Mutagenesis of ACE2 for Enhanced Affinity
Soluble decoy receptors should ideally bind the SARS-CoV-2 spike tighter
than, and out-compete, the wild type (WT) ACE2 receptor. Tight affinity
is also necessary for effective virus neutralization at typical doses
for biologic drugs. The importance of affinity is further emphasized
when soluble decoys are considered against the therapeutic alternative
of affinity-matured monoclonal antibodies, which bind S of SARS-CoV-2
with tight nanomolar to picomolar
KD28,29,30,33; by comparison, WT sACE2
has only moderate affinity (KD ≈ 20 nM)16,43. To this
end, mutagenic approaches, either targeted or through selection of
diverse libraries of variants, provide a means for engineering decoy
receptors with enhanced affinity.
In the first report44 that mutations within ACE2 can
indeed be found that increase S affinity, Chan et al. performed deep
mutagenesis on full length ACE243,45. A library of
ACE2 variants was created that encoded all possible single amino acid
variants at 117 sites, covering both the regions that interface with the
RBD as well as the ACE2 active site. The library was then selected by
fluorescence activated cell sorting (FACS) for ACE2 surface expression
in human cells and tight binding to the RBD of SARS-CoV-2 protein S
(Figure 2). ACE2 residues buried at the interface with S tend to be more
conserved, while ACE2 residues at the interface periphery or within the
active site were mutationally tolerant. Mutations of ACE2 residues N90
and T92, which form an N-glycosylation motif, are universally enriched
for high binding affinity (with the exception of T92S), indicating that
N90 glycosylation hinders binding, at least in this experiment’s cell
line. Other mutations were also found dispersed across multiple sites
that increase S binding, and some of these were rapidly screened as
combinations of mutations in soluble ACE2 conjugated to superfolder GFP
(sfGFP)46. Expression medium containing sACE2-sfGFP,
without purification, can be incubated directly with cells expressing S
and bound sACE2-sfGFP detected by flow cytometry. Ultimately, variants
of sACE2 carrying four or three mutations (Figure 2C), dubbed sACE2.v2
and sACE2.v2.4, were engineered for an optimal balance of high affinity,
expression and yield. Impressively, monomeric sACE2.v2, which spans ACE2
residues 19-615 that form the extracellular protease domain, outcompeted
WT sACE2 fused to dimeric Fc of IgG1 and also competed effectively with
anti-RBD Ig derived from Covid-19 positive patient sera. Soluble ACE2
forms a natural dimer (which we refer to as sACE22) if
the construct is expanded up to ACE2 residues 730-740, which includes
the dimerization/collectrin-like domain immediately C-terminal of the
protease domain. Soluble ACE22 has increased apparent
affinity and neutralization of authentic virus through avidity,
demonstrating that each protease domain in sACE22 can
bridge two S proteins. The leading variant from these efforts, dimeric
sACE22.v2.4, had a picomolar KD for the
SARS-CoV-2 RBD and achieved 50% relative infection inhibition of
authentic virus at subnanomolar concentrations. Surprisingly, it showed
similar neutralization efficacy against SARS-CoV-1 despite no
consideration of binding to this virus during mutagenesis. This implied
potential efficacy against other zoonotic coronaviruses that utilize
ACE2 and may spill over to humans in the future. This was confirmed by
demonstrating sACE22.v2.4 tightly binds diverse RBD
sequences from SARS-related bat betacoronaviruses that use ACE2 as an
entry receptor40. Furthermore, an in vitro selection
of mutations within the RBD from SARS-CoV-2 failed to find S variants
that lose affinity for the engineered sACE22.v2.4 decoy
but retain binding to the wild type host ACE2 receptor, thus confirming
the hypothesis that mechanisms for viruses to escape soluble decoy
receptors are limited40.
In a similar deep mutagenesis experiment, Heinzelman and Romero
investigated the influence on S protein binding of single-nucleotide
variants (SNVs) in the ACE2 protease domain, evaluating nearly 4,000
amino acid substitutions47. Mutations were made using
error-prone PCR, a rapid and reliable way of generating library
diversity but which fails to capture multiple nucleotide changes within
a single codon (unless error rates are excessively high). The library of
the ACE2 protease domain was screened by yeast surface display (Figure
2B), and while most mutations decreased binding (it is, after all, easy
to ’break’ a protein through random mutagenesis), approximately 4% of
mutations were found to increase RBD binding. While mutations in the
active site tended to have little effect (with the caveat that the
background ACE2 sequence used for library generation already had
catalytic activity knocked out), the authors found that residues in the
chloride-binding site of ACE2, which regulates its peptidase
activity48 and is over 40 Å from the RBD interface, do
affect binding. In addition, a separate distal cluster of hydrophobic
residues (L236, F588, L591, and L595) was also identified as a critical
site affecting binding despite being over 30 Å from the RBD interface.
By interrogating SNVs, the authors were also able to statistically
evaluate the clinical and epidemiological relevance of ACE2 allelic
variants in the human population, and they estimated that roughly 1 in
10,000 people may have SNVs that increase spike binding and roughly 4 in
1000 having SNVs that decrease spike binding. However, these predictions
will need to be confirmed by genetics studies of patient cohorts,
especially since mutations in ACE2 with altered affinity for S may also
have other effects that impact infection in vivo, such as expression
changes in human tissues. Overall, this work highlights the importance
of considering long-range effects of mutations distal from the
functional binding site.
