41592_2020_799_MOESM1

Articles

https://doi.org/10.1038/s41592-020-0799-7

Multiplexed Cre-dependent selection yields

systemic AAVs for targeting distinct brain cell

types

Sripriya Ravindra Kumar

, Timothy F. Miles

1,5

, Xinhong Chen

1,5

, David Brown

, Tatyana Dobreva

Qin Huang

1,2

, Xiaozhe Ding

, Yicheng Luo

, Pétur H. Einarsson

, Alon Greenbaum

1,3,4

, Min J. Jang

Benjamin E. Deverman

1,2

and Viviana Gradinaru

✉

Division of Biology and Biological Engineering, California Institute of Technology, Pasadena, CA, USA.

Present address: Stanley Center for Psychiatric

Research, Broad Institute, Cambridge, MA, USA.

Present address: Joint Department of Biomedical Engineering, North Carolina State University, Raleigh,

NC, USA.

Present address: University of North Carolina at Chapel Hill, Chapel Hill, NC, USA.

These authors contributed equally: Timothy F. Miles,

Xinhong Chen.

✉

e-mail: viviana@caltech.edu

SUPPLEMENTARY INFORMATION

In the format provided by the authors and unedited.

NATuRE METHoDS

www.nature.com/naturemethods

Supplementary Figure

Extended Schematic f

or Multiplexed

CREATE

nd Analysis o

f Round

1 Selection.

Diagram of the genetic switch used in M

CREATE. The Acceptor Vector shows the position of the forward and reverse p

rimers

between the Lox sites that are used for selective recovery of capsids from the Cre

cells. The Rep

AAP∆Cap

vector

shows

deletion of

480 bp

cap

gene in addition to the

stop codons that

are

designed to prevent synthesis of VP1, VP2, and VP3 proteins. AAP protein

translation is unaffected by these modifications.

Schematic of the pro

tocol to selectively recover rAAV genomes from the target

population using the Cre

Lox flipping strategy and preparation of the sample for deep sequencing.

The library coverage for R1 DNA

and virus libraries obtained from specific sequencing depths.

The percentage of variant overlap within the sampled DNA and virus, or

across different Cre lines within tissues, or across tissues from R1 selection.

The distributions of AAV capsid read counts for libraries

recovered by NGS from brain tissue across di

fferent Cre transgenic mice post R1 selection. The dotted line

is illustrative only and

roughly

separates the signal from noise

(see Methods for estimation of signal v

noise)

here

signal in this context represents the

input for the R2 selection.

rAA

V genome recovery from tissues using different treatments are shown with total rAAV genome recovery

from 0.1 g of liver,

Percentage of rAAV genomes recovered per ng of total extracted DNA, and

The CT value (cycle threshold from

qPCR) of rAAV genome e

xtracted by trizol that were treated with SmaI restriction enzyme or untreated and

CT value of mitochondrial

DNA (internal control for smaller genome recovery

fold change = 10.79 (2

ΔCT

))

recovered from 1 ng of total DNA from liver tissue

. In

= 4

mice; 2 from GFAP

Cre line and 2 from Tek

Cre line, each data point is drawn from the mean of three technical replicates, error

bar is mean±S.E.M., Mann

Whitney test

, two

tailed (exact P

value of 0.0286 (

≤

0.05

)

and 0.1143

(

P > 0.05

CI 95%)

)

The data reported

are from one independent trial, and

from three independent trials.

Supplementary Figure

Analysis o

mer

rAAV Libraries From Round

2 Selections.

The vector yields

obtained per 10 ng of capsid DNA library across R1 and R2 vector productions.

Distributions of the DNA and

virus libraries produced by the

synthetic pool

and

PCR pool

methods by the standard score of NGS read counts. The variants in virus

libraries are

sorted by the decreasing order of standard score and their scores from respective DNA libraries are mapped onto them.

Correlations between the standard scores of read counts for the DNA and virus libraries (n = 1 per library) produced by the

synthetic

ool

and

PCR pool

methods is determined by linear least

squares regression, and the regression line (best fit) and R

representing the

coefficient of determination is shown.

