Abstract
The integrated stress response (ISR) enables cells to cope with a
variety of insults, but its specific contribution to downstream
cellular outputs remains unclear. Using a synthetic tool, we
selectively activate the ISR without co-activation of parallel pathways
and define the resulting cellular state with multi-omics profiling. We
identify time- and dose-dependent gene expression modules, with ATF4
driving only a small but sensitive subgroup that includes amino acid
metabolic enzymes. This ATF4 response affects cellular bioenergetics,
rerouting carbon utilization towards amino acid production and away
from the tricarboxylic acid cycle and fatty acid synthesis. We also
find an ATF4-independent reorganization of the lipidome that promotes
DGAT-dependent triglyceride synthesis and accumulation of lipid
droplets. While DGAT1 is the main driver of lipid droplet biogenesis,
DGAT2 plays an essential role in buffering stress and maintaining cell
survival. Together, we demonstrate the sufficiency of the ISR in
promoting a previously unappreciated metabolic state.
Subject terms: Stress signalling, Organelles, Transcriptomics,
Metabolomics, Lipids
__________________________________________________________________
ISR-specific contributions to stress-induced cellular outputs are not
well understood. Here,
authors use a minimal activation system and multi-omics to define an
ISR-sufficient metabolic state that includes protective accumulation of
lipid droplets.
Introduction
In response to environmental challenges as diverse as viral infection,
mitochondrial dysfunction, amino acid deprivation, or the accumulation
of misfolded proteins, eukaryotic cells engage a common signaling
mechanism called the Integrated Stress Response (ISR)^[46]1,[47]2. This
highly conserved pathway limits the rate of protein synthesis while
simultaneously inducing a stress-responsive gene expression program
(Fig. [48]1a). Four kinases (PERK, HRI, GCN2, and PKR) detect a variety
of insults and trigger the ISR through a shared mechanism, the
phosphorylation of the alpha subunit of eukaryotic translation
initiation factor 2 (eIF2)^[49]3. eIF2 is a central participant in mRNA
translation, bringing the methionyl-tRNA ternary complex (eIF2-GTP-Met
tRNAi) to the ribosome to initiate protein synthesis. Upon initiation
of translation, GTP is hydrolyzed. eIF2 is then recharged with GTP to
enable a new round of initiation by its own dedicated guanine
nucleotide exchange factor (GEF), eIF2B^[50]4,[51]5. When eIF2ɑ is
phosphorylated on serine 51 by any of the four eIF2 kinases, it binds
to eIF2B through a different interface, inhibiting its enzymatic
activity^[52]6,[53]7. This decrease in GEF activity results in reduced
mRNA translation and constitutes a major brake on protein synthesis in
stressed cells.
Fig. 1. Dimerizable PERK enables tunable and selective control of the ISR.
[54]Fig. 1
[55]Open in a new tab
a Schematic illustrating the inputs and outputs of the ISR. The ISR
incorporates input from various cellular stress sensors to increase the
levels of p-eIF2, the central input into the pathway. Its major outputs
include a reduction in protein synthesis, the formation of SGs, and the
induction of a specialized gene expression program that is in part
mediated by the translational induction of the transcription factor
ATF4. b Schematic illustrating the pharmacogenetic Dmr-PERK tool. Upon
addition of a small molecule dimerizer, Dmr-PERK phosphorylates eIF2
and activates the ISR. c AlphaLISA measurement of the ratio of
p-eIF2/eIF2 in Dmr-PERK U2OS cells following 2 h of dimerizer treatment
at the indicated concentrations. Error bars show mean ± SD of three
technical replicates. d Quantification of OP-Puromycin incorporation
(left axis) and G3BP1-positive SGs (right axis) co-labeled in Dmr-PERK
cells following 4 h of dimerizer treatment at the indicated
concentration. Error bars depict mean ± SD of three technical
replicates e Heatmap of protein coding gene expression in untreated
(UT) Dmr-PERK cells or in cells treated with dimerizer (Dmr, 0.2 nM),
thapsigargin (Tg, 100 nM), or arsenite (Ars, 0.05 mM), for the
indicated amount of time. Columns show the mean expression of three
replicates relative to the mean expression at t = 0 on a log[2] scale.
f Heatmap of genes described in e that have a significantly different
time-dependent response between dimerizer (Dmr) and thapsigargin (Tg)
treatments. Selected GO terms with significant enrichment by ORA within
the groups of Tg-responsive genes are indicated. g Heatmap of genes
described in e that have a significantly different time-dependent
response between dimerizer (Dmr) and arsenite (Ars) treatments.
