of 12
Article
https://doi.org/10.1038/s41467-024-48974-y
Engineering viral vectors for acoustically
targeted gene delivery
Hongyi R. Li
1
,ManwalHarb
2
, John E. Heath
1
, James S. Trippett
2
,
Mikhail G. Shapiro
3,4,5
&JerzyO.Szablowski
2,3,6,7
Targeted gene delivery to the brain is a critical tool for neuroscience
research and has signi
fi
cant potential to treat human disease. However, the
site-speci
fi
c delivery of common gene vectors such as adeno-associated
viruses (AAVs) is typically performed via invasive injections, which limit its
applicable scope of research and clinical applications. Alternatively,
focused ultrasound blood-brain-barrier opening (FUS-BBBO), performed
noninvasively, enables the site-speci
fi
c entry of AAVs into the brain from
systemic circulation. However, when used in conjunction with natural AAV
serotypes, this approach has limited transduction ef
fi
ciency and results in
substantial undesirable transduction of peripheral organs. Here, we use
high throughput in vivo selection to engineer new AAV vectors speci
fi
cally
designed for local neuronal transduction at the site of FUS-BBBO. The
resulting vectors substantially enhance ultrasound-targeted gene delivery
and neuronal tropism while reducing peripheral transduction, providing a
more than ten-fold improvement in targeting speci
fi
city in two tested
mouse strains. In addition to enhancing the only known approach to non-
invasively target gene delivery to speci
fi
cbrainregions,theseresults
establish the ability of AAV vectors to be evolved for speci
fi
c physical
delivery mechanisms.
Gene therapy is one of the most promising emerging approaches to
treating human disease. Recently, a number of gene therapies were
approved for clinical use to treat diseases such as blindness
1
,muscular
dystrophy
2
, and metabolic disorders
3
with Adeno-Associated Viral
vectors (AAVs). Gene therapy could also potentially target brain dis-
orders. Unfortunately, gene delivery to the brain remains a major
challenge. The typical approach for the administration of such gene
therapies involves a surgical injection directly into the brain par-
enchyma, which is invasive. Other studies show it may also be possible
to achieve brain-wide gene delivery with systemic
4
6
or intrathecal
injections
7
. However, these approaches, while noninvasive, lack spatial
precision and thus cannot target regionally de
fi
ned neural circuits.
Focused ultrasound blood-brain barrier opening (FUS-BBBO) has
the potential to overcome these lim
itations by providing a route to
noninvasive, site-speci
fi
c gene delivery to the brain
8
12
. In FUS-BBBO
ultrasound is focused through an intact skull
13
,
14
to transiently loosen
tight junctions in the BBB and allow for the passage of AAVs from the
blood into the targeted brain site. Other mechanisms of FUS-BBBO
could include increased transcytosis
15
and decreased levels of ef
fl
ux
transporters
16
. FUS-BBBO can target intravenously administered AAVs
Received: 26 July 2021
Accepted: 21 May 2024
Check for updates
1
Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA.
2
Department of Bioengineering, Rice University,
Houston, TX, USA.
3
Division of Chemistry and Chemical Engineering, California Institute of Technology, Pasadena, CA, USA.
4
Andrew and Peggy Cherng
Department of Medical Engineering, California Institute of Technology, Pasadena, CA, USA.
5
Howard Hughes Medical Institute, Pasadena, CA, USA.
6
Rice
Neuroengineering Initiative, Rice University, Houston, TX, USA.
7
Rice Synthetic Biology Institute, Rice University, Houston, TX, USA.
e-mail:
mikhail@caltech.edu
;
jszab@rice.edu
Nature Communications
| (2024) 15:4924
1
1234567890():,;
1234567890():,;
to millimeter-sized brain sites or cover large regions of the brain
without apparent tissue damage in the tested timeframes
17
,
18
.These
capabilities place FUS-BBBO in contrast with intraparenchymal injec-
tions, which are invasive and deliver genes to a single 2
3 millimeter-
sized region per injection
19
,
20
, requiring a large number of brain
penetrations to cover larger regions of interest. At the same time, the
spatial targeting capability of FUS-BBBO differentiates it from the use
of spontaneously brain-penetrating engineered AAV serotypes which
lack spatial speci
fi
city
5
. In proof of concept studies, FUS-BBBO has
been used in rodents to introduce AAVs encoding reporter genes such
as GFP
8
,
9
,
17
,
21
, growth factors
22
, and optogenetic receptors
10
.Thedeliv-
ery of chemogenetic receptors to the hippocampus provided the
ability to modulate memory formation
11
.