Distributions of capsid libraries from brain tissue of two mice (purple and gree

n) used in each

Cre line selection, as produced by the

synthetic pool

(left) and

PCR pool

(right) designs. The distribution of

spike

library introduced in

the

synthetic pool

library design is shown in red (center).

Correlations of enrichment scores o

f variants from the brain libraries (n = 2

per Cre line, mean is plotted) produced by

synthetic pool

and

PCR pool

methods is determined by the same method described in

Supplementary Figure

Analysis o

f Round

mer

Tissue Libraries From

Synthetic Pool

And

PCR Pool

Methods

Correlation analysis between the enrichment score (log10) of two alternate codon replicates of variants from the GFAP

Cre (left),

SNAP

Cre (center), and Syn

Cre (right) brain libraries

linear

least

squares

regression

(n = 2 per Cre line, mean is plotted). The

dotted line separates

the

high

confidence

signal from noise

High confidence signal (below) is assessed by a linear regression line (best

fit)

and

represents

the coefficient

determination

The difference in enrichment score between the two codon replicates of a

variant, across different brain libraries, with over 8000 variants recovered in replicates.

Heatmaps represent the magnitude (log2 fold

change) of AA bias in “out

put” library 1

normalized to

“input” library 2 that reach statistical significance (boxed if

value ≤ 0.0001

, two

sided, two

proportion z

test, except in R1 DNA normalized to known NNK template where one

proportion z

test was performed, and P

values corrected for multiple comparisons using Bonferroni correction) is shown.

R1 DNA library

normalized

NNK template (top left,

~9 million sequences),

virus

normalized to

DNA

libraries (bottom left, ~10 million sequences), R2 GFAP library with enrichment

score above 1.0 in brain

normalized to

R2 virus (top right, 20 sequences

) and R2 SNAP library

with enrichment score above 1.2

normalized to

R2 virus (bottom right, 17 sequences) are shown

(n = 1 for DNA, virus, and n = 2 for brain libraries)

Clustering

analysis of positively enriched variants from Tek, GFAP, and combined neuron brain libraries

(SNAP and Syn)

by PCR pool

design, and

synthetic pool

design with

spike

library are shown with size of nodes representing their relative enrichment in brain, and the

thickness of edges (connecting lines) representing the extent of shared AA identity

between nodes. A distinct family is highlighted in

yellow with the corresponding AA frequency logo below (AA size reflects prevalence and color coded based on AA properties).

Supplementary Figure

AAV

PHP.V1 Efficient

ly Targets t

he Brain Vasculature.

Expression of AAV9 (above) and AAV

PHP.V1 (below) packaging

ssAAV:CAG

mNeonGreen across all

organs is shown

(n=3,

3x10

vg dose per adult C57BL/6J mouse, 3 weeks of expression). The background auto fluoresc

ence is represented in

magenta

Expression in cortical astrocytes (S100

) after IV delivery of

AAV

PHP.V1 (left) and AAV

PHP.eB (right)

psids carrying

ssAAV:

GfABC1D

2xNLS

mTurquoise2

(1x10

vg dose pe

r adult mouse, 4 weeks of expression). Percentage of cortical S100

cells

that overlapped with mTurquoise2 expression is quantified (

= 2,

each data point is mean from 3 images per mouse

Expression of

AAV9,

AAV

PHP.eB

(d),

AAV

PHP.V1

(e)

packaging

ssAAV:

Ple261

iCre

in Ai14

tdTomato reporter adult mouse (n=2

3 per group,

3x10

vg dose per adult mouse, 3 weeks of expression).

, Expression of

AAV

PHP.V1 carrying

self

complementary (sc)

scAAV:

EGFP

(above) and

scAAV:

CAG

EGFP (below). Magenta represents the lectin DyLight 594 staining (n=2

3, 3x10

vg dose per adult

C57BL/6J mouse, 2 weeks of expression).

Experiments in

are reported from one independent trial from a fresh batch of

viruses,

and tittered in the same assay for dosage consistency,

and

validated in two independent trials (n = 2 per group).