Selected GO terms with significant enrichment by ORA within the groups
of arsenite-responsive genes are indicated. a, b are created with
BioRender.com released under a Creative Commons
Attribution-NonCommercial-NoDerivs 4.0 International license
[56]https://creativecommons.org/licenses/by-nc-nd/4.0/deed.en. Source
data are provided as a [57]Source Data file.
Concomitant with the general inhibition of protein synthesis, the ISR
facilitates the translation of a subset of specialized mRNAs. These
transcripts carry short upstream open reading frames (uORFs) in their
5’ untranslated region (5’UTR) that inhibit translation of the main
open reading frame (ORF) when levels of ternary complex are high but
result in translation of the ORF when ternary complex is low. A key
transcript regulated in this manner encodes ATF4, a bZIP transcription
factor that induces the expression of a wide range of genes, including
metabolic genes that promote amino acid accumulation and glutathione
synthesis^[58]2,[59]8–[60]11. ATF4 is a constitutively expressed
transcript across all cell types^[61]12 and thus, its translation is a
first line of defense when cells encounter an unfavorable environment
in mammalian tissues. Other uORF containing mRNAs include DDIT3, which
encodes the transcription factor CHOP (C/EBP-homologous protein) and is
thought to promote apoptosis during persistent ISR
activation^[62]13,[63]14, and GADD34, a regulatory subunit of the
phosphatase PP1, which dephosphorylates eIF2 and acts as an important
negative feedback loop to derepress translation
initiation^[64]15,[65]16. The reduction in translation initiation also
leads to the formation of stress granules (SG), membraneless organelles
that are formed by the cytosolic assembly of untranslated mRNAs and
their RNA-binding proteins^[66]17–[67]19.
While these molecular and cellular outputs of the ISR are well
documented, it is not known how they are coordinated in response to the
strength and duration of input to the pathway. Given that timing and
signal intensity can determine whether the ISR promotes cell survival
or death, it is critical to define this input-output
relationship^[68]1. Characterization of the outputs of the ISR is
complicated by the fact that it is typically activated as part of a
broader cellular response to stressors that engage multiple signaling
pathways. For example, the accumulation of unfolded proteins in the ER
activates a network of signaling pathways collectively termed the
unfolded protein response (UPR). In addition to activating the eIF2
kinase PERK, unfolded proteins induce two other transmembrane ER
sensors, IRE1 and ATF6 that trigger their respective signaling
programs^[69]20,[70]21. Similarly, studies on ISR-dependent metabolic
responses typically rely on hypoxia, nutrient stresses or mitochondrial
perturbations, which trigger multiple response pathways that have
pleiotropic metabolic consequences^[71]22–[72]26.
To address these challenges, we generated a cell line expressing a
synthetic construct^[73]27–[74]29 that allowed us to selectively
initiate the ISR in a tunable fashion. We harnessed this system to
quantitatively explore the transcriptome, metabolome, and lipidome over
time and define ISR-specific and -sufficient responses. We found that
ATF4, in addition to its well-documented role in promoting amino acid
and glutathione synthesis, inhibits carbon utilization by the TCA cycle
and promotes the reductive carboxylation of glutamine-derived
α-ketoglutarate to support amino acid synthesis. We also found that the
ISR was sufficient to drive an ATF4-independent reorganization of
cellular lipid content that promotes triglyceride (TG) and cholesterol
ester (CE) accumulation in lipid droplets (LDs). Although TG synthesis
by DGAT1 was the main contributor to LD formation, we found that DGAT2
was required to buffer ISR activity and maintain cell viability. These
responses occurred early and at low levels of pathway input, and thus,
we propose that a primary output of the ISR is a metabolic state that
acts on amino acid, central carbon and lipid metabolism.
