Science - USA (2022-05-27)

factors, hormones, cytokines, inflammation, and
stresses. It controls the expression of many
downstream genes related to cell division,
apoptosis, cell migration, and immunity ( 30 ).
In addition to subtype-specific TF identi-
fication, we also predicted the important TFs

in a sample-specific manner. To do this, we

analyzed all TFs’relative expression and chro-

matin accessibility at their motifs for samples

in CRPC-WNT, CRPC-NE, and CRPC-SCL, and

compared these to the average for CRPC-AR

samples. This analysis demonstrated that the

key TFs identified in a subtype-specific man-

ner agree with the sample-specific results (fig.

S9). We also found significant correlation be-

tween the expression of key TFs and acces-

sibility at their motifs, which suggests that

they likely exhibit pioneering activity (fig. S10,

Tanget al., Science 376 , eabe1505 (2022) 27 May 2022 4of13

A

C

B

D

ATAC-seq group

Group 1

Group 2

Group 3

Group 4

AR expression

AR high

AR low

AR neg

Pathway

UMAP for RNA−seq data GSEA enrichment scores: one vs. others

OncoPrint

MSKPCa2

22Rv

1

LNCaPVCaPPDX09aPDX10aPDX16aPDX16b

MSKPCa19MSKPCa22

PDX34aMSKPCa1

MSKPCa16WCM1078WCM1262

MSKEF1

MSKPCa10MSKPCa14

WCM154

H660

MSKPCa24

PDX01a

MSKPCa11MSKPCa12MSKPCa13MSKPCa15MSKPCa17MSKPCa3MSKPCa8MSKPCa9

WCM155

DU145

PC3

MSKPCa18MSKPCa20

ATAC-seq group

ETS_fusion

20 10 0

23%

66%

23%

46%

20%

11%

6%

6%

14%

14%

14%

6%

14%

20%

9%

9%

17%

3%

9%

11%

9%

14%

14%

29%

3%

AR

TP53

RB1

PTEN

FOXA1

NCOR1

AKT1

PIK3CB

PIK3CA

PIK3R1

BRAF

KRAS

APC

CTNNB1

TCF7L2

MSH2

BRCA2

BRCA1

ATM

ATR

FANCA

CDKN1B

CDKN2A

ZFHX3

SPOP

Alternations

Deep deletion

Amplification

Inframe Mutation (putative driver)

Inframe Mutation (unknown significance)

Missense Mutation (putative driver)

Missense Mutation (unknown significance)

Truncating Mutation (putative driver)

ETS_fusion

Negative

Positive

AR-

associated

pathwPI3K

ay

RASRAF

Wnt

pathw

ay

DNA

damage repair

cyclecell

SPOP

ATAC-seq group

CRPC-AR

CRPC-WNT

CRPC-NE

CRPC-SCL

Lost at RNA level (RNA-seq)

Lost at protein level (WB)

CRPC-AR CRPC-WNT

CRPC-NE CRPC-SCL

−0.5 0.0 0.5 −0.5 0.0 0.5

−0.5 0.0 0.5 −0.5 0.0 0.5

LIM_MAMMARY_

STEM_CELL_UP

BELTRAN_

NEPC_UP

HALLMARK_WNT_BETA_

CATENIN_SIGNALING

TCGA_ANDROGEN_

RESPONSE

Enrichment Score

LIM_MAMMARY_

STEM_CELL_UP

BELTRAN_

NEPC_UP

HALLMARK_WNT_BETA_

CATENIN_SIGNALING

TCGA_ANDROGEN_

RESPONSE

Enrichment

neg_enrich(padj < 0.05)

not_sign(padj > 0.05)

pos_enrich(padj < 0.05)

Pathway

(z-score)

Gene expression

Fig. 2. Transcriptomic and genomic characterization of the four CRPC
subtypes defined by ATAC-seq.(A) Unsupervised UMAP on the mRNA
expression values of the 1000 most variably expressed genes across
all samples. (B)EnrichmentscoresandP values from GSEA indicate that
the four signals are significantly positively enriched in specific subtypes but not
in others. (C) Heatmap shows the relative expression of subtype-specific
marker genes and basal/luminal genes across all samples. The ATAC-seq

group and signature scores of the four representative pathways for each

sample are shown at the top. (D) OncoPrint shows the genomic alterations of

the 35 samples with DNA-sequencing data. MSKPCa8 to 20, 22, and 24

were sequenced with MSK-IMPACT. The SNVs and CNVs for cell lines (LNCaP,

22Rv1, VCaP, H660, DU145, and PC3) were collected from CCLE (Cancer

Cell Line Encyclopedia). Whole-exome sequencing (WES) was performed for

other samples.

RESEARCH | RESEARCH ARTICLE