Supplementary Figure

AAV

PHP.V2 Variant Exhibits Biased Transduction Towards Brain Vascu

lar Cells.

Transduction of mouse brain by the AAV

PHP.V2 variant and control AAV9, carrying the

ssAAV:

CAG

mNeonGreen (

= 3, 3x10

IV dose per C57BL/6J adult mouse, 3 weeks of expression) is shown. The sagittal brai

n images (left) are matched in fluorescence

intensity to the sagittal brain images in Fig.

. Higher magnification images of AAV

PHP.V2 transduced brain sections stained with

αGLUT or αS100 or αOlig2 (magenta) are shown.

Transduction of brain vasculature by AAV

PHP.V2 carrying

ssAAV:

CAG

DIO

EYFP (green) in Tek

Cre adult mice (left, 1x10

vg IV dose per mouse, 4 weeks of expression) i

s shown, and its efficiency (right) is

determined by the overlap of αGLUT1 staining (red) with EYFP expression across different brain areas

(

= 2,

mean of 3 images per

brain region per mouse)

Transduction of astrocytes by AAV

PHP.V2 in GFAP

Cre adult m

ouse (1x10

vg IV dose per mouse, 4

weeks of expression) is shown. Percentage of cortical S100

cells that overlapped with EYFP expression is quantified (

= 2,

mean of 3

images per mouse).

Supplementary Figure

Furt

her Validation o

f Synthetic Pool a

nd PCR Pool Variants

Demonstrates Higher Confidence i

n Synthetic Pool NGS Data.

Transduction levels of liver hepatocytes quantified as the percentage of DAPI

cells that are EGFP

(

= 3, vectors packaged with

ssAAV:

CAG

2xNL

EGFP, 1x10

vg IV dose/adult C57BL/6J mouse, 3 weeks of expression

mean±S.E.M

, 4 images per mouse per

group. One

way ANOVA, non

parametric Kruskal

Wallis test gave an approximate P

value of 0.0088

Transduction of brain tissue

AAV

PHP.B4, B

, A

PHP.X1 (ARQMDLS), and AAV

PHP.X2 (TNKVGNI) packaging

ssAAV:

CAG

mNeonGreen genome (

= 3,

1x10

vg IV dose/adult C57BL/6J mouse, 3 weeks of expression), matched in fluorescence intensity to AAV9 and AAV

PHP.V1 sagittal

brain images in Fig.

Transduction

of the brain by AAV

PHP.B8

using the

ssAAV:

CAG

mRuby2 genome (

= 3, 3x10

vg IV

dose

/adult

C57BL/6J mouse, 3 weeks of expression).

Transduction of AAV9 (left), AAV

PHP.X3 (QNVTKGV) (middle) and AAV

PHP.X4 (LNAIKNI

) (right)

vectors packaging

ssAAV:

CAG

2xNLS

EGFP

(

= 2, 1x10

vg IV dose/

adult C57BL/6J mouse, 3 weeks of

expression).

data is reported from one independent trial.

Supplementary Figure

Evolution

of the

AAV

PHP.B Capsid b

y Diversifying Amino Acid Positions 5

597.

Distributions of R1 and

R2 brain libraries (at AA level

, standard score

(SS)

RCs

sorted in decreasing order of scores)

is shown

The

for AAV

PHP.N and

AAV

PHP.

eB across libraries are mapped on the zoomed

in view of this plot (dotted li

ne box).

Heatmap

of AA distributions across the diversified region of the

enriched variants from R2 liver library (top 100 sequences)

normalized to the

virus (input library).

Clustering analysis of

enriched variants from GFAP

and Vglut2 brain libraries are shown with size of nodes

representing their relative

depletion

in liver, and the thickness of edges (connecting lines) representing their relative identity between

nodes.