Despite its promise, three critical drawbacks currently limit the
potential of FUS-BBBO in research and therapy applications. First, the
BBB effectively limits the transduction of systemically administered
naturally occurring AAVs in non-FUS-targeted regions. peripheral
organs have endothelia that allow AAV entry and consequently receive
a high dose of the virus, which could lead to toxicity
23
.Second,the
relative inef
fi
ciency of AAV entry at the site of FUS-BBBO have led
published studies to use doses that
were higher than those needed for
direct intraparenchymal injections, which in the clinic typically range
from 10
10
to 10
12
viral genomes (VGs) per site injected, compared to
10
12
10
14
VGs per kilogram of body weight for intravenous route
24
.In
our previous work, to achieve transduction ef
fi
ciency comparable to
such injections at 5 × 10
8
VGs, we used 10
10
VGs per gram of body
weight intravenously with FUS-BBBO
11
.TheAAV9dosesusedinother
FUS-BBBO studies to date have ranged from 5 × 10
8
to 1.67 × 10
10
VGs
per gram of body weight
8
,
9
,
11
,
21
,
25
,
26
. Lowering the viral doses would
reduce the chances of peripheral toxicity, and the costs of potential
therapies
24
.
We reasoned that these limitations arise from the fact that wild-
type serotypes of AAV did not evolve to cross physically loosened
biophysical barriers and are therefore not optimal for this purpose. We
hypothesized that we could address these limitations by developing
new engineered viral serotypes speci
fi
cally optimized for FUS-BBBO
delivery. Capsid engineering techniques
27
in which mutations are
introduced into viral capsid prote
ins have been used to enhance gene
delivery properties such as tissue speci
fi
city
5
,
6
,
28
30
, immune evasion
31
,
or axonal tracing
32
. However, they have not yet been used to optimize
viral vectors to work in conjunction with speci
fi
c physical delivery
mechanisms.
To test our hypothesi
s, we performed in vivo selection of muta-
genized AAVs in mice in conjunction with FUS-BBBO (Fig.
1
)by
adapting a recently developed Cre-recombinase-based screening
methodology
6
,
30
.Weidenti
fi
ed 5 viral capsid mutants with enhanced
transduction at the site of FUS-BBBO and not in the untargeted brain
regions. We then performed detailed validation experiments com-
paring each of these mutants to the parent wild-type AAV, revealing a
signi
fi
cant increase in on-target transduction ef
fi
ciency, increased
neuronal tropism, and a marked decrease in off-target transduction in
peripheral organs, with an overall performance improvement of more
than 10-fold. These results demonstrate the evolvability of AAVs for
speci
fi
c physical delivery methods.
Results
High-throughput in vivo screening for AAVs with ef
fi
cient FUS-
BBBO transduction
To identify new AAV variants with improved FUS-BBBO-targeted
transduction of neurons, we generated a library of viral capsid
sequences containing insertions of 7 randomized amino acids between
residues 588 and 589 of the AAV9 capsid protein (Supplementary
Fig. S1). Such 7-mer insertions have been widely used to engineer AAVs
with new properties
5
,
6
,
27
32
. We chose AAV9 as a starting point due to its
use in previous FUS-BBBO studies
8
,
9
,
11
and superior transduction
compared to other naturally occurring AAV serotypes
21
.
To make the screening more ef
fi
cient, we employed
recombination-based AAV selection
6
,
30
.ThisapproachusesaCre
recombinase inside the cells to invert a fragment of the vector
sDNA.