The deep mutational scans of ACE2 are based on a single round of
sequence diversification and selection, whereas others have employed
multiple rounds to direct the in vitro evolution of ACE2 towards
exceptionally tight affinity for SARS-CoV-2 S. Higuchi et al. focused
mutagenesis to ACE2 residues 18-102 and 272-409 in the protease domain,
respectively denoted as PD1 and PD2, that form the RBD
interface49. Binding was assayed using the method
previously developed for deep mutagenesis43, with
human cells expressing the ACE2 library incubated with RBD-sfGFP and
sorted by FACS. Mutations within the ”PD1” region were selected first,
followed by mutagenesis within the ”PD2” region and additional rounds of
selection, although additional mutations within the PD2 region did not
substantially improve affinity. The ACE2 variant identified with highest
affinity, termed 3N39 with 7 substitutions (Figure 2C), bound RBD with
picomolar KD. Despite having a much higher mutational
load than sACE2.v2.4, soluble 3N39-Fc fusions were soluble with minimal
aggregation by size exclusion chromatography. The engineered 3N39-Fc
fusion vastly outperformed the equivalent wild type ACE2-Fc fusion in
virus neutralization.
The Kortemme, Hobman, and Wells groups went even further in using
multiple rounds of engineering, diversification and selection to drive
down the dissociation constant to the picomolar
range50. Computational alanine scanning identified
ACE2 residues H34, Q42 and K353 as hot spots that contribute
disproportionately to the binding energy, and these residues with their
neighbors were the focus for computational mutagenesis. Computationally
designed ACE2 variants were purified as Fc fusions and binding to S was
improved 3- to 11-fold. Four of these ACE2 variants served as parents
that were then matured through yeast surface display, resulting in a
further 14-fold improvement in apparent KD. Deep
mutagenesis data from Chan et al43 was then
considered, from which new ACE2 variants were designed with mutations
that were not alanine scan-based hotspots and included sites outside the
RBD interface that likely stabilize conformation. The final ACE2
variants had exceptionally tight affinity, with the lowest
KD variants carrying 4-8 substitutions (Figure 2C). It
was also independently confirmed that additional increases in apparent
affinity can be achieved through inclusion of the ACE2 dimerization
domain for avid binding and protein stabilization. It is worth noting
that in this study deliberate action was taken to inactivate ACE2
enzymatic activity. This has historically been done by mutating two
critical Zn2+-coordinating residues (H374N and
H378N)51, but because these substitutions resulted in
undesirable destabilization in the variant background, an alternative
mutation (H345L within the substrate-binding cavity) was adopted. While
the rationale for this is the elimination of vasodilatory effects from
excess ACE2 peptidase activity that may cause hypotension, it is worth
considering the potential benefits of keeping enzymatic activity
(discussed below).
Pangolin (Pangolin-CoV-2020) and bat (isolate RaTG13) coronaviruses, in
addition to human SARS-CoV-1 and -2, can initiate membrane fusion using
not only human ACE2 but many other animal orthologs52.
This informed the substitution of human ACE2-D30 for
glutamate52, which is found in ACE2 of other species
and is able to better reach K417 of the RBD for salt bridge
formation16,17. In addition, inclusion of the
dimerization domain together with the protease domain further
facilitated the engineering of a sACE2-Ig fusion protein effective
against multiple betacoronavirus strains. The discovery of the
affinity-enhancing D30E mutation and benefits of including the
dimerization domain overlap with independent work by
others43,50. Another contributing factor was increased
avidity (to be discussed further below) due to design of a tetrameric
sACE2-Ig configuration52. Overall, it is apparent from
all these studies that a truly effective therapeutic approach utilizing
decoy ACE2 receptors will likely incorporate a multifaceted strategy of
affinity enrichment, avidity and Fc fusions.
One final aspect to touch upon regarding affinity enhancement through
mutagenesis is the connection between binding affinity and ACE2
enzymatic activity. For example, the sACE2.v2.4 variant has reduced
catalytic activity even though its 3 amino acid substitutions are not
within the active site43, thus demonstrating
potentially unappreciated coupling between S affinity and ACE2
catalysis. Others have observed that some mutations that knock out ACE2
activity can increase S binding53, although many
mutations in the active site have no effect on S
affinity43,47. In particular, we highlight again the
key finding by Heinzelman and Romero that mutations to a
chloride-binding motif of ACE2, which is 40 Å from the RBD interface and
is directly involved in regulating peptidase activity, can affect spike
affinity, possibly due to conformational effects47.
The binding of trimeric SARS-CoV-2 spike protein to wild type ACE2 also
increases its peptidase activity 3- to 10-fold, with the SARS-CoV-2 RBD
inducing bending of the N-terminal protease subdomain toward the
C-terminal protease subdomain “…reminiscent of a closing
dam”54. Although the precise mechanisms remain
unclear, there are nonetheless multiple clues incriminating ACE2
conformation and dynamics on both catalytic activity and S affinity. We
believe from a therapeutic perspective (to be discussed further below),
maintaining catalytic activity is preferable due to the protective
effects of ACE2 products on the pulmonary and cardiovascular systems.