myelomonocytes (fig. S1A). In wild-type zebra-
fish, hematopoiesis is polyclonal and appears
multicolored ( 6 ) (fig. S1B). Hematopoiesis in
runx1homozygous mutants was driven by
fewer larger clones, consistent with an oligo-
clonal state resulting from a smaller HSPC pool
(fig. S1, B and C). Thus, Zebrabow can detect an
aberrant clonal composition in hematopoiesis
caused by a germline genetic mutation.
We next sought to determine how acquired
mutations affect clonal composition. We com-
bined mosaic mutagenesis using CRISPR-Cas9
with Zebrabow labeling, a technique we call
TWISTR (tissue editing with inducible stem
cell tagging via recombination). Embryos of
Tg(ubi:Zebrabow-M)andTg(drl:CreERT2)crosses
were injected with a pool of guide RNAs (gRNAs)
targeting zebrafish orthologs of genes implicated
in clonal hematopoiesis and two control genes
(Fig. 1A and table S1). HSPC color labeling was
induced by 4-OHT at 24 hpf. We hypothesized
that a clonal outgrowth due to enhanced fit-
ness would be detectable by a dominant single
color. Although zebrafish injected with control
gRNAs had normal multicolored adult hema-
topoiesis, mutant zebrafish often exhibited a
large single-colored cluster (Fig. 1, B and C),
with increased average size of the largest clus-
ter (fig. S2, A and B). Cluster size in mutant
zebrafish was more variable than that in
controls, consistent with abnormal clone size
distribution (fig. S2C). We used the Gini co-
efficient as a measure of inequality ( 8 ) and
found that individual mutant zebrafish had
higher Gini coefficients than controls, indicat-
ing greater clonal skewing (Fig. 1D). There
were no significant differences in hemato-
poietic cell types or morphology between co-
horts ( 9 ) (fig. S2, D and E). These results show
that TWISTR can induce and detect clonal
dominance in the presence of acquired muta-
tions without disturbance of hematopoiesis.
We determined the mutational landscape in
TWISTR mutants to characterize genetic driv-
ers of clonal fitness. Although embryos were
injected at the one-cell stage, translation of
Cas9mRNA delays mutagenesis, generating
a mosaicism of wild-type and mutant alleles
( 10 , 11 ). We collected serial peripheral blood
samples starting at 3 months and marrow sam-ples at 8 to 9 months (fig. S3A). As a control for
overall gRNA targeting, we collected fin tissue
largely composed of nonhematopoietic cells.
Sequencing showed mutations in all targeted
genes, with variable numbers of mutations
per zebrafish (fig. S3, B and C). Although all
gRNAs were tested individually to confirm
target editing and gRNA efficiency was con-
cordant across tissues, the numbers of detected
edits varied between genes (fig. S3, D and E).
This could be due to gene-specific selection or
may represent differences in pooled gRNA
efficiencies. Comparing types of insertions/
deletions (indels) for a given gene in marrow
(hematopoietic) versus fin (nonhematopoietic)
could show tissue-specific selection. We found
that disruptive frameshift mutations were more
likely to occur in marrow than in fin forasxl1,
tet2, andtp53(fig. S4A). These results demon-
strate multiplexed mutagenesis with TWISTR
and reveal that frameshift alleles of specific
genes undergo positive selection in marrow.
To examine temporal changes of variant
allele fractions (VAFs) in mutant zebrafish,
we sequenced and tracked unique variants inSCIENCEscience.org 5 NOVEMBER 2021¥VOL 374 ISSUE 6568 769
Fig. 2. Multiplex mosaic muta-
genesis and Zebrabow labeling
reveal mutant allele dynamics
and identify drivers of clonal
hematopoiesis.(A) Plots of
frameshift indel variant allele
fractions (VAFs) in marrow
cells and Gini coefficient in
myelomonocytes of mutant zebra-
fish. P< 0.02 (SpearmanÕs rank
correlation). (B) Sorting strategy
of individual clusters. (C) Tiling plot
of alleles in sorted clusters from
29 mutant zebrafish. Columns
correspond to single zebrafish in
the same order in each set.
Percent of frameshift (fs) VAF
containing clusters is shown.
P< 0.001 (c^2 test).
AGini coefficientFrameshift VAF (%)B
Dominant
clustersC
Non-dominant
clusters
32
64
61
21
25
21
14250
000144asxl1*
tp53
dnmt8
zrsr2
piga
ezh2
tet2stag2adnmt4
dnmt6ezh1dnmt5rad21btet379
69
62
38
31
17
17103
3037321
7dnmt7
rad21a14
14dnmt3 (^77)
Frameshift In-frame No alteration
tp53
p=0.002
stag2a
p=0.462
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
dnmt8
p=0.015
zrsr2
p=0.156
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
0.2 0.4 0.6 0.8 1
0
20
40
60
80
(^100) piga
p=0.863^
tet2
p=0.227
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
asxl1
*p=0.0039^
ezh2
p=0.147
0.2 0.4 0.6 0.8 1
0
20
40
60
80
100
(^0) 0.2 0.4 0.6 0.8 1
20
40
60
80
100
B
R G
sorted
non-dominant
sorted
dominant
% fs % fs
0
10
20
30
40
50
60
Cluster size (%)
No data
RESEARCH | REPORTS