Available via license: CC BY-NC 4.0
Content may be subject to copyright.
1
Conserved energetic changes drive function in an ancient protein fold
Malcolm L. Wells1*, Chenlin Lu1*, Daniel Sultanov1*, Kyle C. Weber1*, Zhen Gong1, and Anum
Glasgow1†
1 Department of Biochemistry and Molecular Biophysics, Columbia University Irving Medical
Center, New York, NY, United States
* These authors made equal contributions.
† Corresponding author. Email: ag4522@cumc.columbia.edu
Abstract
Many protein sequences occupy similar three-dimensional structures known as protein folds. In
nature, protein folds are well-conserved over the course of evolution, such that there are
100,000 times as many extant protein sequences than there are folds. Despite their common
shapes, similar protein folds can adopt wide-ranging functions, raising the question: are protein
folds so strongly conserved for the purpose of maintaining function-driving energetic changes in
protein families? Here we show that a set of key energetic relationships in a family of bacterial
transcription factors (TFs) is conserved using high-resolution hydrogen exchange/mass
spectrometry, bioinformatics, X-ray crystallography, and molecular dynamics simulations. We
compared the TFs to their anciently diverged structural homologs, the periplasmic binding
proteins (PBPs), expecting that protein families that share the same fold and bind the same
sugars would have similar energetic responses. Surprisingly, our findings reveal the opposite:
the “energetic blueprints” of the PBPs and the TFs are largely distinct, with the allosteric
network of the TFs evolving specifically to support the functional requirements of genome
regulation, versus conserved interactions with membrane transport machinery in PBPs. These
results demonstrate how the same fold can be adapted for different sense/response functions
via family-specific energetic requirements – even when responding to the same chemical
trigger. Understanding the evolutionarily conserved energetic blueprint for a protein family
provides a roadmap for designing functional proteins de novo, and will help us treat aberrant
protein behavior in conserved domains in disease-related drug targets, where engineering
selectivity is challenging.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
2
Over 700 million natural proteins (1) fold into only a few thousand spatial arrangements
of secondary structures, or protein folds (2–6). Extant protein folds have been maintained over
billions of years despite low sequence identity (7). To evolve a new function, a protein typically
samples sequence changes that preserve its fold while varying its conformational dynamics (8).
Sequence changes to a protein that disrupt essential allosteric regulation are disfavored (9).
Thus, proteins are robust to mutations that do not disrupt essential dynamics or function,
enabling sequence diversity while preserving the basic topology.
Unfortunately, methodological limitations in observing energetic relationships in proteins
limit our capacity to identify the functionally important elements of the fold for most proteins. For
example, protein functions may depend on perturbation-driven changes in the local stability of
specific residues, secondary structures, intramolecular interactions, and quaternary interfaces.
Consequently, we do not generally understand how new functions emerge from old folds.
Using a high-resolution integrative structural biology strategy, we determined the
evolutionarily conserved, functionally important energetic relationships in two bacterial protein
families that share the Venus Flytrap (VFT) fold. VFT-fold proteins are found across the
domains of life (7, 10, 11). Five proteins from E. coli serve as our model system for probing
energetic redistribution in the families: three dimeric LacI/GalR transcription factors (TFs) and
two monomeric periplasmic binding proteins (PBPs). The TFs regulate sugar metabolism genes
in the cytoplasm, while the PBPs are periplasmic transporters for the same sugars (12, 13). The
PBPs and TFs most likely evolved ligand binding specificity independently after a DNA binding
domain (DBD) acquisition event that occurred before the divergence of eubacteria (7, 14).
Despite their shared topologies, in experimentally solved structures, the sugar-induced
conformational changes in the VFT domains of TFs are subtler than in PBPs due to the dimer
state of the TFs (12, 13, 15). The family’s namesake TF, the lac repressor (LacI), reweights
specific intramolecular contacts to switch between its transcriptionally repressed and active
functional states without dramatic changes to the VFT domain structure (16–18). The DBD of
each TF has a helix-turn-helix (HTH) motif that is connected by a hinge helix to the VFT domain.
TFs switch between their repressed and active states by allosterically restructuring their DBDs
in response to binding inducer molecules in their VFT domains. We hypothesized that the VFT-
HTH architecture might be so strongly conserved in the TFs because they require the DBD and
VFT domain to be coupled via a minimal set of shared energetic relationships. Because the
PBPs bind the same sugars, we hypothesized that they evolved similar sugar-driven energetic
changes within the constraints of the VFT fold.
LacI/GalR TFs and PBPs have distinct sequence conservation
To investigate whether evolution of separate functions in PBPs versus LacI/GalR TFs
precipitated amino acid changes in the VFT fold that are specific to either family, we built a gene
tree based on 2,222 diverse LacI/GalR TF and PBP sequences using only their VFT domains
(Figs. 1A, S1). Although the large number of sequences and the small number of inspected
amino acid sites limit the resolution of the more internal branches, all TFs form a separate clade
from all PBPs with high confidence (bootstrap support 100%), corroborating a previous study (7)
and suggesting independent evolution of the VFT domain in the two lineages.
Comparing conservation scores (see Methods) for each protein family revealed distinct
patterns of sequence conservation (Fig. 1B). While the PBPs are uniformly highly conserved,
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
3
Figure 1. Sequence conservation and structure of LacI/GalR TFs and PBPs. A, Reconstructed gene tree shows
distinct clades of LacI/GalR TFs versus PBPs. Bootstrap values >90% are in turquoise. B, Sequence conservation is
higher and more uniform for PBPs than for TFs. C, TF sequence conservation echoes mutational data for LacI: more
conserved positions are less tolerant to mutations, causing severe changes to LacI function. D, The TF dimer interface
is among the least conserved regions of the fold (e.g., LacI, right). The homologous region is well-conserved for PBPs
(e.g., RbsB, left). E, We solved 2.1 Å X-ray crystal structures of the RbsR complex with operator DNA rbsO and F, with
ribose (right). The inducer-bound states of LacI and RbsR VFT domains are structurally similar by X-ray crystallography.
G, Zoomed view of key RbsR interactions with ribose with 2Fo–Fc maps for sidechains and ligand at 1σ.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
4
the TFs exhibit “valleys” of higher sequence diversity. As expected, there is high conservation in
the DBD. However, the N-terminal lobe of the VFT domain is more permissive to mutations than
the C-terminal lobe (Fig. 1B). The conservation pattern recapitulates experimental phenotypic
outcomes in LacI mutants (Fig. 1C) (19).
Interestingly, we found substantial sequence variability in the TF dimer interface,
reflecting the biochemical diversity of interfaces across many TFs (Figs. 1D, S2) (15, 20–22).
Although several solvent-exposed helices in the TFs are also sequence-diverse, the variability
in the dimer interface is surprising because the LacI/GalR TFs are obligate homodimers, with no
known monomeric functional state. By contrast, the same region is highly conserved in PBPs
(Fig. 1D), likely due to interactions with nutrient transporters (23). Differences in sequence
conservation in the two families reflect their distinct functional constraints while sharing a fold.
LacI/GalR TFs are structurally similar in different functional states
Bioinformatics-derived sequence-level differences do not reveal the structural basis for
ligand regulation in the TFs, which requires comparing high-resolution X-ray crystal structures of
related TFs in different functional states (24, 25). Of the 41 experimentally solved structures of
full-length LacI/GalR TFs, 31 are solved to <3 Å resolution (Table S1). Among all 100 published
TF structures that include at least the VFT domain, 83 are solved to <3 Å resolution. Overall, all
TF VFT domain structures have <4 Å pairwise root-mean-square deviation (RMSD) across their
protomers (Fig. S3). We are interested in TFs that activate transcription by binding small
molecules, and of this subset, only the LacI VFT domain structure is available in both operator-
and inducer-bound states (26–28, 15). In summary, an evolutionarily conserved activation
mechanism in the LacI/GalR family cannot be gleaned from the available structural information.