Expression of AAV

PHP.B (above) and AAV

PHP.N (below) packag

ed with ssAAV:CAG

mNeonGreen across all organs is

shown (n

3, 3x10

vg IV dose per adult C57BL/6J mouse, 3 weeks of expression). The background auto fluorescence is in

magenta

Transduction of mouse brain by the AAV

PHP.N variant,

carrying the CAG promoter that drives the expression of mNeonGreen (

3, 1x10

vg IV dose per C57BL/6J adult mouse, 3 weeks of expression) is shown. Fluorescence

in situ

hybridization chain reaction

(FITC

HCR) was used to label excitatory neurons with

glut1

and inhibitory neurons with G

ad1

Few

cells

where

EGFP expression

localized

with

specific

cell markers

are

highlighted by asterisks symbol.

Supplementary Figure

Investigation o

f AAV

PHP Variants Across Diff

erent Mouse Strains

and

In Vit

ro Human Brain M

icrovascular Endothelial Cells.

Transduction of AAV9, AAV

PHP.eB and AAV

PHP.V1 in human brain microvascular endothelial cell culture (HBMEC) is shown. The

vectors were packaged with ssAAV:CAG

mNeongreen. T

he mean fluorescence intensity across the groups were quantified (n=3 tissue

culture wells of 0.95 cm

surface area per group, 3 images per well per group per dose was imaged after three days of expression,

doses 1x10

vg and 1x10

vg per 0.95 cm

surface

area). A two

way ANOVA with correction for multiple comparisons using Tukey’s

test gave adjusted P

value of 0.0051 for AAV9 vs PHP.V1, 0.0096 for PHP.eB vs PHP.V1, 0.8222 for AAV9 vs PHP.eB for 1x10

vg,

and 0.0052 for AAV9 vs PHP.V1, 0.0049 for PHP.eB vs

PHP.V1, 0.9996 for AAV9 vs PHP.eB for 1x10

vg (**P

≤ 0.01, is shown and P

> 0.05 is not shown on the plot; mean ± S.E.M., 95% CI).

The transduction of cortex brain region by AAV

PHP.B, AAV

PHP.C2 and

AAV

PHP.C3 across two different mouse strains: C57BL/6J and BALB/cJ are shown. The vectors were packaged

with ssAAV:CAG

mNeongreen (

= 2

3 per group, 1x10

vg IV dose/ adult mouse, 3 weeks of expression). The fluorescence intensity is matched across

all the images. The data reported in

a,b

are from one independent trial where all viruses were freshly prepa

red and tittered in the same

assay for dosage consistency, with additional validation for AAV

PHP.C2 and AAV

PHP.C3 in an independent trial for BALB/cJ.

SUPPLEMENTARY TABLES 1

Supplementary Table 1:

Comparison between the two methods for R2 selection.

The table summarizes the pros and cons of selection design parameters by the

synthetic

pool

and

PCR pool

R2 selection methods.

AAV

Variants

Synthetic pool

enrichment rank

PCR pool

enrichment rank

PCR pool

read count rank

PHP.V1

PHP.V2

PHP.B4

PHP.B7

PHP.B8

PHP.C1

PHP.C2

293

PHP.C3

Not recovered

Supplementary Table 2:

Ranking

of AAV

PHP capsids across methods.

Ranks of selected variants among all capsids recovered from

R2 Tek

Cre selection by

synthetic

pool

enrichment score (representing M

CREATE),

PCR

pool

enrichment score (representing

closer to M

CREATE), or

PCR pool

read counts (representing CREATE),

the highest ranks of

which starts from 1,

and

“Not recovered”

represent absence of the variant from R2 sequencing

data.

