified in all vertebrates and even in more distantly related species such as sponges 34 and Cnidaria 35 , whose last common ancestor with Homo sapiens existed over 700 million years ago 5 .
We selected ten MCL-1 homologs (Fig. 1b) from species, whose last common ancestors with Homo sapiens are distributed equidistantly on an evolutionary time axis up to a billion years from present day to the past.We opted for a horizontal approach by comparing se- quences of currently living species, as opposed to a vertical approach involving the reconstruction of ancestral proteins 36,37 .Besides Homo sapiens, we included Mus musculus, Bos taurus, Gallus gallus, Alligator mississippiensis, Xenopus laevis, Danio rerio, a Petromyzon marinus candidate 38 , Lingula unguis, and Hydra vulgaris.
Before exploring the protein dynamics for this homolog selection, our objective was to unequivocally establish the conservation of both the structure and function of MCL-1.By comparing the amino acid sequences of the homologs to their human equivalent, we found that sequence identity dramatically decreased as a function of evolutionary divergence (Fig. 1c), approaching a level of saturation at 25% where homology becomes challenging to detect 39 .The conserved amino acid residues are mostly associated with the canonical binding groove (Extended Data Fig. 1a), consistent with the prevailing scientific perspective 40 , or are localized at the hydrophobic core of the protein.As solely the human and murine homologs bear experimentally acquired structures (e.g.PDB: 6QFM, 2ROC), we used two structure prediction models, AlphaFold 27 and RosettaFold 28 , to compute the structures for the remaining homologs (Extended Data Fig. 1b).In comparison to their experimental equivalents, we found conserved topologies (TM scores ≥ 80% 41,42 , Fig. 1d) and only small spatial differences between the predicted protein backbone (RMSD ≤ 2.5 Å, Fig. 1e).A subtle correlation between inferior structural conservation and increased divergence time became visible.Nevertheless, the predictions show that, although sequences might differ strikingly, MCL-1 structure did not substantially change over a long evolutionary time scale 43 .
The primary function of MCL-1, i.e., the ability to strongly bind the BH3 domain in its binding pocket, which makes it a pivotal anti-apoptotic regulator, is also conserved.We experimentally determined MCL-1's binding affinity for a uniform PUMA BH3 ligand (bearing mu-tations for crosslinking, see extended Data Fig. 2), with K D values ranging from 100 nM to 1 µM for most homologs.We detected a weak correlation of log K D , which refers to the binding free energy, with evolutionary divergence time (Fig. 1f).Notably, homologs from both Hydra vulgaris and Homo sapiens, separated by an evolutionary distance of over 700 million years, bound the same ligand with comparable affinities (K D,Hydra = 220 nM, K D,Homo = 480 nM).Given its critical function as a 'life/death switch' 3 in numerous animal species, this result confirms that MCL-1 indeed exhibits a high degree of structural and functional conservation, manifesting in minor differences at the molecular level.
MCL-1's role as a prime example of structural and functional conservation raises the question of whether the dynamics of the protein are also conserved.Are the nanosecond processes occurring in human MCL-1 also present in Hydra vulgaris MCL-1?

CONSERVATION OF PROTEIN DYNAMICS
To examine the impact of extremely slow evolutionary processes on the fast-paced protein dynamics of MCL-1, we used transient infrared spectroscopy in combination with a photoswitchable azobenzene moiety that is covalently bound to the PUMA BH3 ligand (Fig. 2a, Methods).In its cis-state, the crosslinked photoswitch additionally stabilizes the ligand inside the binding pocket (Extended Data Fig. 2m).Conversely, the light-induced transition from the cis-to the trans configuration leads to a reduction in α-helicity (Extended Data Fig. 2n), indicating a destabilization of PUMA BH3.