To properly compare LacI with another sugar-inducible LacI/GalR TF, we solved X-ray
crystal structures of the ribose repressor (RbsR) in its inducer- and operator-bound forms at 2.1
Å resolution (Figs. 1E-G, S4, Table S2). RbsR is induced by ribose and has 33% sequence
identity with LacI (29). While LacI controls lactose metabolism by binding the lac operator DNA
(lacO1), RbsR controls ribose metabolism by binding the rbs operator DNA (rbsO).
Like the LacI complexes with inducer isopropyl β-D-1-thiogalactopyranoside (IPTG) and
lacO1, RbsR-ribose and RbsR-rbsO have 1.58 Å backbone RMSD in the VFT domain (Fig. 1F).
The C-terminal VFT lobe is almost identical between the operator- and inducer-bound states of
both TFs (0.40 Å and 0.61 Å RMSD for RbsR and LacI, respectively). The two TFs are also
highly structurally similar, with 1.6 Å RMSD in the VFT domain (Fig. 1E, F). The DNA-bound
states for both TFs show the hinge helices buried in the minor groove of the operator, which is
bent at 45° (Figs. 1E, S4-1). Notably, the extra base insertion between the palindromic repeats
in lacO1 compared to rbsO does not change the crystallographic DNA or protein conformation.
Measuring the local stabilities of VFT protein functional states
For transcriptional regulation to occur, the TFs must sample different conformations with
different probabilities in each state, leading to conformational ensembles that are energetically
distinct and not clear from crystal structures (30). To determine how the energy landscape is
conserved in TFs (the “energetic blueprint”), we performed hydrogen exchange/mass
spectrometry (HX/MS) on three paralogous sugar-inducible LacI/GalR TFs in their apo, DNA-
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
5
Figure 2. High-resolution HX/MS. A, We chose two PBPs and three TFs to study by HX/MS, including two PBP/TF
pairs that respond to the same ligands. B, Experimental design and data summary. C, Overview of HX/MS experiment
and analysis using PIGEON-FEATHER, resulting in site-resolved ∆Gop for many residues. D, ∆Gop heatmap for the
proteins ordered by structurally aligned residues, compared to LacI solvent-accessible surface area (SASA) and TF
and PBP sequence conservation. Gray marks an alignment gap. Right: VFT-fold structure schematic. Several regions
are marked by matching numbers on the heatmap and schematic. (1) The DBD has higher ∆Gop in the operator-bound
state than in apo or sugar-bound states in the TFs. (2) A binding pocket loop responds differently to sugar binding in
each protein. (3) Sugar binding stabilizes helix 20 in each protein. (4) Sugar binding stabilizes a C-terminal VFT lobe
segment in the PBPs. (5) Sugar binding stabilizes the N-terminal region of helix 36 for all proteins.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
6
bound, and inducer-bound states: LacI, RbsR, and the galactose repressor (GalR) (31) (Figs.
2A, S5-S7, Tables S3-S5). GalR induces transcription of the gal operon by binding galactose
and has 25% sequence identity to LacI and 31% to RbsR (32). To account for differences in
hydrogen exchange among the TFs that might arise from binding different sugars, we also
performed HX/MS on two PBPs that bind the same sugars as RbsR and GalR: ribose binding
protein (RbsB) and galactose binding protein (MglB) (Figs. 2A, S8, Tables S6, S7). These five
VFT proteins bind the sugars with affinities in the 100 nM to 1 mM range. Both TF/PBP pairs
have 28% sequence identity in the VFT domain.
HX/MS measures the rate of exchange of backbone amide hydrogens with deuterium,
revealing the relative structuredness of regions of the protein in various functional states. Using
PIGEON-FEATHER (17), a high-resolution HX/MS analysis method, we determined the single-
residue free energies of opening (∆Gop) for all three TFs in three states (apo, sugar-bound, and
operator-bound) and both PBPs in two states (apo and sugar-bound) (Fig. 2B, C, Tables S8-
S12). The ∆Gop is a measure of local stability; the highest ∆Gop values in a protein correspond
with its global free energy of unfolding (33). Ligand-driven shifts in ∆Gop report local energetic
changes in the context of the protein ensemble.
The ∆Gop datasets show local and global protein responses to binding sugars and
operators (Figs. 2D, S9, Supplemental Note 1), demonstrating that some effects occur only in
PBPs or TFs (Fig. 2D-1, 2D-4), while others are conserved in all VFT-fold proteins (Fig. 2D-3,
2D-5). Some regions of the VFT fold are differentially stabilized by ligand binding in each
protein. For example, a structurally conserved binding pocket loop (Fig. 2D-2) responds
differently to each sugar in each protein, suggesting protein-specific evolution of a functional
role for the loop to allow E. coli to tune ligand binding affinity and specificity. We discuss state-,
family-, and fold-specific energetic changes in the next sections.
Binding to operator DNA has long-range allosteric effects in all TFs
To learn how the TFs respond to binding operator DNA, we compared their ∆Gop in
operator-bound and apo states (∆∆Gop, DNA-APO) (Fig. 3A, B). As expected, the DBD of each TF
has higher average ∆∆Gop, DNA-APO than the VFT domain due to interactions with DNA that
structure the DBD (Fig. 3C) (34). Operator binding also stabilizes the hinge helix (35) and
several VFT-DBD interactions (Fig. 3C, D).
Remarkably, binding to operator DNA also allosterically stabilizes the TFs far beyond the
DBD. One example is the structuring of a C-terminal dimer interface helix (the “metastable
helix”, H32 in Fig. 2D), which points to increased dimer stability in LacI/GalR TFs upon operator
binding regardless of the DNA sequence, length, or symmetry (Fig. 3B-1, D-1). This region is
identical in crystal structures of LacI and RbsR in all functional states (Figs. 1F, S4-3) and in
structural models of GalR (36). The behavior of H32 exemplifies a functionally important,
evolutionarily conserved shift in a protein’s ensemble that occurs without a crystallographic
conformational change. While the regions flanking the metastable helix are conserved in the
LacI/GalR MSA (Fig. 1D, Fig. S2), the helix itself is not, suggesting that evolution might tune this
region to control DNA binding affinity and inducer sensitivity. In LacI, mutations to H32 affect the
quaternary structure (37). By contrast, H32 is well-conserved in PBPs (Fig. 1D), likely because it
forms part of the binding interface for membrane transporters.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
7
Figure 3. Conserved allosteric effects of DNA binding in TFs. A, Each TF responds to binding its own operator
DNA with global changes in local stability. B, Conserved energetic effects of operator binding in the three TFs. C,
Although there is global restructuring in the TFs, operator binding causes a larger increase in average ∆Gop in the DBD
than the VFT domain. D, conserved increase in ∆Gop for each TF in key operator-responsive regions.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
8
We also observed other conserved effects of operator binding at the dimer interface,
including stabilization in the “transducer helix” H3 (Figs. 3D-2, S4-2), which bridges the binding
pocket with the DBD and experiences ligand-dependent shifts in ∆Gop that are discussed further
in the next section. Operator binding also stabilizes the central residue of the N-terminal VFT
beta strand E5 in each TF, together with its hydrogen bond partner in the neighboring strand
and in the DBD (Fig. 3B, D-3).
X-ray crystal structures of LacI-lacO1 and RbsR-rbsO showed that TF binding bends
operator DNA (Fig. 1E, S4-1). NMR spectroscopy on the LacI DBD corroborated this result and
further demonstrated that DBD binding to off-target DNA maintains a linear DNA conformation
(38). However, the structural explanation for DNA operator binding specificity by LacI remains
mysterious because the behavior of the VFT domain in the presence of off-target DNA could not
be studied using either technique. To explore LacI response to on-target vs. off-target DNA in its
VFT domain, we compared peptide-level HX/MS of LacI in solution with lacO1 (on-target,
symmetric), rbsO (off-target, symmetric), and scrO, a scrambled version of the lacO1 sequence
(off-target, not symmetric) (Fig. S10, Table S13). While rbsO produced attenuated effects in the
DBD compared to lacO1, only lacO1 could cause the allosteric changes in the metastable helix,
70 Å from the DBD. These structural rearrangements, which are not observed in LacI incubated
with weaker (rbsO) or nonspecific (scrO) DNA binding motifs, promote stronger binding to
operator DNA as a demonstration of DNA sequence-induced protein allostery.
