Nature - USA (2020-09-24)

606 | Nature | Vol 585 | 24 September 2020

Article

In addition, exome sequencing revealed only one non-synonymous
somatic mutation in FR2 cells: a small frameshift deletion in TLR7,
which was not a hit in our CRISPR screens and has no reported
ferroptosis-relevant functions (Supplementary Data 8). However, RNA
sequencing revealed that, among 87 known peroxisome and ether lipid
biosynthesis genes, AGPS and TMEM189 (also known as PEDS1) were
significantly downregulated in FR2 cells—a decrease that was validated
at the protein level (Fig. 3i, j, Extended Data Fig. 10j, k, Supplementary
Data 8). The expression levels of other known ferroptosis-modulating
genes—including ACSL4, LPCAT3 and AIFM2 (which encodes FSP1)^22 ,^23 —
were not significantly altered (Extended Data Fig. 10k, l).
TMEM189 encodes a 1-O-alkyl-PE desaturase that converts 1-O-alkyl
ethers into 1-O-alkenyl ethers^24 , and displays a co-dependency with
ACSL4, PEX3 and FAR1 in the cancer dependency map^25 (Extended
Data Fig. 11a). However, TMEM189 did not register as a significant hit
in our CRISPR screens. In addition, perturbing TMEM189 levels using
CRISPR–Cas9, shRNA or cDNA did not significantly alter the sensitiv-
ity to ferroptosis in the cell lines tested (Extended Data Fig. 11b–g),
implying that the pro-ferroptotic role of PUFA-ePLs is independent of
the double bond present in the alkenyl-ether linkage. By contrast, we
found that AGPS inactivation—which depletes both 1-O-alkyl-lipids
and 1-O-alkenyl-lipids—is sufficient to restore the growth of GPX4−/−
tumours. This suggests that the spontaneous downregulation of

AGPS—but not TMEM189—contributes to the observed emergence of

ferroptosis resistance in the ccRCC model.

We then explored whether the ferroptosis-sensitizing role of

PUFA-ePLs is relevant in non-neoplastic settings. We chose to look

at neurons and cardiomyocytes as major cell types from the brain

and heart, respectively—vital organs that exhibit high ePL levels^18 and

have been reported to undergo ferroptosis under certain pathologi-

cal conditions^26 ,^27. In the SH-SY5Y neuronal differentiation model^28 ,

we found that differentiated neurons exhibited higher sensitivity

to GPX4-inhibition-induced lipid peroxidation and ferroptosis than

the parental cells, a difference that was associated with upregulated

PUFA-ePEs and PUFA-ePCs in the neurons (Fig. 4a–c, Extended Data

Fig. 12a–e, Supplementary Data 9).

We also found that cardiomyocytes derived from human induced

pluripotent stem cells (iPS cells) were more sensitive to GPX4 inhi-

bition than were cardiac progenitors derived from iPS cells (Fig. 4d,

Extended Data Fig. 12f, g, Supplementary Video 2). Cell death with

ferroptosis-like morphology was seen in cardiomyocytes treated

with ML210, and was prevented by co-treatment with liproxstatin-1

or Fer-1, but not by co-treatment with the  necroptosis inhibitor

necrostatin-1 or the apoptosis inhibitor z-VAD-FMK (Extended Data

Fig. 12h, i, Supplementary Video 3). Our lipidomic analysis further

revealed that cardiomyocytes, in comparison with cardiac progenitors,

cd

020406080

0

200

400

600

800

Time after injection (d)

786-O-Cas9 xenografts

GPX4+/+

GPX4–/–

β-Actin

GPX4

GPX4–/–

GPX4+/+ FS #a #b #c #d

Ferroptosis-resistant (FR1)

–1.5 –1.0 –0.5 0 0.5 1.0 1.5 2.0

0

2

4

6

8

log 2 (FR2#d/WT)

log 2 (FR2#d/WT)

log 2 (FR2#d/WT)