Considering a time frame from pico-to microseconds 25 , we studied the protein dynamics in a pump-probe experiment where the cis-to-trans isomerization of the photoswitch is triggered by an ultrashort UV/VIS laser pulse at 420 nm and the protein vibrational spectrum is probed in the mid infrared region around 1650 cm −1 (Fig. 2b).In this spectral region, C=O stretch vibrations of the protein backbone can be observed.Negative (blue) and positive (red) absorption changes serve as indicator for structural alterations 46 .We obtained homolog-specific kinetic footprints for the ten investigated species (Extended Data Fig. 3, exemplified for M. musculus in Fig. 2c).Analogous to fossil footprintsthe paleontologic counterpart -, the kinetic footprints display comparable elements.All of them are similarly shaped, displaying a blue shift of a band at 1645 cm −1 , which reveals a negative bleach towards a new (positive) band at 1675 cm −1 (Fig. 2c, triangle).The signal appears within the low nanosecond time frame for all of the homologs and can be attributed to α-helix unfolding 7,9,47 .More strikingly, the kinetic footprints exhibit diverging, species-specific details, which are particularly well visible for the spectral feature between 1655 cm −1 and 1685 cm −1 (Fig. 2c, circle), and a late negative feature at 1620 cm −1 forming at around 100 ns (Fig. 2c, square).The first-mentioned feature (circle) is especially pronounced for mammalian/avian/reptilian homologs (Extended Data Fig. 3a-e), but loses its distinct appearance more and more for species with higher evolutionary divergence (P.marinus, L. unguis, H. vulgaris (Extended Data Fig. 3h-j), displaying a solitary, less emphasized maximum at 1660 cm −1 .Furthermore, the kinetic footprints of the non-Gnathostomata P. marinus, L. unguis, and H. vulgaris, lack the late negative feature at 1620 cm −1 .
All kinetic footprints are dominated by three phases of dynamic activity, an early-, mid-, and late phase, where the intensity of spectral features grows or decreases significantly (exemplified for M. musculus in Fig. 2d, for all other species in Extended Data Fig. 4).To fathom these three dynamic processes and their corresponding time constants, we analyzed the kinetic footprints with global multiexponential fitting (Fig. 2d, details in Methods section) 7,44,45,48 .Our analysis demonstrates that there are four states of molecular rearrangement upon photoperturbation, populated with time constants τ early , τ mid , and τ late .The time intervals in which the three observed processes take place are very narrow for the ten homologs we investigated, evidencing that not only the structure and function of MCL-1 are conserved across a wide and diverse range of today's living animals (Fig. 1) but also the underlying protein dynamics (Fig. 2e).This stands in stark contrast to the significant alterations that we observe for the primary structure of the protein homologs (Fig. 1c).When we plot the time constants against an evolutionary time scale (Fig. 2f), we find that the processes populated with τ mid , correlate with the evolutionary divergence.In contrast, we did not detect similar protein dynamic drifts for the other two time constants, showing an absence of correlation of early-and late protein response with evolutionary divergence.On the other hand, if the time constants are plotted in dependence of the experimental binding affinities, it becomes evident that the processes populated with τ late are strongly correlated with the protein's affinity (Fig. 2g).