Two VFT-fold protein families respond differently to binding the same sugars
We reasoned that comparing ∆Gop for structurally aligned regions would highlight
similarities and differences in how binding sugars redistributes the energy in TFs vs. PBPs to
mediate distinct sense-response behaviors (Fig. 4) (17). Similarly, comparing ∆Gop for TFs and
PBPs that bind the same ligands can reveal how the same fold can be adapted for different
functional responses to the same ligand (Figs. 4, S11A-C). While sugar binding causes unique
local effects in each protein due to each sugar’s specific biochemical characteristics (Fig. 4C,
D), we also observed long-range pan-VFT and family-specific patterns (Figs. 4B, E, S10D).
The hinge loops of PBPs stabilize their sugar-bound conformational state (Fig. S12)
(13), whereas the TFs show no conformational changes in these loops in apo vs. sugar-bound
structures. Accordingly, although the PBPs and TFs bind the same sugars, they share few
conserved effects by HX/MS (Figs. 4, S11C). All hinge loops are stabilized in both PBPs; some
positions have ∆∆Gop, HOLO-APO>20 kJ/mol, on par with the strongest effects in the TFs that arise
from protein-sugar interactions (Fig. 4B, C).
Each TF is globally stabilized by inducer binding (Fig. 4D-F), particularly in the binding
pocket, pocket-periphery, and dimer interface. The N-terminal VFT dimer interface experiences
pervasive inducer-triggered allosteric structuring: part of the transducer helix is stabilized, and
strands E5, E1, and E9 unite to form a super-sheet with a hydrogen bond pattern distinct from
the apo and operator-bound states (Fig. S4-4, S13). The differences in how inducer and
operator binding stabilize this region suggest that TF induction includes a strand slipping
mechanism that favors an energetically similar yet configurationally distinct beta sheet with a
different strand register (Figs. S4-4, S11A, S13). Strand slippage is unlikely to occur
spontaneously because it requires a substantial energetic disruption to other local structural
elements, but was previously implicated in mediating conformational changes in a membrane
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
9
Figure 4. Ligand binding in PBPs and TFs. A, Color-coded secondary structures of the VFT domain. B, Conserved
energetic effects of sugar binding in the two PBPs. Green regions are stabilized by sugar binding in both PBPs. C,
Energetic redistribution in the PBPs upon sugar binding. The radar plots represent the ligand binding site. Each box at
the radar plot edge, colored by secondary structure as in A, corresponds to a different amino acid lining the binding
pocket. The spokes, colored by functional state, indicate the ∆Gop for these amino acids. For example, in MglB, residues
in loop C14 are stabilized by sugar binding (increased ∆Gop). This loop is colored green in A. D, Energetic redistribution
in the TFs upon sugar binding, colored as in C. E, Conserved energetic effects of inducer binding in the three TFs. F,
Molecular details of the TF binding pockets. E and F are colored by ∆∆Gop, HOLO-APO, as in B.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
10
protein (39). Helices 16 and 20 at the binding pocket periphery are also critical to inducer
binding, corroborating mutational studies of LacI (Fig. 4F) (16, 19). Finally, four positions are
destabilized in response to inducer binding in all three TFs to compensate for locally increased
helical stability.
To understand the baseline requirements for sugar binding in the VFT fold, we collated
conserved effects in all five proteins (Fig. S11D). All proteins are stabilized by sugar binding in
the binding pocket helices (∆∆Gop, HOLO-APO>10 kJ/mol). However, while there are many family-
specific energetic changes, we observed few pan-VFT effects, which underscores how the fold
is uniquely co-opted by each family.
Water molecules structure the induced states of LacI/GalR TFs
HX/MS shows that inducer binding stabilizes the transducer helix in the binding pocket of
each TF (Fig. 4D), but crystal structures suggest that this requires water-mediated hydrogen
bonds (Fig. S4-2) (15, 16). We were curious whether structural water molecules, defined as
waters that form three or more hydrogen bonds with protein or inducer atoms, are evolutionarily
conserved to maintain the induced TF ensemble (16). To explore this possibility, we used
RosettaECO (Explicit Consideration of cOordinated water) (40) to model structural waters in the
induced states of several sugar-inducible TFs for which experimental evidence of sugar-binding
and induction is available. We identified a high-confidence water molecule in the binding pocket
of sugar-binding LacI/GalR TFs near the dimer interface, bridging the sugar with the transducer
helix (Fig. S14). In support of the models, this water is found with very low b-factors in RbsR-
ribose (b=11.73) and LacI-IPTG (b=23.80) crystal structures (Fig. S4-2) (15). As an interesting
exception, in the trehalose repressor (TreR), the phosphoryl group of the inducer trehalose-6-
phosphate acts as the bridge (20) (Fig. S14C), whereas unphosphorylated trehalose is an anti-
inducer that stabilizes binding on the operator DNA (41). The LacI anti-inducer ONPF similarly
disrupts the water network established by IPTG (16).
Using the crystal structures of LacI-IPTG and RbsR-ribose and our model for GalR-
galactose, we performed atomistic MD simulations as an orthogonal method to explore how
structural waters contribute to the induced TF ensembles (Fig. 5B). We extracted the waters
within 3.5 Å of the sugars and performed a clustering analysis to find clusters with >80%
occupancy (Fig. S15). Echoing the RosettaECO results, the simulations identified the same
water near the transducer helix to have high occupancy in all TFs (Fig. 5C). The highest
duration/occupancy for this structural water was in LacI, due to two hydroxyl groups in IPTG
positioned towards the dimer interface, rather than only one in ribose; in GalR, it has near 100%
occupancy. The water does not appear in operator-bound TFs (Fig. S4-2) (27), and we expect
only transient waters in the binding pockets of apo TFs. These data support the hypothesis that
the sugar-inducible TFs evolved to maintain not just the VFT fold or functionally important
conformational changes, but also specific interactions with water that contribute to the induced
state ensembles. PBPs also make a similar water-mediated interaction with sugars (Fig S16),
suggesting that the pan-VFT-fold stabilization of the N-terminus of the transducer helix upon
sugar binding relies partially on a structured water molecule.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
11
Figure 5. Energetic changes in TFs vs. PBPs. A, Binding pocket waters in TF structures and models. B, Waters
sampled in MD simulations cluster together. C, Durations and occupancies for each water cluster in MD simulations.
Red dots correspond to circled structural waters in B. D, Model for induction in LacI/GalR TFs based on conserved
∆∆Gop, IND-DNA (left), with molecular switches labeled and schematized on the right. E, Conserved mechanism of sugar
response in both PBPs. F, Pan-VFT energetic relationships center on sugar interactions with binding pocket helices.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
12
Molecular switches toggle between TF functional states
PBPs respond to sugars with a hinge-mediated conformational change that is visible in
structures (Fig. S12). Sugar binding stabilizes them in a “closed” conformational ensemble that
they can already sample in the apo state, with locked inter-lobe loops forming a perfect binding
interface for sugar transporters. But in TFs, the operator- and sugar-bound states have
incompatible energetic requirements along the dimer and VFT-DBD interfaces, and the
transition between states is invisible in structural models. Ultimately, the TFs evolved an
energetic blueprint that is not a smaller-scale version of PBP response to sugars as we
originally hypothesized, but which is instead unique to the family (Figs. 5D, S17).
