Science - USA (2020-09-04)

1166 4 SEPTEMBER 2020 • VOL 369 ISSUE 6508 sciencemag.org SCIENCE

INSIGHTS | PERSPECTIVES

PROTEIN DESIGN

Can proteins be truly designed sans function?

A new unit of local protein structure can aid in the de novo design of ligand-binding proteins

By Anna Peacock

P

roteins come in a wide range of sizes,

shapes, and folds and perform a broad

range of functions. Investigations

into how and why proteins fold from

peptide sequences to yield a par-

ticular structure have continued for

decades ( 1 ) and have inspired efforts to de-

sign proteins de novo—that is, to rationally

design structured miniature protein folds

from first principles. Ultimately, the goal

is not merely to design specific folds but to

create proteins that execute functions—ide-

ally, functions beyond the repertoire found

in nature. When the desired function in-

volves binding of small molecules, as is the

case in many applications, this requirement

adds an additional level of complexity and

challenge. On page 1227 of this issue, Polizzi

and DeGrado ( 2 ) have developed a search

algorithm, Convergent Motifs for Binding

Sites (COMBS), by which ligand-selective

binding proteins can be designed truly de

novo, thereby providing a much-needed tool

for advancing functional protein design.

The de novo design of a truly artificial

protein fold was first reported for TOP7, a

93–amino acid mixed a/b-fold ( 3 ). Since

then, the de novo design of protein struc-

ture has made impressive progress through

enhanced understanding, expansion of

the experimentally validated structures in

the Protein Data Bank, greater computing

power, and affordable access to synthetic

genes that allow for greater experimental

validation. Artificial peptides have also

been designed to assemble into previ-

ously unknown architectures, including

a-helical barrels featuring accessible chan-

nels through their core ( 4 ), large spherical

cages ( 5 ), and polyhedral shapes ( 6 ). More

recently, the development of protein fold-

ing and design online computer games has

even allowed citizen scientists to design

previously unexplored protein folds ( 7 ).

Protein function can be achieved by

the actual structure, such as collagen,

but more often by the binding of complex

small-molecule ligands generating sen-

sors, receptors, or catalysts. A cavity must

be designed with a size and shape to serve

as a complementary “lock” for the target

ligand “key” that aligns the target for fa-

vorable interactions but does not interfere

with proper folding. The common strategy

of designing a structure and then introduc-

ing a function often comes at the expense

of proper structural folding or stability and

is often not successful.

A more appealing approach would be to

simultaneously design structure and func-

tion de novo. Previously reported strategies

tend to position the target ligand relative

to the interacting atoms of the amino acid

side chain ( 8 , 9 ). This approach can gener-

ate a large number of structures to evalu-

ate computationally, and can also lead to

combinations of coordinates and ligand

rotamers that are not actually observed ex-

perimentally. Generally, approaches to date

do not initially achieve strong binding, and

subsequent rounds of experimental valida-

tion and redesign are often required.

Polizzi and DeGrado designed an artifi-

cial protein fold that shares no sequence

homology to native proteins and also has

a binding site selective for a complex small

molecule—in this case, the blood-thinning

drug apixaban (see the figure). They de-

veloped a new unit of protein structure,

which they call a van der Mer (vdM), that

directly maps ligand chemical group func-

tionality (such as carbonyl, carboxylate,

carboxamide, and amine) to peptide resi-

due backbone coordinates. Crucially, vdMs

are generated from close contact with the

side chain, the main chain, or both. Ligand

chemical-group locations relate to back-

bone coordinates and not side chains, so

vdMs link directly to the protein fold. The

vdMs were extracted from the experimen-

tally determined structures deposited in

the Protein Data Bank, and the incidence

of the various vdMs across the Protein Data

Bank was used to score them. Surprisingly,

only a modest number of vdMs are highly

prevalent, making it computationally at-

tractive to adopt this approach.

Binding sites were engineered into de-

signed four-helix bundle folds that are

structurally unrelated to factor Xa, for

which apixaban is an inhibitor ( 10 ), and

that do not generally bind to small mol-

ecules. The authors searched for combina-

tions of vdMs (favoring those with higher

scores) that could be superimposed onto

protein backbone templates and that

presented chemical groups that could be

School of Chemistry, University of Birmingham,

Birmingham, UK. Email: [email protected]

Functional groups on the ligand are identifed

and a target peptide fold selected.

Selecting targets

Flexible backbone design of the

remainder of the sequence

Ab initio folding of the sequence to

screen for promising designs

C=O

+

C=O

CONH 2

Apo protein

structure

Ligand-bound

x-ray structure

Ligand

Gln/CONH 2

Thr/C=O vdM

His/C=O vdM

From experimental His/C=O

structure

Selecting vdMs

Design and synthesis

Experimental testing and validation of design

Validation

Complementary vdMs are sampled and the

highest- scoring ones is selected.

His

HisHis/

GRAPHIC: N. DESAI/

SCIENCE

De novo protein design

Polizzi and DeGrado developed a unit of local

protein structure called a van der Mer (vdM) that

links ligand group chemical functionality and

main-chain backbone coordinates to help design

ligand-binding proteins.

Published by AAAS