Other lipids

PUFA-ePC/ePE

PUFA-TAGs

C52:7 TAG

C54:9 TAG

C54:8 TAG

C56:10 TAG

C56:9 TAG

C58:11 TAG

LPC/LPEs

C38:3 ePE

C38:7 ePCC40:7 ePCC38:7 ePE

C36:4 ePC

C38:6 ePE C14:0 LPC

C16:1 LPC

C16:0 LPC

C18:2 LPC

C18:1 LPC

C20:5 LPC

C20:0 LPE

ef

g

GPX4

β-Actin

(^001428425670)
500
1,000
Time after injection (d)
786-O xenograft
GPX4+/+WT (1 × 106 cells)
GPX4–/– FR1#d (5 × 106 cells)
GPX4–/– FR1#a (5 × 106 cells)
GPX4–/– FS (5 × 106 cells)
–2 –1 012
–15
–10
–5
0
TMEM189
AGPS
ABCD1
LR
Tumour-isolated cells (786-O)
GPX4+/+(WT)FR2#d
LRLR
FR2#a
37
25
TMEM189
100
AGPS 75
50
β-Actin 37
kDa
h
i j
–4 –2 024
0
2
4
6
8
Free fatty acids
C20:2n6
C20:4n6
C20:3n6
C22:5n6
C22:6n6
C20:5n3
a
010203040
0
100
200
300
400
500
Time after injection (d)
Tumour volume (mm
3 )
Tumour volume (mm
3 )
Tumour volume (mm
3 )
OVCAR-8 xenograft
024487296
0
2
4
6
8
10
Time after Fer-1 withdrawal (h)
Relative viability
OVCAR-8
GPX4–/– + sgNC
GPX4–/– + AGPS sg1
GPX4–/– + AGPS sg2
GPX4–/– + FAR1 sg2
GPX4–/– + PEX3 sg1
GPX4–/– + PEX3 sg2
GPX4–/– + PEX10 sg1
GPX4–/– + PEX10 sg2
GPX4+/++ sgNC
b
GPX4GPX4+/+–/– + sgNC+ WT
GPX4–/– + AGPS sg2
GPX4GPX4–/––/– + + FAR1PEX3 sg2 sg1
GPX4–/– + PEX10 sg1
–log
(adj. 10
P)
–log
(adj. 10
P)
log
(adj. 10
P)
LR
Tumour-isolated cells (786-O)
GPX4+/+(WT)FR2#d
LRLR
FR2#a
Fig. 3 | Cancer cells initially dependent on GPX4 downregulate PUFA-ePLs
to evade ferroptosis. a, Relative viability of the indicated OVCAR-8 cells after
Fer-1 withdrawal. n = 8 biologically independent samples. GPX4+/+-sgNC (black
dashed curve) was analysed independently and included as a reference.
P values for 90-h viability of GPX4−/− derivatives are as follows: GPX4−/−-sgNC
versus GPX4−/− AGPS sg1, P = 9.07 × 10−8; versus GPX4−/− AGPS sg2, P = 9.66 × 10−1 5;
versus GPX4−/− FA R 1 sg2, P = 3.79 × 10−5; versus GPX4−/− PEX3 sg1, P = 5.9 × 10−1 2;
versus GPX4−/− PEX3 sg2, P = 2.92 × 10−17; versus GPX4−/− PEX10 sg1, P = 1.99 × 10−16;
versus GPX4−/− PEX10 sg2, P = 3.53 × 10−1 3. b, Tumour growth rates of the
indicated OVCAR-8 xenografts. GPX4−/−-sgNC, n = 8 mice; other conditions,
n = 5 mice. P values for day-37 tumour sizes are as follows: GPX4+/+ versus
GPX4−/−-sgNC, P = 4.03 × 10−7. GPX4−/−-sgNC vs GPX4−/− AGPS sg2, P = 1.15 × 10−4;
versus GPX4−/− FA R 1 sg2, P = 0.0446; versus GPX4−/− PEX3 sg1, P = 2.13 × 10−5;
versus GPX4−/− PEX10 sg1, P = 0.0197. c, Tumour growth rates of the indicated
786-O xenografts. n = 5 mice. For day-27 tumours, GPX4+/+ versus GPX4−/−,
P = 1.4 × 10−8. d, Immunoblot showing GPX4 levels in GPX4+/+ 786-O cells, the
original ferroptosis-sensitive (FS) GPX4−/− clone, and the first round of
ferroptosis-resistant (FR1) cells. e, Tumour growth curves of the indicated
786-O xenografts. n = 5 mice. P values for day-30 tumours are as follows: GPX4−/−
FR1#a versus FS, P = 1.96 × 10−8; GPX4−/− FR1#d versus FS, P = 3.42 × 10−7.
f, Immunoblot showing GPX4 levels in FR2 cells. FR2#a, from a FR1#a tumour;
FR2#d, from a FR1#d tumour. L/R, left/right tumour. g, Volcano plot of the
lipidomic analysis comparing GPX4−/− FR2#d and GPX4+/+ 786-O cells. LPE,
lysophosphatidylethanolamine; LPC, lysophosphatidylcholine; TAG,
triacylglycerol. n = 6 biologically independent samples. h, Volcano plot of the
free fatty-acid lipidomic analysis comparing GPX4−/− FR2#d and GPX4+/+ 786-O
cells. n = 6 biologically independent samples. i, Volcano plot showing the
relative mRNA expression (analysed by RNA sequencing) of peroxisome and
ether-lipid-metabolism-related genes comparing GPX4−/− FR2#d and GPX4+/+
786-O cells. n = 4 biologically independent samples. See Methods for data
analysis methods. j, Immunoblot showing TMEM189 and AGPS levels in the
indicated cells. For immunoblots, β-actin was used as a loading control.
a, c, d, f, j show representative results of experiments performed twice.
See Supplementary Information for uncropped immunoblot images. Data are
mean ± s.d. P values were calculated using two-tailed Student’s t-tests.
Multiple-testing adjustment was performed using the Benjamini–Hochberg
method. Two tumours were injected per mouse in animal experiments.