(Supplementary Fig. S1a). Because Cre is only present inside the cells,
this approach allows for the identi
fi
cation of capsid variants that can
enter the cells and deliver their DNA to the nucleus. These Cre-inverted
DNA sequences can then be detected by PCR using primers speci
fi
cto
the inverted section of the DNA (Supplementary Fig. S1b). Here, we
used transgenic mice that expressed Cre in neurons, to select for AAVs
with improved neuronal transduction
5
,
6
,
33
.
To ensure we selected for AAVs transduced speci
fi
cally within the
FUS-BBBO-targeted areas we started with a library of 1.3 × 10
9
AAV
candidates delivered to one hemisphere with FUS-BBBO (Fig.
1
a, b). We
then extracted the viral DNA that was delivered to the targeted
hemisphere, and re-screened the extracted variants again to quantify
speci
fi
city and ef
fi
ciency of FUS-BBBO-mediated transduction. We
targeted 4 sites within one hemisphere using magnetic resonance
imaging (MRI) guidance, and con
fi
rmed the successful BBB opening
through imaging of gadolinium contrast agent extravasation (Fig.
2
a).
We employed FUS parameters below tissue damage limits
11
,
34
(0.33 MPa at 1.5 MHz, 10 ms pulse length, 1 Hz repetition frequency,
0.22
μ
l dose of microbubbles per gram of body weight). The AAV
libraries were delivered intravenously (IV) immediately following FUS
application to the brain at a dose of at a dose of 6.7×10
9
VGs per gram
of body weight. We then allowed for 2 weeks of expression, euthanized
the mice for tissue collection. Immediately after, we extracted the viral
DNA from the brain and used Cre-dependent PCR ampli
fi
cation to
selectively amplify the Cre-modi
fi
ed viral DNA, with a goal of
fi
nding
Control
FUS-BBBO
I.V. AAV library
14 days
DNA
extraction
Cre-
dependent
PCR
Enriched
by FUS
Untargeted
hemisphere
Next-Gen. sequencing:
Synthesize sequences specific to FUS-site + clone into new library
repeat
FUS
round 1 library
1.3E9
sequences
round 2 library
2.1E3
sequences
testing
5 candidates
AAV.FUS
ab
Fig. 1 | Screening methodology for generation of an AAV for improved site-
speci
fi
c noninvasive gene delivery to the brain. a
Summary of the high-
throughput screening and selection process. AAV library is administered intrave-
nously (I.V.) and delivered to one brain hemisphere through FUS-BBBO. After
14 days mice are euthanized, their brain harvested, and the DNA from selected
hemispheres is extracted. The DNA is then ampli
fi
ed by Cre-dependent PCR that
enriches the viral DNA modi
fi
ed by Cre. In our case, neurons expressed Cre
exclusively, and the Cre-dependent PCR enriched viral DNA of AAVs that
transduced neurons. We subjected the obtained viral DNA to next-generation
sequencing for the targeted hemisphere (round 1) or both targeted and control
hemispheres (round 2). The process is then repeated for the next round (steps
exclusive to round 2 indicated by the gray text).
b
Overall, 1.3 billion clones were
screened in the
fi
rst round, and 2098 clones in the second round of selection. Out
of these clones, we selected 5 that were tested in low-throughput to yield AAV.-
FUS.3
a vector with enhanced FUS-BBBO gene delivery.
Article
https://doi.org/10.1038/s41467-024-48974-y
Nature Communications
| (2024) 15:4924
2
AAVs selectively transducing neurons. We then sequenced the
obtained DNA with next-generation sequencing (NGS) of the region of
the 7-mer insertion and selected the 2098 most abundant sequences
for subsequent evaluation. This screen selected for AAVs which could
enter the neurons. However, these variants could not be quantitatively
compared at this stage, due to large number of vectors in library
compared to the total administered dose. As a result, each AAV clone
existed in the library in a small copy number preventing statistically
meaningful comparisons between each AAV candidates.
Instead, to quantitatively compare our 2098 down-selected cap-
sid variants, we re-synthesized and packaged them as a new AAV library
at a dose of 1.3 × 10
9
viral genomes per gram of body weight, corre-
sponding to ~1.5
3×10
7
viral genomes of each clone being injected
into each mouse. In each of the two hSyn-CRE mice, we injected the
AAV library intravenously and opened the BBB in one hemisphere
using MRI-guided FUS as in round 1. Two weeks after treatment, we
performed a series of procedures on each mouse. First, we removed
the brain and separated the two hemispheres. We then extracted DNA
from both the hemisphere that was targeted by the FUS and the
hemisphere that was not. The DNA extract was ampli
fi
ed by the CRE-
dependent PCR to enrich for viral genomes that transduced neurons.