A set of “molecular switches” transition the LacI/GalR TFs to the induced state by
restructuring the hinge helix (35), but also, crucially, the VFT dimer interface, VFT cross-lobe
interactions, and the VFT-DBD interface (Figs. 5D, S17, Supplemental Note 2). Inducer- and
operator-driven conformational changes at switch positions have substantial ∆∆Gop,DNA-ind of 10-
20 kJ/mol. Inducer binding to the DNA-bound TFs rigidifies the transducer helix at its N-terminus
at the expense of the stability of the C-terminus, stabilizes inter-lobe interactions, and
destabilizes the metastable helix, like an “undo” button for operator binding. Simultaneously, the
hydrogen bond network in the N-terminal VFT beta sheet is rearranged, and several helices
adjacent to the DBD are restructured. In the VFT-DBD interface, inducer binding stabilizes H36
residues to reduce hydrophobic interactions with the loop that joins the hinge helix to the VFT
domain, thereby destabilizing the DBD. Simultaneously, reduced hydrogen bonding between H7
and the hinge helices releases them from the minor groove of the operator DNA, linearizing it. In
support of this mechanism, mutations in this region modulate LacI-operator DNA affinity (19,
42–44). Destabilization of the DBD is conserved in the induced state, resulting in decreased
∆Gop at recognition helix positions key to operator specificity (38). Additionally, nonspecific
charged interactions with DNA are destabilized, as are DBD helical packing interactions. While
sugars stabilize some of the binding pocket helices in both PBPs and TFs, few of these long-
range effects that define the energetic blueprint of TFs are manifested in PBPs (Fig. 5E, F).
Discussion
Coordinating switching between operator DNA- and specific sugar-bound states for
LacI/GalR TFs is a life-or-death necessity for bacteria. Here, we discovered how the distribution
of ensemble energies in the LacI/GalR TFs is dramatically different when bound to operators vs.
sugars, though their X-ray crystal structures are similar. We found specific residues in the TF
family that behave as molecular switches by occupying different, mutually exclusive energies in
different functional states. By the evolutionary conservation of these switches and their state-
specific interactions with protein, ligand, and water molecules, the TFs undergo a shift in their
conformational ensembles that forbids prolonged operator DNA binding simultaneously with
inducer binding. This is the invisible family secret that enables transcriptional regulation.
The VFT fold is not unique to LacI/GalR TFs. By comparing the TFs to their distant
cousins, the PBPs, we determined how one protein fold performs two different functions in
response to the same small molecule ligands. This yields a quantitative description of how
specific energetic relationships persisted and adapted over the evolutionary timescale to protect
functionally important conformational changes from being lost, even as new chemical triggers
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
13
challenged the remarkable plasticity of the fold. We can use this blueprint to rationally and
precisely engineer VFT-fold sensors, which has historically proven challenging for both TFs and
PBPs despite their promise in biotechnology (18, 45–48).
The story of conserved energetic relationships in a classic TF and its cousins
demonstrates how regulators of the genome have evolved alongside the genome itself, and that
the energetic blueprint for the family, buried in the evolutionary record, is key to designing
transcriptional regulators de novo. Previous efforts to design proteins that bind to DNA
sequences specifically and with high affinity are mired by the repetitive degeneracy of DNA
structures and a bias towards modeling linear DNA. In one recent effort to design DNA-binding
proteins, fourteen sequence-specific protein binders were identified from screening more than
140,000 design models, many of which were designed with an experimentally solved target
DNA structure available (49). Perhaps designing transcriptional regulators de novo requires
considering two-way allostery: how the TF changes the conformational ensemble of DNA, and
how different DNA sequences reshape the conformational ensemble of the TF.
Acknowledgements
We thank members of the Glasgow Lab for fruitful discussions, Dr. Wayne Hendrickson for
advice on X-ray crystallography, Dr. Naomi Latorraca for feedback on the manuscript, and Dr.
Rinat Abzalimov at the Advanced Science Research Center at the City University of New York
in his role as the MS Facility Manager. We thank the staff at NYX beamline at NSLS-II for their
support during data collection. We acknowledge support from the National Institutes of Health
(R00GM135529 to AG). KCW was funded by a National Science Foundation Graduate
Research Fellowship.
Data availability statement
The HX/MS deuterium uptake plots, FEATHER-derived centroid fits to HX/MS data, and a
PyMOL session file showing protein-specific differences in ∆Gop are available in the
Supplementary Materials. The HX/MS data have been deposited to the ProteomeXchange
Consortium via the PRIDE (50) partner repository with the dataset identifier PXD062370. The X-
ray crystal structures are uploaded to the PDB with codes 9NY7 and 9NY8. The AlphaFold 3-
RosettaECO models, MD trajectories, conservation score matrices, gene trees, MSA files, and
X-ray diffraction images are available in Zenodo (dataset identifier 10.5281/zenodo.15091462).
Methods
Calculation of conservation scores
Multiple sequence alignment (MSA). The MSA was generated using MMSeqs2 (52) separately
for TFs (LacI as the query) and PBPs (RbsB as the query) with bacterial sequences from
UniRef90 as the search database, using the following command line:
mmseqs filtertaxseqdb ./Uniref90_DB UniRef90_bacteria --taxon-list 2 -s 6 --max-seqs 20000 -a
1 -e 0.01 --min-aln-len 150 --allow-deletion 1 --threads 60.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
14
Conservation scores. We used the global epistatic model GEMME (53) which takes into account
both MSA and evolutionary history of the sequences using the MSA generated above. We used
the following command line with the provided docker:
docker run -ti --rm --mount type=bind,source=$PWD,target=/project elodielaine/gemme:gemme
then
python2.7 $GEMME_PATH/gemme.py MSA.fasta -r input -f MSA.fasta
The combined prediction scores were normalized within each group of proteins (TFs or PBPs)
such that the highest conservation score is 1.
Gene tree construction
For the input sequence alignment, we combined MSAs from the previous step and added the
sequence for LuxP (Uniprot P54300) to initially use as an outgroup (7). We used only the VFT
portion of the MSA and further excluded positions with >10% gaps. We used IQTree (54) for the
tree construction. We estimated a substitution model using the following command line:
iqtree -s ./MSA.fasta -m MFP -B 1000 -bnni -T AUTO .
We chose Q.pfam+R10 (55) as the best-fit model based on Bayesian information criterion (BIC)
(56) and ran the following command:
iqtree -s ./MSA.fasta -m Q.pfam+R10 -B 1000 -bnni -T AUTO -safe -cptime 600 .
Structural alignment of TFs
We collected experimentally solved structures of LacI/GalR TFs from the PDB (57) using the
following query:
( ( Lineage Name = "GalR/LacI-like bacterial regulator" AND Annotation Type = "SCOP" ) OR (
Lineage Name HAS ANY OF WORDS "LacI,GalR" AND Annotation Type = "ECOD" ) OR (
Lineage Identifier = "IPR046335" AND Annotation Type = "InterPro" ) ) AND Polymer Entity
Sequence Length > 90
The pairwise RMSD was calculated using US-align (58).
Plasmid construction
DNA sequences encoding RbsR, and GalR were synthesized as double-stranded genes by IDT
DNA and cloned into a pET9a vector with an N-terminal 6× histidine tag and TEV protease
sequence using a Golden Gate strategy with type IIs restriction enzyme BsaI (59). In the case of
LacI, we used a previously described plasmid with the same architecture (16, 17). The MglB
and RbsB expression plasmids were also constructed with N-terminal 6× histidine tag and TEV
site using the same strategy in the same vector, but the genes were amplified by polymerase
chain reaction from the E. coli DH10B genome. All genetic constructs were confirmed by
sequencing. All gene sequences are available in Table S15.
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
15
Protein expression and purification
Proteins were purified as described previously (17), with the exception that GalR buffers for cell
lysis, purification, and storage included 600 mM NaCl to avoid protein aggregation (60).