We specified which parts of the protein complex contribute to which process by recording kinetic footprints for 13 C-15 N-labelled MCL-1 of M. musculus in complex with non-labelled PUMA BH3 (Extended Data Fig. 5).Separating in this way the protein from the peptide response, we found that time constant τ early (= 0.9-3.5 ns) can be attributed to the α-helical unfolding of the PUMA BH3 peptide.(Fig.2h) The time constant τ mid (= 21-50 ns) corresponds to spectral features which shift ≈ 50 cm −1 for isotope labelled MCL-1 (Extended Data Fig. 5), and can thus be traced back to a initial response of MCL-1, that potentially allows to rearrange and cope with the conformational destabilization originating from the binding pocket.Apparently, mammalian homologs exhibit an earlier MCL-1 adaption upon destabilization than non-mammalian/vertebrate species and again other non-vertebrata (Fig. 2f, red).Finally, the terminal time constant τ late (= 0.7-3.6 µs) corresponds to mutual re-FIG.2. The conservation of ultrafast protein dynamics in MCL-1.(a) The protein MCL-1 in complex with the photoswitchable PUMA BH3 peptide.(b) Transient infrared spectroscopy of the photo-perturbed MCL-1/PUMA BH3 complex results in kinetic footprints for all homologs, exemplarily displayed in (c) for Mus musculus.The symbols serve as reference points for explanations in the main text.(d) Three dominating phases of increased dynamic activity are assessed (early-, mid-, late phase; dashed lines).Global multiexponential fitting with three time constants yields fits (red/blue) that cover the raw data (grey) well.Evolution-associated difference spectra 44,45 (lower panel) were calculated for state S1 (red), S2 (yellow), S3 (green), and for St (blue) with time constants τ early , τ mid , and τ late .(e) The difference spectra of all homologs display a high degree of similarity.(f) Time constants of increased dynamic activity τ early , τ mid , and τ late against evolutionary divergence in million years, Ma.(g) Time constants τ early , τ mid , and τ late against MCL-1's affinity for PUMA BH3.Data in (f) and (g) are displayed with linear fits ± standard deviation (yellow) and correlation coefficients r (Pearson).(h) Isotope labeling (Extended Data Fig. 5) helped to separate the signal contribution of MCL-1 and PUMA BH3 spatially and temporally.The time constants were assigned to dynamic processes in the protein complex (schematic overview).
arrangements in the whole complex.The results are in line with previous observations for the isotope labelled human MCL-1/BIM complex 7 .
From our results, one might speculate whether MCL-1's initial response (τ mid ) has met with less selective pressure in the past, causing it to drift.This hypoth-esis is supported by the absence of any discernible correlation between τ mid and the protein affinity (Fig. 2g), implying that the function of the protein is seemingly not entangled with this dynamic process.In contrast, a robust correlation between τ late and the K D (Fig. 2g) indicates that the late mutual rearrangements of MCL-1 and PUMA BH3 (in the microsecond regime) are connected to the function of the protein.From our observations, it seems that the relationship between the late dynamic response and the protein affinity is conserved and cannot be inferred from the evolutionary separation of species; other factors must be at play.Irrespective of how to evaluate the given correlations, what remains truly remarkable is that our results provide an unprecedented opportunity to gain insights into the speed and extent of the impact of evolution on dynamical processes.

CONCLUSION
MCL-1 is a critical player in apoptosis 29 , not only in human beings but also in a great variety of animals 30 .By experimentally studying ten MCL-1 homologs and their interactions with a photo-switchable ligand PUMA BH3, we gained valuable insights into the dynamics of the proteins on a broader evolutionary time scale.Using time-resolved infrared spectroscopy, we successfully recorded the kinetic footprints of the MCL-1/PUMA BH3 complex and analytically compared them -similar to bones, skulls, and footprints in the classic field of paleontology 49,50 , or protein structures and genetic information in its molecular form [51][52][53] .
Our findings reveal a remarkable degree of conservation for the protein dynamics across the homologs, highlighting the importance of these processes in preserving their anti-apoptotic function over a span of nearly a billion years.Of particular interest is the correlation we observed between one of these ultrafast processes and the evolutionary divergence among the protein homologs, a drift in protein dynamics in the nanosecond range.This discovery challenges the prevailing focus on resolving protein structures 22 and dynamics in equilibrium 16 or analyzing genomic data 53 to understand evolution.Instead, our work highlights the importance of considering nanosecond protein dynamics as a crucial factor in unraveling the evolutionary history of these proteins.With this approach, we build a bridge between the shortest (1 ns = 10 −9 s) and the largest conceivable timescales in living matter (300 Ma ≈ 10 16 s).
Overall, our study defines a starting point for exploring the dynamics of countless other proteins with varying degrees of conservation.By investigating different systems that are more or less conserved, we can gain valuable insights into the extent of evolution's impact on nanosecond processes, and how these rapid processes translate to slow-paced protein function.