After FUS-BBBO delivery, DNA extraction, CRE-dependent PCR, and
NGS, we recovered 1433 sequences.
To identify the most improved candidates, we examined their
copy number in each hemisphere (Fig.
2
b). To identify AAVs that
selectively transduced sites that underwent FUS-BBBO, we
fi
rst
looked for variants that were at least 100-fold more represented in
the targeted hemisphere relative to the untargeted hemisphere.
From this list, we further selected candidates for which the 100-fold
difference was maintained in both mice. To ensure that the
sequences were not the result of sequencing error, we selected
candidates that were found with two alternative codon sequences
corresponding to its 7-mer peptide. In the end, 35 sequences met
these criteria (dark gray symbols in Fig.
2
b). Among these FUS-
BBBO-speci
fi
c variants, we chose the 5 most common sequences,
which we hypothesized would code for AAV capsids with the most
ef
fi
cient neuronal transduction. We re-synthesized these sequences
(Supplementary Table 1), cloned them into the AAV9 capsid
between amino acids 587
588, and packaged them for detailed
evaluation, naming them AAV.FUS 1 through 5.
AAV.FUS candidates show enhanced transduction of neurons in
targeted brain regions and reduced transduction of
peripheral organs
An ideal AAV vector for ultrasound-mediated gene delivery to the brain
would ef
fi
ciently transduce targeted neurons while avoiding the
transduction of peripheral tissues, such as the liver which is highly
transduced by the naturally-occurring AAV serotypes
35
.Additionally,
such a vector should only transduce the brain at the FUS-targeted sites.
Of the natural AAV serotypes, AAV9 is most commonly used in FUS-
BBBO because it transduces neurons at the ultrasound target with
relatively high speci
fi
city and ef
fi
ciency compared to untargeted brain
regions
8
,
10
,
11
,
21
. However, AAV9 also shows peripheral transduction and
is typically administered at doses higher than those used in direct
intraparenchymal injection
8
,
10
,
11
, leaving room for improvement. To
evaluate our engineered vectors, we used AAV9 as a benchmark and an
internal control for each tested animal.
We performed FUS-BBBO while intravenously co-administering
each AAV.FUS candidate alongside AAV9 in individual comparison
experiments at 1E10 VGs per gram of body weight. Consequently, each
mouse had an internal control where the injected volume, targeted
brain site, and the ef
fi
ciency of FUS-BBBO were identical for both
serotypes, leaving the ef
fi
ciency of the vector as the independent
variable. To quantify the transduction ef
fi
ciency, we encoded the
fl
uorescent proteins mCherry and EGFP in AAV9 and each AAV.FUS
variant, respectively, under a cell-type nonselective CaG promoter
36
.
After 2 weeks of expression, we counted the numbers of mCherry and
EGFP-expressing cells within the sites of FUS-BBBO. We established the
reliability of this quanti
fi
cation method by comparing cell counts in the
brain for co-administered AAV9-EGFP and AAV9-mCherry (Supple-
mentaryFig.S2).Ourquanti
fi
cation showed that AAV.FUS.1, 2, 3, and 5
had signi
fi
cantly improved transduction ef
fi
ciency compared to AAV9
(
p
= 0.0274, 0.0003, 0.0052, 0.0087, respectively, two-way ANOVA
with Sidak
s multiple comparisons test,
F
(4,24) = 59.49, Fig.