X-ray crystallography sample preparation
To prepare the rbsO-RbsR sample, self-complementary single-stranded 30-base-pair
primers encoding rbsO (5′-GTGGGTCAGCGAAACGTTTCGCTGATGGAG-3′) were synthesized
(IDT DNA) and annealed using a previously described protocol (16). RbsR dimer (10 mg/mL)
was then incubated at 4 °C for 12 h with a two-fold molar excess of the annealed DNA. After
incubation, the mixture was centrifuged (14,000 rpm, 4 °C, 10 min), and crystals were grown by
hanging-drop vapor diffusion using 400 μL of the protein–DNA complex solution and 300 μL of
reservoir solution containing 0.2 M sodium sulfate, 0.1 M bis-tris-propane (pH 6.5), and 20%
PEG 3350 (PACT Premier 2-20) at 20 °C.
To prepare the ribose-RbsR sample, RbsR (10.1 mg/mL) was incubated with a six-fold
molar excess of ribose (1.3 mM) for 12 h at 4 °C and subsequently centrifuged (14,000 rpm, 4
°C, 10 min). The resulting mixture was crystallized by hanging-drop vapor diffusion using 400 μL
of the protein–ribose complex solution and 300 μL of reservoir solution containing 0.2 M sodium
malonate (pH 4.0) and 20% PEG 3350 (Pegion2 A5) at 20 °C.
X-ray crystallography data collection and model building
Diffraction data were collected at the NYX beamline 19-ID of the National Synchrotron Light
Source II (NSLS-II) at Brookhaven National Laboratory. The raw diffraction HDF5 images were
processed using Xia2 (61) with DIALS (62) pipeline, as implemented in CCP4 suite (63), for
ribose-bound RbsR structure and XDS (64) for rbsO-bound structure to perform indexing,
integration and initial scaling. Data scaling and merging were carried out using AIMLESS or
STARANISO (65). Molecular replacement was performed with Phaser (66), using an AlphaFold
2 model (67) as the search model for the ribose-bound RbsR structure, and an AlphaFold 3
model (36) as the input model for the rbsO-bound structure. Coot (68) was used for manual
modelling. Structural refinement was conducted using both Phenix.refine (69) and REFMAC
(70). Final validation and refinement were completed for both using PDB-Redo (71).
HX/MS sample preparation and experiment
Sample preparation was performed as described previously (17), except for the following
variables: sample buffer composition, protein stock composition, and quench composition, as
specified in Tables S3-S7. All proteins were purified by size exclusion chromatography,
confirmed to be >95% pure by SDS-PAGE, concentrated to 5-10 µM, and flash-frozen in
aliquots. HX/MS experiments were performed by diluting freshly purified proteins or thawed
aliquots by tenfold in an equivalent Tris-NaCl D2O buffer at pH 8.0. IPTG, galactose, and ribose
were added to protein samples at 10 mM. Operator DNAs were added at twofold molar excess
of TF dimer concentrations. HX experiments were performed as described previously for LacI
(17) except for the following variables: reaction and injection volumes, and protease columns,
specified in Tables S3-S7. After incubation in D2O buffer, the exchange reaction was quenched
by 1.9-fold sample dilution in quench buffer (3% acetonitrile, 1% formic acid in H2O, with GuHCl
concentrations noted in Tables S3-S7 for all proteins, all MS-grade, Fisher Scientific).
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
16
HX/MS fully deuterated sample preparation
Fully deuterated controls were prepared as described previously (17) (Tables S3-S7). Of note,
due to some observed aggregation, we removed GuHCl from the preparation of fully deuterated
control samples for GalR samples. We allowed them to exchange at 37 ºC for at least 24 hours
and used a quench buffer containing GuHCl as in the other HX experiments.
The protein stock solutions were diluted ten-fold in deuterated buffer and exchanged at 15 ºC
for the specified time, then diluted in quench solution (75 µl, 3 M GuHCl, 3% acetonitrile, 1%
formic acid in H2O, all MS-grade, Fisher Scientific) of equal volume to the deuterated buffer at 2
ºC. All but 10 µl of quenched reaction was injected into the protease column at 7 ºC. Solvent A
(3% acetonitrile, 0.15% formic acid in H2O) was pumped (UltiMate 3000, ThermoFisher
Scientific) at 150 μL/min through the protease column. Peptides eluted from the protease
column were trapped on a 1x10 mm C18 column (Hypersil GOLD, 3 μM pore size,
ThermoFisher Scientific).
HX/MS data analysis
MS/MS peak picking and deconvolution, disambiguation by peak picking, peptide
disambiguation by PIGEON, FEATHER rate fitting, PIGEON error analysis, and quality control
of all fit spectra were performed as described previously for LacI (17). Conservation among
FEATHER-derived ∆Gop was determined by treating as nonzero only state-state differences that
exceeded our significance threshold and occurred in the same direction for all proteins in the
comparison set. We calculated the mean of these positional ∆∆Gop, wherever mean{∆∆Gop} > 2
kJ/mol. For a given protein, state-state differences at each site were considered significant if the
“mini-peptide” (the set of residues including that site covered indistinguishably by a set of fit
exchange rates) was the same for both states and satisfied mean{∆Gop,a,i-∆Gop,b,i} > std{∆Gop,a,i-
∆Gop,b,i}, where the {∆Gop,a,i} are the sorted ∆Gop corresponding to that site for each exchange
rate in the mini-peptide, in state a. In all structural models presented in this work that are
colored by ∆∆Gop conserved among 2+ proteins, we calculated the average positional ∆∆Gop for
those proteins, if the selection criteria described above were met. Otherwise, we colored those
positions white to indicate that the exchange behavior was not conserved. Gray coloring
indicates a lack of data for at least one protein or state, or high standard deviation (noise) in the
significance calculation. The full sets of ∆Gop, their standard deviations, and their resolutions are
available for all proteins in Tables S8-S12.
Structural water prediction using RosettaECO
The experimental structures or AlphaFold 3 (AF3) (36) models were first relaxed using the
Rosetta biomolecular modeling suite (72). The experimental structures were relaxed with
backbone and sidechain constraints using the following command line to produce three models
for each input file:
./relax.default.linuxgccrelease -s input.pdb -out:suffix .constrain.relax -nstruct 3 -
relax:default_repeats 5 -out:path:pdb ./ -out:path:score ./ -in:file:extra_res_fa ./tre.params -
constrain_relax_to_start_coords true -relax:coord_constrain_sidechains true
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
17
Three models were produced for each input protein. The structural model with lowest score in
Rosetta Energy Units (REU) was chosen as the best model. For the AF3 models, we performed
unconstrained relaxation using the command:
./relax.default.linuxgccrelease -s input.pdb -out:suffix .relax -nstruct 3 -relax:default_repeats 5 -
out:path:pdb ./ -out:path:score ./ -in:file:extra_res_fa ./param_fle.params
The ligand was specified through the “param_file.params” file where needed. To solvate the
models using Rosetta Explicit Consideration of cOordinated water (RosettaECO) (40), we used
an .xml protocol (see Data availability) to generate 50 solvated models per input unsolvated
relaxed model:
./rosetta_scripts.default.linuxgccrelease -parser:protocol solvate.xml -s input.pdb -nstruct 50 -
beta_nov16 -out:suffix .solvate -out:path:pdb ./ -in:file:extra_res_fa ./param_file.params
Water molecules within 0.5 Å of each other were clustered and reported in the final solvated
model (see Data availability; the number of waters in each model is reported as “b-factor” in the
final .pdb file).
Molecular dynamics simulations
Five sugar-bound VFT-fold systems (LacI-IPTG, GalR-galactose, RbsR-ribose, MglB-galactose,
and RbsB-ribose) were built for MD simulations using the following structures by PDB ID: 2P9H,
AF3 model, 9NY7, 1GLG, and 2DRI as the input models, respectively. The CHARMM36 force
field was applied and the simulations were performed in the GROMACS package (version
2022). We energy minimized the systems using the steepest descent minimization method until
the maximum force was less than 1000 kJ/mol/nm2, followed by 100 ps restrained MD
simulations in NVT ensemble by constraining the heavy atoms of the protein and ligand at
298.15 K. The resulting simulation system was further equilibrated for a 1000 ps NPT simulation
at 1 atm and 298.15 K with a 1000 kJ/mol/nm2 position restraints on the protein and ligand.