Phylogeny and Bioinformatics
From the countless species in the tree of life we chose ten MCL-1 homolog sequences (Fig. 1b).Alongside the Homo sapiens homolog, our selection encompasses a variety of species, including mammalian (Mus musculus, Bos taurus), avian (Gallus gallus), reptile (Alligator mississippiensis), amphibian (Xenopus laevis), bony fish (Danio rerio), and other farther related eumetazoan homologs (Petromyzon marinus, Lingula unguis).Notably, we also incorporated a homolog from Hydra vulgaris, one of the most distantly related organisms known to exhibit BCL-2 regulated apoptosis 35 .The curated selection represents species whose last common ancestors existed at quasi-equidistant intervals spanning nearly a billion years of evolutionary history.
To assemble our dataset, we accessed amino acid sequences from the Uniprot database.All entries shared the common identifier "MCL-1" in their title or description.The amino acid sequence for P. marinus was added to the selection with the help of Jeramiah Smith (gene on Chr52: 9161036..9167581, + strand; annotated: PMZ 0059412-RA) 38 .The sequences were aligned to the human variant (soluble domain, ∆N-∆C aa 171-327 33 ) and harmonized in length (≈150-160 aa) (Extended Data Fig. 1a).The chosen sequences (refer to Extended Data Tab.I) were initially controlled for by predicting their structure using AlphaFold and RosettaFold (see below) and aligning them with experimental structures from Homo sapiens and Mus musculus.From the sequences, we generated a multiple sequence allignement using Clustal Omega (EMBL-EBI) 54 (Fig. 1a).
The phylogeny in Fig. 1b was obtained from the Time-Tree database (http://timetree.org) 5 .It was not computed from the investigated MCL-1 sequences.In contrast, the given phylogeny was constructed from median and adjusted divergence times which were estimated by TimeTree based on values from an abundance of published studies.The divergence times, always related to H. sapiens and tabulated in Extended Data Tab.I alongside their corresponding confidence interval, reflect the most current scientific understanding (June 2023).In figures Fig. 1c-f, and Fig. 2f, the divergence time is given in million years, Ma.
The experimental structures of MCL-1/PUMA were retrieved from PDB (Mus musculus: 2ROC; Homo sapiens: 6QFM).In addition, we predicted the structures of all MLC-1 homologs with AlphaFold 27 and RosettaFold 28 .We used ColabFold 55 to generate AlphaFold-predicted structures, and the Robetta server 28 for RosettaFold-predicted structures, both with default parameters.To estimate the structural similarity between all protein pairs, we performed an all-againstall alignment of the predicted structures and computed the TM score and root-mean-square deviation (RMSD) of each protein pair using TM-align 41 .For both Al-phaFold and RosettaFold, we selected the top ranked structure out of the five predictions for downstream analyses.We evaluated the quality of the predicted structure using AlphaFold predicted local distance difference test (pLDDT), a per-residue confidence metric which estimates how well the predicted structure would match with an experimental one and which has been shown to be well-calibrated 27 .All our predicted structures have high average pLDDT values, ranging from 0.83 to 0.93, indicating good quality predictions.

Protein preparation
Examining protein function and dynamics, we expressed ten different MCL-1 homologs using a Escherichia coli BL21 expression strain (Fig. 2a).Initially, the bacterial cells were transformed with a pET-30a(+) plasmid containing the corresponding MCL-1 homolog gene, using electroporation.Positive clones were selected through Kanamycin resistance.For standard expression, bacterial cultures were cultivated in lysogeny broth medium until reaching an optical density of OD 600 = 0.6.The expression was induced by adding 700 µM IPTG, followed by incubation at 30 °C for 20 hours.Cell harvest was carried out through centrifugation (3000 xg).In order to generate heavy, uniformly 13 C 15 N-labeled MCL-1, bacterial cultures were grown in minimal medium supplemented with solely heavy carbon and nitrogen sources.The cells were cultivated to an OD 600 of 0.6, induced with 1 mM IPTG, and then further incubated at 30 °C.The expression was stopped after 4 hours with cell harvest as described above.Cell lysis was achieved by subjecting the harvested cells to sonication (20 kHz, 4 x 1 min pulses).The lysed cell suspension was purified using Ni-affinity chromatography and a His 6 -Tag located at the N-terminus of the protein.Purification was carried out under native conditions.The N-terminal His 6 -Tag was removed by 3C protease cleavage.Throughout this study, all analytical procedures were performed in a sample buffer composed of 50 mM Tris (pH 8) and 125 mM NaCl.Mass spectrometry was used to assess the protein's integrity and sample purity.For long-term storage, the samples were kept at -80 °C.In total, we could express the homologs of ten species given in the main text (Extended Data Fig. 6).Under identical conditions, however, we could not express Ornithorhynchus anatinus, Orchesella cincta, and Acanthaster planci homologs at adequate concentrations.

