CRISPR Gene Editing in Stem Cells

and 6I). All clones showed a marked decrease of the CD41a

CD45 population (Figure 6I). These phenotypes reflect progres-

sion to low-risk MDS driven by

GATA2

inactivation and to high-

risk MDS driven by the del(7q), suggesting that the del(7q) is a

marker of adverse prognosis in the clinic.

We then set to model disease progression along sequential

stages driven by the stepwise acquisition of genetic lesions

starting from a normal cell. To this end, we first introduced trun-

cating mutations in the

ASXL1

gene and selected two clones

with monoallelic truncations (Figures 6J and S7D).

ASXL1

C-ter-

minal truncation is an early event in myeloid malignancies and

one of the most common mutations in individuals with CHIP

(Link andWalter, 2016; Steensma et al., 2015). We subsequently,

in a second step, deleted one copy of chr7q in one of the

ASXL1

engineered clones (Figures 6J and S7E–S7G). This set of clones

recapitulated stepwise progression from normal to preleukemia

(

ASXL1

mutation) to high-risk MDS (

ASXL1

mutation + del7q),

assessed by CD45 and CD90 marker expression, cell viability,

and colony formation (Figures 6K–6N).

These results collectively show that our phenotypic roadmap

can be used to model disease stage transitions across the spec-

trum of myeloid malignancy driven by a variety of genetic lesions

and their combinations.

Modeling Disease-Stage-Specific Effects of

Therapeutic Interventions

5-Azacytidine (5-AzaC) is a hypomethylating agent that is used

as first-line therapy in MDS. 30%–50% of MDS patients show

some response, but there are currently limited biomarkers to

predict the responders (Bejar and Steensma, 2014). Further-

more, the mechanism by which 5-AzaC exerts its therapeutic ef-

fects is not clear. Potential mechanisms may include induction of

differentiation or preferential inhibition of the growth of the MDS

clone. To first test for potential effects of 5-AzaC in inducing dif-

ferentiation, we cultured HPCs derived from the different iPSC

lines in methylcellulose in the presence or absence of 5-AzaC

(Figure 7A). Strikingly, treatment with 5-AzaC resulted in a

marked rescue of BFU-E and CFU-GEMM colonies in low-risk

MDS-iPSCs (Figures 7B, S7H, and S7I). In contrast, it had no ef-

fect in colony growth from normal iPSCs or any other iPSC line

from other disease stages. We then tested for selective effects

in the growth of the MDS clone using a competitive growth

assay. Intriguingly, 5-AzaC had an inhibitory effect in the growth

of high-risk MDS-iPSC-derived HPCs, but not of those derived

from other disease stage iPSCs or normal iPSCs (Figures 7C

and 7D). These results suggest that 5-AzaC may primarily affect

differentiation in earlier stages of the disease, whereas its main

therapeutic action later on might be mediated through selective

inhibition of the MDS clone. DNA methylation analysis of low-risk

MDS-iPSC-derived HPCs (MDS-1.12 line) treated with 5-AzaC

for 3 days revealed striking genome-wide hypomethylation

following 5-AzaC treatment, which included gene promoters,

suggesting that hypomethylation may underlie the rescue of col-

ony formation in these cells (Figures 7E and S7J).

To further test for stage-specific drug responses, we treated

HPCs derived from two MDS/AML lines from patient 4, capturing

a less and a more advanced disease stage, the AML-4.24 line

derived from the dominant clone, and the AML-4.10 line derived

from the KRAS mutated subclone (Figure 1) with rigosertib, a

small-molecule inhibitor of RAS signaling pathways that is

currently in clinical trials for high-risk MDS (Athuluri-Divakar

et al., 2016). As predicted, AML-4.10 HPCs showed marked

sensitivity to rigosertib, whereas AML-4.24 cells were marginally

affected (Figure 7F). These results collectively support the use of

our disease progression model in drug testing.

DISCUSSION

Here, we used an approach integrating cell reprogramming and

cancer genetics to establish iPSC lines representative of distinct

stages during the cellular transformation from normal cells to

AML through an MDS stage. Detailed genetic and clonal charac-

terization of the starting cell population and the derived iPSC

lines allowed us to make additional observations regarding the

degree to which the output of reprogramming is representative

of the clonal composition of the primary cells. Our results show

that the clonal representation of the original cells in the iPSCs

is skewed, often in favor of residual normal cells over cells

of the premalignant or malignant clone (Table S1). They also

show that it is reprogramming per se and not the in vitro stimu-

lation and expansion that accounts for this bias, which seems

to be conferred by some MDS- and AML-associated genetic le-

sions, but not others, while some genetic abnormalities seem to

be incompatible with reprogramming (Figures S1B–S1E). Among

the ones tested here, del(5q) and monosomy 7 could never be

captured in iPSCs, despite cells harboring them comprising

over 80% of the starting cell pool. It might be possible to over-

come this refractoriness by using alternative reprogramming fac-

tor cocktails, which we did not test here. A negative or positive

impact of specific cancer-associated gene mutations on the re-

programming ‘‘fitness’’ of the cells would not be surprising given

well-studied positive and negative effects, respectively, of TP53

inactivation and Fanconi anemia pathway mutations on reprog-

ramming (Papapetrou, 2016). Importantly, despite the skewed

clonal and subclonal representation, we were able to capture

normal and preleukemic cells, as well as malignant clones and

subclones, and thus compile a panel of lines carrying genomes

representative of different disease stages from normal to fully

(C) Schematic of growth competition assay to test the effects of 5-AzaC in cell proliferation relative to normal cells. The cells were mixed 1:1 with the N-2.12 line

stably expressing GFP at day 9 of hematopoietic differentiation in the presence or absence of 5-AzaC and followed for an additional 2 days by flow cytometry.

(D) The relative population size was calculated as the percentage of GFP

cells in the treated cells relative to the percentage of GFP cells in the untreated cells at

each time point. iPSC lines from left to right: N-2.12, N-3.10, MDS-1.12, MDS-2.13, and AML-4.24.

(E) Volcano plot showing differences in DNA methylation in four HPC samples independently treated with 5-AzaC derived from the MDS-1.12 line in two inde-

pendent differentiation experiments compared to two untreated controls.

(F) HPCs derived from the AML-4.24 and the AML-4.10 iPSC lines treated with rigosertib. The relative population size was calculated as the number of treated

cells relative to the number of untreated cells at each time point. Mean and SEM from triplicate experiments are shown.