Design Parameters

Synthetic pool

design

PCR pool

design

Carryover of R1 selection bias among

variants

No, likelihood of false positives is low

Yes, potential to minimize by

normalization

Carryover of R1 selection induced

mutants

Yes

Confidence in

library performance

High, using alternate codon replicates

Low

Customize library or add internal

controls

Yes, in an unbiased manner

Yes, with greater risk of bias

Control library size

Yes, without reducing libraries or pooling

Yes, with libraries

reduced for pooling

Cost for R2 library generation

High

Low

AAV Variants

Reference / Selection

method

Tropism

Production

Rounds of evolution

from parent capsid

PHP.B, B2, B3

Deverman et al, 2016 /

CREATE

road CNS

transduction

Good

1 round from AAV9

PHP.A

Deverman et al, 2016 /

CREATE

strocyte transduction

Poor; prone to

precipitate upon

storage at 4

1 round from AAV9

PHP.eB

Chan et al, 2017 / CREATE

nhanced

road CNS

transduction

Good

2 rounds from AAV9

1 round from PHP.B

PHP.S

Chan et al, 2017 / CREATE

ensory neuron

transduction

Good

1 round from AAV9

PHP.V1,

Current study / M

CREATE

BBB

ascula

r cells

and

astrocytes

transduction

Good

1 round from AAV9

PHP.B4, B

, B

Current study / M

CREATE

road CNS

transduction

Good

1 round from AAV9

PHP.B

, B

Current study / M

CREATE

and CREATE

road CNS

transduction

Good

2 rounds from AAV9 or

1 round from PHP.B

PHP.C1, C2, C3

Current study / M

CREATE

Broad

NS transduction

across mouse strains

Good; PHP.C1 prone

to precipitate upon

storage at 4

1 round from AAV9

PHP.N

Current study / M

CREATE

and CRE

ATE

euron transduction

Average

2 rounds from AAV9 or

1 round from PHP.B

PHP

variants

–

named

in memory of

late Professor

aul

atterson

, Caltech

Supplementary Table 3: AAV

PHP vectors identified by CREATE and M

CREATE.

The table provides a

ummary of the variants that have been

identified

o far using CREATE

and M

CREATE, along with their tropism and the evolutionary steps from the parent capsid that

was involved in their discovery.

Supplementary Table 4:

Primers used in M

CREATE selection.

The table provides

a list of primers used in M

CREATE across the different steps of the

selection process as described in the Methods.

Primer name

equence

(5’

3’)

Forward or

reverse direction

ACTCATCGACCAATACTTGTACTATCTCTCTAGAAC

Forward

7xMNN

588i

GTATTCCTTGGTTTTGAACCCAACCGGTCTGCGCCTGTGCMNNMNNMNNMNNMNNMNNM

NNTTGGGCACTCTGGTGGTTTGTG

Reverse

588

R2lib

CACTCATCGACCAATACTTGTACTATCTCTCT

Forward

588

R2lib

GTATTCCTTGGTTTTGAACCCAACCG

Reverse

mer

588i

GTATTCCTTGGTTTTGAACCCAACCGGTCTGCGCXXXXXXMNNMNNMNNMNNMNNMNNMN

NXXXXXXACTCTGGTGGTTTGTG

Reverse

71F

CTTCCAGTTCAGCTACGAGTTTGAGAAC

Forward

CDF/R

CAAGTAAAACCTCTACAAATGTGGTAAAATCG

Forward/Reverse,

see Methods

588i

lib

PCR1

6bpUID

CACGACGCTCTTCCGATCTAANNNNNNAGTCCTATGGACAAGTGGCCACA

Forward

588i

lib

PCR1

GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTTCCTTGGTTTTGAACCCAACCG

Reverse

1527

ACACTCTTTCCCTACACGACGCTCTTCCGATCTGACAAGTGGCCACAAACCACCAG

Forward

1532

GTGACTGGAGTTCAGACGTGTGCTCTTCCGATCTCCTTGGTTTTGAACCCAACCG

Reverse

mNeonGreen

CGACACATGAGTTACACATCTTTGGCTC

Forward

mNeonGreen

GGAGGTCACCCTTGGTGGACTTC

Reverse

Mito

CCCAGCTACTACCATCATTCAAGT

Forward

Mito

GATGGTTTGGGAGATTGGTTGATGT

Reverse

CapF

ATTGGCACCAGATACCTG ACTCGTAA

Forward

Cre

GTCCAAACTCATCAATGTATCTTATCATGTCTG

Reverse

NGS

AATGATACGGCGACCACCGAG

Forward

NGS

CAAGCAGAAGACGGCATACGA

Reverse