Peptide preparation
PUMA BH3 (EEQWAREIGAQLRCMADDLNCQY-ERV) was synthesized using solid-state peptide synthesis on a Liberty 1 peptide synthesizer (CEM corporation, Matthews, NC, USA).In this study, the peptide was deliberately modified by introducing two mutations -replacing Arg143 and Ala150 with Cys residues -compared to the native mammalian version.These Cys residues were incorporated distal to the hydrophobic binding interface, to enable the covalent linkage of a photoswitchable azobenzene moiety.To achieve this linkage, the watersoluble photoswitch (3,3'-bis(sulfonato)-4,4'-bis(chloroacetamido)azobenzene) 56 and the peptide with reduced Cys residues were together incubated in a 20 mM Tris (pH 8.5) at a temperature of 50 °C, under continuous stirring for a duration of 20 hours.Hereafter, the reaction product underwent purification using both anion exchange and reversed-phase chromatography (C18 10µm) to isolate the successfully linked peptide.For final preparation, the buffer of the isolated linked peptide was exchanged through dialysis against the sample buffer (50 mM Tris pH 8, 125 mM NaCl).The linkage's success, as well as the peptide's purity and integrity, were controlled via mass spectrometry.

Circular dichroism spectroscopy
The expressed MCL-1 homologs have in common that they contain eight α-helical elements 33 , and exhibit a circular dichroism spectrum that is typical for α-helical structures (Extended Data Fig. 2b, yellow).In contrast, their peptide ligand PUMA BH3 is intrinsically disordered in isolation 57 (Extended Data Fig. 2b, grey).When in complex with MCL-1, PUMA BH3 assumes an α-helical shape (Extended Data Fig. 2b, black).
We utilize circular dichroism spectroscopy to accomplish two distinct objectives: (i) to evaluate the α-helical content of the MCL-1 and PUMA BH3 complex at a constant concentration, thereby assessing whether they are correctly folded, and (ii) to generate binding curves and determine dissociation constants (K D ) for all analyzed MCL-1 homologs.To record binding curves and assess the K D values, we exploited the nature of PUMA BH3 which is intrinsically disordered in solution and only exhibits an α-helical secondary structure when bound by MCL-1's binding groove.Hence, for an increasing concentration of bound PUMA BH3, and a constant concentration of MCL-1, the α-helical content added by titration reflects the fraction of bound peptide.
For the first aspect (i), a quartz glass cuvette with a 1 mm path length was employed, and the sample concentration was maintained at 20 µM.We measured the spectrum between 200-260 nm at room temperature.Hereby, we examined the α-helical content of the MCL-1 and PUMA BH3 complex which served as a control to for their correct structural conformation (displayed in Extended Data Fig. 2b).
For the second aspect (ii), MCL-1 was brought to a concentration of 2 µM.A quartz glass cuvette with a path length of 1 cm was used, and continuous stirring was maintained during the spectroscopic measurements at room temperature.To record the binding curves, we titrated both the linked and unlinked forms of the PUMA BH3 peptide to the MCL-1 homolog, offering a comprehensive understanding of the binding affinity of photoswitchable and non-photoswitchable complexes.The circular dichroism was recorded at 222 nm as a function of increasing PUMA BH3 concentration.In both scenarios (i) and (ii), measurements involving the photoswitch-able PUMA BH3 were conducted for both the cis-state (achieved through illumination with a 375 nm laser) and the dark-adapted trans-state.