Finally, we conducted the NPT production MD simulations using the same conditions without
restraints, with a time step of 2 fs, nonbonded cutoff of 12 Å, and particle-mesh Ewald long-
range electrostatics. Five 300 ns simulation replicates were collected for each system. The MD
trajectories were analyzed using MDAnalysis (73), VMD (74), or MDTraj (75).
Water clustering analysis
The trajectories from the production runs for each system were aligned to its initial structure
based on the protein backbone atoms, and the coordinates of water molecules located within
3.5 Å of the ligands were extracted. These coordinates were subsequently clustered using the
k-means algorithm implemented in the scikit-learn (76) Python package. The number of the
clusters was determined by the elbow method.
References
1. Z. Lin, H. Akin, R. Rao, B. Hie, Z. Zhu, W. Lu, N. Smetanin, R. Verkuil, O. Kabeli, Y.
Shmueli, A. dos Santos Costa, M. Fazel-Zarandi, T. Sercu, S. Candido, A. Rives,
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
18
Evolutionary-scale prediction of atomic-level protein structure with a language model.
Science 379, 1123–1130 (2023).
2. C. Chothia, One thousand families for the molecular biologist. Nature 357, 543–544
(1992).
3. H. Li, C. Tang, N. S. Wingreen, Are protein folds atypical? Proceedings of the National
Academy of Sciences 95, 4987–4990 (1998).
4. P. Cossio, A. Trovato, F. Pietrucci, F. Seno, A. Maritan, A. Laio, Exploring the Universe of
Protein Structures beyond the Protein Data Bank. PLOS Computational Biology 6,
e1000957 (2010).
5. B. Chitturi, S. Shi, L. N. Kinch, N. V. Grishin, Compact Structure Patterns in Proteins.
Journal of Molecular Biology 428, 4392–4412 (2016).
6. S. Minami, N. Kobayashi, T. Sugiki, T. Nagashima, T. Fujiwara, R. Tatsumi-Koga, G.
Chikenji, N. Koga, Exploration of novel αβ-protein folds through de novo design. Nature
Structural and Molecular Biology 30, 1132–1140 (2023).
7. K. Fukami-Kobayashi, Y. Tateno, K. Nishikawa, Parallel evolution of ligand specificity
between LacI/GalR family repressors and periplasmic sugar-binding proteins. Molecular
Biology and Evolution 20, 267–277 (2003).
8. J. A. Kaczmarski, M. C. Mahawaththa, A. Feintuch, B. E. Clifton, L. A. Adams, D. Goldfarb,
G. Otting, C. J. Jackson, Altered conformational sampling along an evolutionary trajectory
changes the catalytic activity of an enzyme. Nature Communications 11, 5945 (2020).
9. K. R. Torgeson, M. W. Clarkson, D. Granata, K. Lindorff-Larsen, R. Page, W. Peti,
Conserved conformational dynamics determine enzyme activity. Science Advances 8,
eabo5546 (2022).
10. J. Cao, S. Huang, J. Qian, J. Huang, L. Jin, Z. Su, J. Yang, J. Liu, Evolution of the class C
GPCR Venus flytrap modules involved positive selected functional divergence. BMC
Evolutionary Biology 9, 67 (2009).
11. L. Swint-Kruse, K. S. Matthews, Allostery in the LacI/GalR family: variations on a theme.
Current Opinion in Microbiology 12, 129–137 (2009).
12. B. H. Shilton, M. M. Flocco, M. Nilsson, S. L. Mowbray, Conformational Changes of Three
Periplasmic Receptors for Bacterial Chemotaxis and Transport: The Maltose-,
Glucose/Galactose- and Ribose-binding Proteins. Journal of Molecular Biology 264, 350–
363 (1996).
13. S. L. Mowbray, A. J. Björkman, Conformational changes of ribose-binding protein and two
related repressors are tailored to fit the functional need. Journal of Molecular Biology 294,
487–499 (1999).
14. D. J. Parente, L. Swint-Kruse, Multiple Co-Evolutionary Networks Are Supported by the
Common Tertiary Scaffold of the LacI/GalR Proteins. PLOS ONE 8, e84398 (2013).
15. R. Daber, S. Stayrook, A. Rosenberg, M. Lewis, Structural Analysis of Lac Repressor
Bound to Allosteric Effectors. Journal of Molecular Biology 370, 609–619 (2007).
16. A. Glasgow, H. T. Hobbs, Z. R. Perry, M. L. Wells, S. Marqusee, T. Kortemme, Ligand-
specific changes in conformational flexibility mediate long-range allostery in the lac
repressor. Nature Communications 14, 1179 (2023).
17. C. Lu, M. L. Wells, A. Reckers, A. Glasgow, Site-resolved energetic information from
HX/MS experiments. bioRxiv [Preprint] (2024). https://doi.org/10.1101/2024.08.04.606547.
18. P. Kröger, S. Shanmugaratnam, N. Ferruz, K. Schweimer, B. Höcker, A comprehensive
binding study illustrates ligand recognition in the periplasmic binding protein PotF.
Structure 29, 433-443.e4 (2021).
19. P. Markiewicz, L. G. Kleina, C. Cruz, S. Ehret, J. H. Miller, Genetic studies of the lac
repressor. XIV. Analysis of 4000 altered Escherichia coli lac repressors reveals essential
and non-essential residues, as well as “spacers” which do not require a specific sequence.
Journal of Molecular Biology 240, 421–433 (1994).
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
19
20. U. Hars, R. Horlacher, W. Boos, W. Welte, K. Diederichs, Crystal structure of the effector-
binding domain of the trehalose-repressor of Escherichia coli, a member of the LacI family,
in its complexes with inducer trehalose-6-phosphate and noninducer trehalose. Protein
Science 7, 2511–2521 (1998).
21. M. A. Schumacher, A. Glasfeld, H. Zalkin, R. G. Brennan, The X-ray Structure of the PurR-
Guanine-purF Operator Complex Reveals the Contributions of Complementary
Electrostatic Surfaces and a Water-mediated Hydrogen Bond to Corepressor Specificity
and Binding Affinity. Journal of Biological Chemistry 272, 22648–22653 (1997).
22. M. A. Schumacher, G. S. Allen, M. Diel, G. Seidel, W. Hillen, R. G. Brennan, Structural
basis for allosteric control of the transcription regulator CcpA by the phosphoprotein HPr-
Ser46-P. Cell 118, 731–741 (2004).
23. M. L. Oldham, D. Khare, F. A. Quiocho, A. L. Davidson, J. Chen, Crystal structure of a
catalytic intermediate of the maltose transporter. Nature 450, 515–521 (2007).
24. J. J. Birktoft, D. M. Blow, Structure of crystalline α-chymotrypsin: V. The atomic structure of
tosyl-α-chymotrypsin at 2 Å resolution. Journal of Molecular Biology 68, 187–240 (1972).
25. J. A. Endicott, M. E. M. Noble, L. N. Johnson, The Structural Basis for Control of
Eukaryotic Protein Kinases. Annual Review of Biochemistry 81, 587–613 (2012).
26. A. M. Friedman, T. O. Fischmann, T. A. Steitz, Crystal Structure of lac Repressor Core
Tetramer and Its Implications for DNA Looping. Science 268, 1721–1727 (1995).
27. M. Lewis, G. Chang, N. C. Horton, M. A. Kercher, H. C. Pace, M. A. Schumacher, R. G.
Brennan, P. Lu, Crystal Structure of the Lactose Operon Repressor and Its Complexes
with DNA and Inducer. Science 271, 1247–1254 (1996).