3
a, b)
ab
Fraction of sequences in NGS
(targeted hemisphere)
1E-08
1E-07
1E-06
1E-05
1E-04
1E-03
1E-02
1E-08 1E-07 1E-06 1E-05 1E-04 1E-03 1E-02
Fraction of sequences in NGS pool
(untargeted control)
Fig. 2 | High throughput screening yields vectors with improved FUS-BBBO
gene delivery. a
An MRI image showing mouse brain with 4 sites opened with FUS-
BBBO in one hemisphere. The bright areas (arrowheads) indicate successful BBB
opening and extravasation of the MRI contrast agent Prohance into the brain. This
BBB opening was used for delivery of the AAV library.
b
Sequencing results of round
2 of screening show a fraction of NGS reads within the DNA extracted from brains of
Syn1-Cre mice subjected to FUS-BBBO and injected with a focused library of 2098
clones. Each dot represents a unique capsid protein sequence, and the position on
each axis corresponds to the number of times the sequence was detected in the
FUS-targeted and untargeted hemispheres. Markers below the dotted line
represent sequences that on average showed 100-fold higher enrichment in the
targeted hemisphere as compared to the control hemisphere. Dark gray dots
represent 22 clones that are enriched in the FUS targeted hemispheres at least 100-
fold in every tested mouse and DNA sequence encoding the 7-mer insertion pep-
tide. Additional 13 clones had zero detected transduction in the untargeted hemi-
sphere and could not be presented on the log-log plot. Yellow dots represent 5
clones (AAV.FUS.1-5) selected for low-throughput testing. Due to the use of a
logarithmic plot, clones that had zero copies detected in either of the hemispheres
are not shown. Data from one male and one female mouse.
Article
https://doi.org/10.1038/s41467-024-48974-y
Nature Communications
| (2024) 15:4924
3
whereas AAV.FUS.4 showed no improvement (
p
= 0.2556). The fold-
change in transduction relative to AAV9 was greatest for AAV.FUS.2,
and lowest for AAV.FUS.4 (Supplementary Fig. S3). None of the AAV.-
FUS candidates produced substantial off-target expression within the
brain at sites not insonated by FUS, with AAV9 producing 0.29 ± 0.1%
neuronal transduction (
n
= 40 mice), AAV.FUS.3 0.17 ± 0.1% (
n
=17
mice), and other AAV.FUS candidates between 0.24 ± 0.12% (
n
=6),
0.37 ± 0.26% (
n
= 5), 0.2 ± 0.26% (
n
= 6), 0.026 ± 0.05% (
n
=6) for
AAV.FUS.1, 2, 4, and 5 respectively (Fig.
3
c).
Next, we evaluated the extent to which AAV.FUS candidates
transduce off-target peripheral organs. In mice that received
intravenous co-injections of AAV9-mCherry and each variant of
AAV.FUS-EGFP, we counted transduced cells in the liver, a periph-
eral organ known to be targeted by AAVs and a potential source of
dose-limiting toxicity
37
,
38
. Two weeks after injection, we imaged liver
sections and counted cells expressing each
fl
uorophore (Fig.
3
d, e).
We found markedly reduced liver transduction among the AAV.FUS
candidates compared to AAV9 (Fig.
3
e). AAV.FUS 3 showed the
AAV.FUS.1
AAV.FUS.2
AAV.FUS.3
AAV.FUS.4
AAV.FUS.5
0
1
2
3
AAV.FUS / AAV9 transduciton
ns
a
b
d
e
c
f
AAV.FUS.1
AAV.FUS.2
AAV.FUS.3
AAV.FUS.4
AAV.FUS.5
0
5
10
15
20
Fold-improvement
AAV.FUS.1
AAV.FUS.2
AAV.FUS.3
AAV.FUS.4
AAV.FUS.5
0.0
0.2
0.4
0.6
0.8
1.0
AAV.FUS / AAV9 transduciton
Liver
Liver
Brain
Brain
AAV9
AAV.FUS.3
AAV9
AAV.FUS.3
NeuN
AAV.FUS.3
AAV9
AAV.FUS.3
AAV9
AAV.FUS.1
AAV.FUS.2
AAV.FUS.3
AAV.FUS.4
AAV.FUS.5
0.00
0.02
0.04
0.06
0.08
0.10
ratio of transduced cells
to number of neurons
AAV9
AAV.FUS
ns
ns
ns
ns
ns
1 mm
Article
https://doi.org/10.1038/s41467-024-48974-y
Nature Communications
| (2024) 15:4924
4