By recording the α-helical content at 222 nm as a function of increasing PUMA BH3 concentration, we received binding curves for all MCL-1 homologs.In order to calculate the dissociation constant K D , we fitted the data to a two-state binding equilibrium 58,59 : where [M] is the initial concentration of the MCL-1, [P] is the initial concentration of PUMA BH3 given to the solution, and [MP] is the concentration of the proteinpeptide complex.For a constant [M] = 2 µM and a variable [P], the fraction of bound peptide can be understood as: The covalently bound photoswitch in the cis-state stabilizes PUMA BH3 inside the binding pocket (Fig. 1a), with significantly lower K D values for all of the homologs (Extended Data Fig. 2m).For PUMA BH3 in the cis state, Homo sapiens, Bos taurus, and Alligator mississippiensis homologs showed the highest affinities, with K D values in the low nanomolar regime (<10 nM), a region that was classified as physiological for wild type PUMA 31 .Switching the photoswitch from its cis-to its trans configuration results in a loss in α-helicity (Extended Data Fig. 2n) and in the destabilization of PUMA BH3.

Transient infrared spectroscopy
MCL-1 and PUMA BH3 were mixed in a 1:1 ratio prior to the spectroscopic experiment.To ensure high signal strength in transient infrared spectroscopy, both the protein and peptide were brought to concentrations of 600 µM each in the final sample.The overall sample volume was 800 µL.Considering concentrations >500 µM, it is expected that the PUMA BH3 peptide will be predominantly bound within MCL-1's binding pocket, as illustrated in Extended Data Fig. 2. To exclude H 2 O from spectroscopic experiments, the employed buffer was exchanged to a corresponding buffer containing D 2 O. Stringent precautions were taken to avert H 2 O contamination by preserving the sample within a nitrogen atmosphere devoid of water vapor.
For pump-probe measurements, a pair of electronically synchronized 2.5 kHz Ti:sapphire oscillator/regenerative amplifier femtosecond laser systems (Spectra Physics) were employed, offering a maximal delay of 45 µs 25,60 .One of these laser systems, featuring frequency-doubled pulses (420 nm, 3 µJ per pulse, focused to an approximate beam diameter of 140 µm within the sample, and stretched to ∼60 ps to minimize sample deposition on the sample cell windows), were used to induce the cis to trans-isomerization of the photoswitch.The second laser system was applied to generate infrared probe pulses via an optical parametric amplifier (100 fs, spot size 110 µm, center wavenumber 1660 cm −1 ).To ensure a consistent sample environment, the sample was continuously circulated within a closed-cycle flow cell system.This system consisted of a reservoir and a CaF 2 cell featuring a 50 µm optical path length.Before entering the measurement cell, the sample was irradiated in a pre-illumination step using a 375 nm continuous wave diode laser (90 mW, CrystaLaser), in order to optimally prepare the sample with >85% in the cis-state.

Data analysis
From time resolved infrared measurements, we obtained kinetic footprints in the form of 2D data sets d(ω i , t j ) as a function of probe frequency ω i and pumpprobe delay time t j (Fig. 2c, Extended Data Fig. 3).For each homolog, we subjected the 2D dataset to a global multiexponential fitting 45 , operating under the premise that the investigated system can be understand as interconverting discrete states with time-invariant spectra.
We employed multiexponential functions with amplitudes a(ω i , τ k ) and a global set of time constants τ k for fitting the experimental data 48,61,62 : The time constants τ k were treated as free fitting parameters, with the constraint of a minimal number of exponential terms.Based on observations with similar systems 7,32 we excluded data before 300 ps, to prevent the influence of the pump pulse (60 ps pulse length), which results in a strong "heat signal" at 100 ps, induced by azobenzene photoisomerization, which can be universily observed for this kind of experiment 7,24,32 .Three time constants τ early , τ mid , and τ late were needed to adequately fit the data, dissecting the dynamic response in an early-, mid-, and late phase (Extended Data Tab.II).The one exception is D. rerio, which required a fourth time constant τ D.rerio = 300 ns to adequately fit the data.