28. C. E. Bell, M. Lewis, A closer view of the conformation of the Lac repressor bound to
operator. Nature Structural Biology 7, 209–214 (2000).
29. C. a. Mauzy, M. a. Hermodson, Structural and functional analyses of the repressor, RbsR,
of the ribose operon of Escherichia coli. Protein Science 1, 831–842 (1992).
30. V. J. Hilser, An Ensemble View of Allostery. Science 327, 653–654 (2010).
31. B. von Wilcken-Bergmann, B. Müller-Hill, Sequence of galR gene indicates a common
evolutionary origin of lac and gal repressor in Escherichia coli. Proceedings of the National
Academy of Sciences 79, 2427–2431 (1982).
32. A. Majumdar, S. Adhya, Probing the structure of gal operator-repressor complexes.
Conformation change in DNA. Journal of Biological Chemistry 262, 13258–13262 (1987).
33. Y. Bai, T. R. Sosnick, L. Mayne, S. W. Englander, Protein Folding Intermediates: Native-
State Hydrogen Exchange. Science 269, 192–197 (1995).
34. C. G. Kalodimos, N. Biris, A. M. J. J. Bonvin, M. M. Levandoski, M. Guennuegues, R.
Boelens, R. Kaptein, Structure and Flexibility Adaptation in Nonspecific and Specific
Protein-DNA Complexes. Science 305, 386–389 (2004).
35. C. A. E. M. Spronk, M. Slijper, J. H. van Boom, R. Kaptein, R. Boelens, Formation of the
hinge helix in the lac represser is induced upon binding to the lac operator. Nature
Structural and Molecular Biology 3, 916–919 (1996).
36. J. Abramson, J. Adler, J. Dunger, R. Evans, T. Green, A. Pritzel, O. Ronneberger, L.
Willmore, A. J. Ballard, J. Bambrick, S. W. Bodenstein, D. A. Evans, C.-C. Hung, M.
O’Neill, D. Reiman, K. Tunyasuvunakool, Z. Wu, A. Žemgulytė, E. Arvaniti, C. Beattie, O.
Bertolli, A. Bridgland, A. Cherepanov, M. Congreve, A. I. Cowen-Rivers, A. Cowie, M.
Figurnov, F. B. Fuchs, H. Gladman, R. Jain, Y. A. Khan, C. M. R. Low, K. Perlin, A.
Potapenko, P. Savy, S. Singh, A. Stecula, A. Thillaisundaram, C. Tong, S. Yakneen, E. D.
Zhong, M. Zielinski, A. Žídek, V. Bapst, P. Kohli, M. Jaderberg, D. Hassabis, J. M. Jumper,
Accurate structure prediction of biomolecular interactions with AlphaFold 3. Nature 630,
493–500 (2024).
37. A. Schmitz, U. Schmeissner, J. H. Miller, Mutations affecting the quaternary structure of
the lac repressor. Journal of Biological Chemistry 251, 3359–3366 (1976).
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
20
38. C. A. Spronk, A. M. Bonvin, P. K. Radha, G. Melacini, R. Boelens, R. Kaptein, The solution
structure of Lac repressor headpiece 62 complexed to a symmetrical lac operator.
Structure 7, 1483–1492 (1999).
39. K. Makabe, S. Yan, V. Tereshko, G. Gawlak, S. Koide, β-Strand Flipping and Slipping
Triggered by Turn Replacement Reveal the Opportunistic Nature of β-Strand Pairing.
Journal of the American Chemical Society 129, 14661–14669 (2007).
40. R. E. Pavlovicz, H. Park, F. DiMaio, Efficient consideration of coordinated water molecules
improves computational protein-protein and protein-ligand docking discrimination. PLOS
Computational Biology 16, e1008103 (2020).
41. R. Horlacher, W. Boos, Characterization of TreR, the major regulator of the Escherichia
coli trehalose system. Journal of Biological Chemistry 272, 13026–13032 (1997).
42. C. M. Falcon, K. S. Matthews, Glycine Insertion in the Hinge Region of Lactose Repressor
Protein Alters DNA Binding*. Journal of Biological Chemistry 274, 30849–30857 (1999).
43. L. G. Kleina, J. H. Miller, Genetic studies of the lac repressor. Journal of Molecular Biology
212, 295–318 (1990).
44. P. Campitelli, L. Swint-Kruse, S. B. Ozkan, Substitutions at Nonconserved Rheostat
Positions Modulate Function by Rewiring Long-Range, Dynamic Interactions. Molecular
Biology and Evolution 38, 201–214 (2021).
45. M. A. Dwyer, H. W. Hellinga, Periplasmic binding proteins: a versatile superfamily for
protein engineering. Current Opinion in Structural Biology 14, 495–504 (2004).
46. N. D. Taylor, A. S. Garruss, R. Moretti, S. Chan, M. A. Arbing, D. Cascio, J. K. Rogers, F.
J. Isaacs, S. Kosuri, D. Baker, S. Fields, G. M. Church, S. Raman, Engineering an
allosteric transcription factor to respond to new ligands. Nature Methods 13, 177–183
(2016).
47. S. Tungtur, S. M. Egan, L. Swint-Kruse, Functional consequences of exchanging domains
between LacI and PurR are mediated by the intervening linker sequence. Proteins:
Structure, Function, and Bioinformatics 68, 375–388 (2007).
48. D. H. Richards, S. Meyer, C. J. Wilson, Fourteen Ways to Reroute Cooperative
Communication in the Lactose Repressor: Engineering Regulatory Proteins with Alternate
Repressive Functions. ACS Synthetic Biology 6, 6–12 (2017).
49. C. J. Glasscock, R. Pecoraro, R. McHugh, L. A. Doyle, W. Chen, O. Boivin, B. Lonnquist,
E. Na, Y. Politanska, H. K. Haddox, D. Cox, C. Norn, B. Coventry, I. Goreshnik, D.
Vafeados, G. R. Lee, R. Gordan, B. L. Stoddard, F. DiMaio, D. Baker, Computational
design of sequence-specific DNA-binding proteins. bioRxiv [Preprint] (2023).
https://doi.org/10.1101/2023.09.20.558720.
50. Y. Perez-Riverol, J. Bai, C. Bandla, D. García-Seisdedos, S. Hewapathirana, S.
Kamatchinathan, D. J. Kundu, A. Prakash, A. Frericks-Zipper, M. Eisenacher, M. Walzer,
S. Wang, A. Brazma, J. A. Vizcaíno, The PRIDE database resources in 2022: a hub for
mass spectrometry-based proteomics evidences. Nucleic Acids Research 50, D543–D552
(2022).
51. Y. Bai, J. S. Milne, L. Mayne, S. W. Englander, Primary structure effects on peptide group
hydrogen exchange. Proteins: Structure, Function, and Bioinformatics 17, 75–86 (1993).
52. M. Steinegger, J. Söding, MMseqs2 enables sensitive protein sequence searching for the
analysis of massive data sets. Nature Biotechnology 35, 1026–1028 (2017).
53. E. Laine, Y. Karami, A. Carbone, GEMME: A Simple and Fast Global Epistatic Model
Predicting Mutational Effects. Molecular Biology and Evolution 36, 2604–2619 (2019).
54. B. Q. Minh, H. A. Schmidt, O. Chernomor, D. Schrempf, M. D. Woodhams, A. von
Haeseler, R. Lanfear, IQ-TREE 2: New Models and Efficient Methods for Phylogenetic
Inference in the Genomic Era. Molecular Biology and Evolution 37, 1530–1534 (2020).
55. B. Q. Minh, C. C. Dang, L. S. Vinh, R. Lanfear, QMaker: Fast and Accurate Method to
Estimate Empirical Models of Protein Evolution. Systematic Biology 70, 1046–1060 (2021).
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
21
56. G. Schwarz, Estimating the Dimension of a Model. The Annals of Statistics 6, 461–464
(1978).
