Liver De-Targeting
Cyagen has developed an AI platform that integrates big data, cloud computing, machine
learning,
and other technologies to optimize the AAV9 capsid protein. This has generated a large number of
candidate
variants. The results show a high level of confidence in the predicted liver de-targeting data
(Figure 1), with
a PearsonR correlation coefficient as high as 0.884.
Figure 1. Predicted Liver Infection Efficiency
Top sequences selected from AI-predicted sequences were individually validated by tail vein
injection in
mice (5E11 vg/each) and examined after 21 days. In vivo imaging results (Figure 2) demonstrate
that PM167
exhibits significantly better liver de-targeting than PHP.eB, while PM170 shows significantly
better
liver de-targeting than AAV9 wild-type (WT) and slightly higher than PHP.eB.
Figure 2. In Vivo Validation of Liver Infection Efficiency (In
Vivo Imaging
- Luc)
After tail vein injection in mice (5E11 vg/each) and a 21-day incubation period, frozen section
results
(Figure 3) reveal that PM167 exhibits significantly lower green fluorescent protein signals in
the liver
compared to PHP.eB. PM170, on the other hand, shows slightly higher green fluorescent protein
signals in the
liver than PHP.eB but still significantly lower than AAV9 wild-type (WT).
Figure 3. In Vivo Validation of Liver Infection Efficiency (Frozen Section
EGFP)
High Targeting to the Central Nervous System
Cyagen has developed an AI platform that integrates big data, cloud computing, machine learning,
and other
technologies to optimize the AAV9 capsid protein. This has generated a large number of candidate
variants.
The results show a high level of confidence in the predicted central nervous system targeting
data (Figure
4), with a PearsonR correlation coefficient as high as 0.843.
Figure 4. Predicted Central Nervous System
Infection Efficiency
Top sequences selected from AI-predicted sequences were individually validated by tail vein
injection in
mice (5E11 vg/each) and examined after 21 days. In vivo imaging results (Figure 5) show a high
accumulation
of Top sequences in the brain. The Luc signal intensity in the brain expressed by PM167 is
approximately two
times that of PHP.eB, while the Luc signal intensity in the brain expressed by PM170 is
approximately 1.5
times that of PHP.eB.
Figure 5. In Vivo Validation of Brain Infection Efficiency (In
Vivo Imaging,
Luc)
To further investigate the distribution of Top sequences in different regions of the central
nervous system,
we conducted another examination in mice after tail vein injection (5E11 vg/each) and a 21-day
incubation
period. Frozen section results (Figure 6) demonstrate that PM167 exhibits significantly higher
green
fluorescent protein signals than PHP.eB in various brain regions (cortex, corpus callosum,
hippocampus,
midbrain) and the spinal cord. PM170 also shows higher green fluorescent protein signals than
PHP.eB in
various brain regions and the spinal cord, except for the corpus callosum.
Cortex
Corpus callosum
Hippocampus
Midbrain
Spinal cord
Figure 6. In Vivo Validation of
Brain and
Spinal Cord Infection Efficiency (Frozen Section EGFP)
Top sequences selected from AI-predicted sequences were pooled in equal ratios into a single test
article
and used to deliver barcoded transgene reporters respectively. We injected the mixed test
article (5E12
vg total) into the cisterna magna of a non-human primate (♀, 3.6 kg). Following 16 days in-life,
animal
was sacrificed and tissues were processed for next generation sequencing (NGS) and and
histology.
Figure 7. In Vivo Validation of Central Nervous
System Infection Efficiency (NGS)
Figure 8. In Vivo Validation of Central Nervous
System Infection Efficiency (Frozen Slice EGFP)
Using the Cyagen AI-AAV platform, we constructed a high-capacity mutant plasmid library,
packaged a virus
library, and performed NGS sequencing. We built a DualConvLSTM network to establish an AAV2
production
prediction model. The model's credibility was validated on the test set, achieving a high
correlation with
Pearson=0.929 and Spearman=0.859 (Figure 9). Additionally, the AI-generated retinal targeting
model showed a
correlation of Pearson=0.874 and Spearman=0.871 on the test set (Figure 10).
Figure 9. Correlation of the Production
Prediction Model
on the Test Set
Figure 10. Correlation of the Retinal
Targeting Model on
the Test Set
We used the production model and the retinal targeting AI model to predict variants with high
production and
expression capabilities. We selected the top sequences and constructed RC mutant plasmids. These
plasmids
were separately packaged with wild-type AAV2 plasmids and 7M8 plasmids to produce Luciferase
viruses. After
virus packaging, purification, and QPCR titer testing, all three variants showed higher yields
compared to
AAV2 and AAV2.7M8. Specifically, PM054 had the highest yield, which was 3.48 times that of AAV2.
PM021 and
PM077 had yields 1.5 times and 2.01 times that of AAV2, respectively.
Figure 11. AI Predicted AAV2 Variant Virus Yields
The packaged Luciferase viruses were injected into the vitreous cavity of mice at a dosage of
3E+9 vg (viral
genomes) per eye. After 3 weeks, Luciferase expression was detected using both in vivo imaging
(Figure 12)
and chemiluminescence assays. The in vivo imaging results showed that the signal intensity of
Luciferase for
all three variants was higher than that of AAV2 and AAV2.7M8.
For a more precise quantification, mice were euthanized, and their eyeballs were collected and
homogenized
for chemiluminescence detection. The PM054 variant exhibited the highest Luciferase expression
level, which
was 15 times that of AAV2. PM021 and PM077 had Luciferase expression levels of 12.5 times and 10
times that
of AAV2, respectively.
Figure 12. In vivo Validation of Whole Eye
Expression
Capability (In Vivo Imaging)
Figure 13. In vivo Validation of Whole Eye
Expression Capability (Chemiluminescence)
To further validate the in vivo infection efficiency of the variants and explore the cell types
infected, EGFP viruses were packaged and injected into the vitreous cavity of mice (3E+9 vg per
eye). After 3 weeks, the overall EGFP expression was detected through fundus fluorescence
photography, and eyeball samples were collected for pathological examination. The fundus
photography results (Figure 14) showed that the GFP signal of PM054 was the strongest, and the
fluorescence signals of all three variants were significantly higher than those of AAV2 and
AAV2.7M8.
Figure 14. In vivo Validation of Whole Eye
Infection Capability (Fundus Photography)
DAPI staining was performed on frozen sections of eyeballs, and the results showed that the
infection range of all three variants was greater than that of AAV2-WT and AAV2.7M8. In
particular, PM021 and PM054 could infect almost the entire retinal area. In terms of the
infection depth, AAV2-WT only infected the RGC layer cells, while AAV2.7M8 had some penetration
capability, infecting a small number of optic nerve cells in the posterior retina. All three
variants exhibited greater penetration capability than AAV2-WT and AAV2.7M8, infecting cells in
various layers of the retina from RGC to PRC. Among them, PM054 had the best infection rate and
expression in optic nerve cells, with an infectivity in the posterior retina approximately 10
times that of AAV2.7M8.
Figure 15. In vivo Validation of Whole Eye
Expression Capability and Retinal Posterior Layer Penetration Ability
Should the in vivo NHP validation prove successful, researchers may opt to purchase the remaining 50% of the IP rights.
If the NHP in vivo validation fails, we will offer a certain amount of virus packaging services free of charge.
Join Cyagen's AAV Innovation 100 Plan
Validation Data
Platform Advantages