Under the assumption of a sequential, unidirectional process with four states S 1 , S 2 , S 3 , S t and three time constants τ early , τ mid , τ late connecting them, we calculated concentration profiles for each state as well the corresponding evolution-associated difference spectra 44 , which are depicted in Fig. 2d,e.Commencing with state S 1 , all subsequent evolution-associated difference spectra exhibited a blue shift from 1645 cm −1 to around 1675 cm −1 .Equally to our observations for the raw data, a very distinct positive feature at 1660 cm −1 was detected in the evolution-associated difference spectra of the latest two states (Fig. 2d,e, green and blue).
With the help of isotope labeling 63 (Fig. 5), we could assign this distinct spectral maximum, populated with τ mid , to the initial response of MCL-1 upon photodestabilization of its ligand.The early response at τ early exclusively originates from the unfolding of PUMA BH3.The terminal, late response at τ late results from mutual, heterogeneous rearrangements in MCL-1 and PUMA BH3.The kinetic footprints display comparable elements.We observe fine differences between closer related homologs and diverging, gradually emphasized features for farther related homologs, indicating an evolutionary timescale in which protein dynamic processes have changed.
Extended Data Fig. 4. The dynamic response upon photo-perturbation can be subdivided into an early-, mid-, and late phase, exemplified here for kinetic traces at 1640 cm −1 and 1660 cm −1 .When analyzed with global multiexponential fitting and three time constants τ early , τ mid , and τ late (given as dashed lines), the resulting fits (red/blue) are congruent with the time traces (grey).The one exception is D. rerio, which requires an additional fourth time constant τD.rerio = 300 ns (*) to adequately fit the data.Extended Data Fig. 5. Transient infrared spectroscopy with isotope labeled MCL-1/PUMA BH3 complexes.From ten homologs, M. musculus was chosen as a representative, as 13 C 15 N-labeling is highly cost intensive.(a) Kinetic footprints of unlabeled samples (upper panel), and of samples with 13 C 15 N-labeled MCL-1 (lower panel).The early protein response (triangle) is not shifted for the labeled sample.The mid protein response (circle) manifests in a distinct sharp feature that shifts from 1660 cm −1 to 1610 cm −1 upon isotope labeling (dashed lines).Isotope labeling did not separate any spectral features at the late phase of the protein response.(b) Evolution-associated difference spectra (see Data analysis in the Methods section) display a relatively sharp positive band in state S3 and S4 (circle).In the non-labeled complex, the sharp maximum coincides with the broad positive band of the blue shift.For the labeled complex, it is shifted by 50 cm −1 from 1660 cm −1 to 1610 cm −1 , as expected for 13 C 15 N-labeling 63 .Extended Data Fig. 6.Expression yield of various MCL-1 homologs from 10 L Escherichia coli culture.A yield of 1 µmol was seen crucial for spectroscopic analysis.

FIG. 1 .
FIG. 1. Structure and function of MCL-1 is conserved.(a) NMR structure of MCL-1 (grey) complexed with PUMA BH3 (yellow) (PDB: 2roc) 26 .(b) Phylogeny of ten species whose MCL-homologs were selected for this study.The phylogeny and the corresponding evolutionary divergence time in million years (Ma) were taken from TimeTree5 and cover the current state of science (July 2023) 5 .(c) Sequence identity of all investigated MCL-1 homologs (compared to H. sapiens) against evolutionary divergence time of the corresponding species.(d,e) Structural similarity between MCL-1 homologs (compared to H. sapiens), predicted with AlphaFold 27 (blue) and RosettaFold 28 (red).(f) MCL-1 homolog binding free energy for the PUMA BH3 peptide, plotted against the evolutionary divergence time.Yellow, linear fit ± standard deviation.The Pearson correlation coefficient r = 0.39 indicates that the binding free energy correlates weakly with evolutionary divergence time.

ABBREVIATIONS
ABBREVIATIONS