57. H. M. Berman, J. Westbrook, Z. Feng, G. Gilliland, T. N. Bhat, H. Weissig, I. N. Shindyalov,
P. E. Bourne, The Protein Data Bank. Nucleic Acids Research 28, 235–242 (2000).
58. C. Zhang, M. Shine, A. M. Pyle, Y. Zhang, US-align: universal structure alignments of
proteins, nucleic acids, and macromolecular complexes. Nature Methods 19, 1109–1115
(2022).
59. C. Engler, R. Gruetzner, R. Kandzia, S. Marillonnet, Golden Gate Shuffling: A One-Pot
DNA Shuffling Method Based on Type IIs Restriction Enzymes. PLoS ONE 4, e5553
(2009).
60. E. D. Agerschou, G. Christiansen, N. P. Schafer, D. J. Madsen, D. E. Brodersen, S.
Semsey, D. E. Otzen, The transcriptional regulator GalR self-assembles to form highly
regular tubular structures. Scientific Reports 6, 27672 (2016).
61. G. Winter, xia2: an expert system for macromolecular crystallography data reduction.
Journal of Applied Crystallography 43, 186–190 (2010).
62. G. Winter, D. G. Waterman, J. M. Parkhurst, A. S. Brewster, R. J. Gildea, M. Gerstel, L.
Fuentes-Montero, M. Vollmar, T. Michels-Clark, I. D. Young, N. K. Sauter, G. Evans,
DIALS: implementation and evaluation of a new integration package. Acta
Crystallographica Section D 74, 85–97 (2018).
63. J. Agirre, M. Atanasova, H. Bagdonas, C. B. Ballard, A. Baslé, J. Beilsten-Edmands, R. J.
Borges, D. G. Brown, J. J. Burgos-Mármol, J. M. Berrisford, P. S. Bond, I. Caballero, L.
Catapano, G. Chojnowski, A. G. Cook, K. D. Cowtan, T. I. Croll, J. É. Debreczeni, N. E.
Devenish, E. J. Dodson, T. R. Drevon, P. Emsley, G. Evans, P. R. Evans, M. Fando, J.
Foadi, L. Fuentes-Montero, E. F. Garman, M. Gerstel, R. J. Gildea, K. Hatti, M. L.
Hekkelman, P. Heuser, S. W. Hoh, M. A. Hough, H. T. Jenkins, E. Jiménez, R. P. Joosten,
R. M. Keegan, N. Keep, E. B. Krissinel, P. Kolenko, O. Kovalevskiy, V. S. Lamzin, D. M.
Lawson, A. A. Lebedev, A. G. W. Leslie, B. Lohkamp, F. Long, M. Malý, A. J. McCoy, S. J.
McNicholas, A. Medina, C. Millán, J. W. Murray, G. N. Murshudov, R. A. Nicholls, M. E. M.
Noble, R. Oeffner, N. S. Pannu, J. M. Parkhurst, N. Pearce, J. Pereira, A. Perrakis, H. R.
Powell, R. J. Read, D. J. Rigden, W. Rochira, M. Sammito, F. Sánchez Rodríguez, G. M.
Sheldrick, K. L. Shelley, F. Simkovic, A. J. Simpkin, P. Skubak, E. Sobolev, R. A. Steiner,
K. Stevenson, I. Tews, J. M. H. Thomas, A. Thorn, J. T. Valls, V. Uski, I. Usón, A. Vagin, S.
Velankar, M. Vollmar, H. Walden, D. Waterman, K. S. Wilson, M. D. Winn, G. Winter, M.
Wojdyr, K. Yamashita, The CCP4 suite: integrative software for macromolecular
crystallography. Acta Crystallographica Section D 79, 449–461 (2023).
64. W. Kabsch, XDS. Acta Crystallographica Section D 66, 125–132 (2010).
65. STARANISO anisotropy & Bayesian estimation server.
https://staraniso.globalphasing.org/cgi-bin/staraniso.cgi.
66. A. J. McCoy, R. W. Grosse-Kunstleve, P. D. Adams, M. D. Winn, L. C. Storoni, R. J. Read,
Phaser crystallographic software. Journal of Applied Crystallography 40, 658–674 (2007).
67. J. Jumper, R. Evans, A. Pritzel, T. Green, M. Figurnov, O. Ronneberger, K.
Tunyasuvunakool, R. Bates, A. Žídek, A. Potapenko, A. Bridgland, C. Meyer, S. A. A. Kohl,
A. J. Ballard, A. Cowie, B. Romera-Paredes, S. Nikolov, R. Jain, J. Adler, T. Back, S.
Petersen, D. Reiman, E. Clancy, M. Zielinski, M. Steinegger, M. Pacholska, T.
Berghammer, S. Bodenstein, D. Silver, O. Vinyals, A. W. Senior, K. Kavukcuoglu, P. Kohli,
D. Hassabis, Highly accurate protein structure prediction with AlphaFold. Nature 596, 583–
589 (2021).
68. P. Emsley, K. Cowtan, Coot: model-building tools for molecular graphics. Acta
Crystallographica Section D 60, 2126–2132 (2004).
69. P. V. Afonine, R. W. Grosse-Kunstleve, N. Echols, J. J. Headd, N. W. Moriarty, M.
Mustyakimov, T. C. Terwilliger, A. Urzhumtsev, P. H. Zwart, P. D. Adams, Towards
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
22
automated crystallographic structure refinement with phenix.refine. Acta Crystallographica
Section D 68, 352–367 (2012).
70. P. Skubák, A. A. Lebedev, N. S. Pannu, R. A. Steiner, R. A. Nicholls, M. D. Winn, F. Long,
A. A. Vagin, REFMAC5 for the refinement of macromolecular crystal structures. Acta
Crystallographica Section D 67 (2011).
71. R. P. Joosten, F. Long, G. N. Murshudov, A. Perrakis, The PDB_REDO server for
macromolecular structure model optimization. IUCrJ 1, 213–220 (2014).
72. R. F. Alford, A. Leaver-Fay, J. R. Jeliazkov, M. J. O’Meara, F. P. DiMaio, H. Park, M. V.
Shapovalov, P. D. Renfrew, V. K. Mulligan, K. Kappel, J. W. Labonte, M. S. Pacella, R.
Bonneau, P. Bradley, R. L. Dunbrack, R. Das, D. Baker, B. Kuhlman, T. Kortemme, J. J.
Gray, The Rosetta All-Atom Energy Function for Macromolecular Modeling and Design.
Journal of Chemical Theory and Computation 13, 3031–3048 (2017).
73. N. Michaud-Agrawal, E. J. Denning, T. B. Woolf, O. Beckstein, MDAnalysis: A toolkit for
the analysis of molecular dynamics simulations. Journal of Computational Chemistry 32,
2319–2327 (2011).
74. W. Humphrey, A. Dalke, K. Schulten, VMD: Visual molecular dynamics. Journal of
Molecular Graphics 14, 33–38 (1996).
75. R. T. McGibbon, K. A. Beauchamp, M. P. Harrigan, C. Klein, J. M. Swails, C. X.
Hernández, C. R. Schwantes, L.-P. Wang, T. J. Lane, V. S. Pande, MDTraj: A Modern
Open Library for the Analysis of Molecular Dynamics Trajectories. Biophysical Journal 109,
1528–1532 (2015).
76. F. Pedregosa, G. Varoquaux, A. Gramfort, V. Michel, B. Thirion, O. Grisel, M. Blondel, P.
Prettenhofer, R. Weiss, V. Dubourg, J. Vanderplas, A. Passos, D. Cournapeau, M.
Brucher, M. Perrot, É. Duchesnay, Scikit-learn: Machine Learning in Python. Journal of
Machine Learning Research 12, 2825–2830 (2011).
.CC-BY-NC 4.0 International licenseavailable under a
was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made
The copyright holder for this preprint (whichthis version posted April 7, 2025. ; https://doi.org/10.1101/2025.04.02.646877doi: bioRxiv preprint