Results
Dmr-PERK provides specific and tunable control of ISR activation
To specifically profile the cellular effects of ISR signaling, we
generated a U2OS-derived cell line stably expressing a synthetic
construct, dimerizable PERK (Dmr-PERK), consisting of the cytosolic
eIF2ɑ kinase domain-containing portion of mouse PERK fused to a
chemically inducible DmrB dimerization domain^[75]28. Upon addition of
ligand AP20187 (dimerizer), the fusion protein dimerizes, leading to
its activation and phosphorylation of eIF2ɑ (Fig. [76]1b). These cells
showed a low baseline level of p-eIF2 that increased with
dose-dependent addition of dimerizer (Fig. [77]1c) and initiated
canonical downstream signaling events. Inhibition of protein synthesis,
as measured by OP-Puromycin incorporation, was dose-dependent and
sensitive to Dmr-PERK activation, occurring at even the lowest
concentrations of dimerizer (0.04 nM) (Fig. [78]1d). In contrast,
accumulation of SGs, as detected by anti-G3BP1 staining, was only
measurable at much higher doses (EC50 of 4.7 nM) (Fig. [79]1d),
consistent with visible G3BP1-positive structures only appearing when
inhibition of mRNA translation is significantly reduced^[80]18.
Translation of uORF-containing transcripts was also sensitive to
minimal levels of Dmr-PERK activation, with measurable accumulation of
ATF4, GADD34, and CHOP protein at low concentration (0.01 nM), and
further increases with higher doses (Supplementary Fig. [81]1a, b).
Protein expression of ATF4 increased within 1 h of Dmr-PERK activation,
even at low doses of dimerizer (0.04 nM), likely reflecting the rapid
translation of pre-existing mRNA species (Supplementary Fig. [82]1c).
This tunable system therefore allows dose- and time-dependent control
of ISR activation.
To evaluate how the transcriptional response elicited by this minimal
system compared to that of commonly used pleiotropic ISR-inducing
agents, we performed RNA-seq on Dmr-PERK cells treated with a time
course of dimerizer (0.2 nM), thapsigargin (100 nM), a SERCA inhibitor
that induces ER stress by depleting its Ca^++ stores, or sodium
arsenite (0.05 mM), which generates ROS and causes oxidative stress.
For each compound, we chose a dose that elicited comparable levels of
p-eIF2ɑ (Supplementary Fig. [83]1d). Consistent with more targeted ISR
activation, triggering Dmr-PERK induced only a subset of the
transcriptional changes elicited by arsenite or thapsigargin
(Fig. [84]1e). This included canonical genes associated with ISR
activation^[85]8 and the ISR target gene signature of coherently
co-expressed genes (ISR GeneCLIC signature), as defined from unbiased,
tissue-independent Clustering by Inferred Co-Expression analysis by
Wong et al.^[86]30,[87]31 (Supplementary Fig. [88]1e, Supplementary
Data [89]1). Of the 89 genes with a significantly different response to
dimerizer as compared to thapsigargin (Fig. [90]1f), Gene Ontology
Biological Process terms associated with the UPR were uniquely enriched
by over-representation analysis with thapsigargin treatment, as was
expression of known target genes of the ATF6 and IRE1 branches of the
UPR^[91]32 (Supplementary Fig. [92]1f), demonstrating that these
pathways were not engaged by Dmr-PERK activation. Similarly, 1686 genes
were uniquely regulated with arsenite treatment, including genes
involved in protein folding, chaperone pathways and non-coding RNA
processing (Fig. [93]1g, Supplementary Data [94]1). Dimerizer treatment
on parental cells that do not carry the Dmr-PERK construct had no
significant transcriptional effects over untreated cells (Supplementary
Fig. [95]1g). These results establish that this pharmacogenomic tool
elicits a targeted ISR, enabling us to specifically assess downstream
cellular events without the contribution of the parallel pathways
engaged by pleiotropic stressors.
Transcriptional targets of the ISR show distinct responses to the strength
and duration pathway input
ISR signaling helps cells adapt to changes in their environment but
severe and/or chronic stress can trigger cell death pathways and impair
physiological processes^[96]1,[97]28,[98]30,[99]33. To determine how
the transcriptional outputs of the ISR vary with time (duration) and
input strength (stress level), we performed RNA-seq on cells treated
for 1, 2, 4, 8, 16, or 24 h with one of three doses of dimerizer. We
chose a low dose of dimerizer (0.01 nM) where protein synthesis
inhibition would be minimal (<15%) but ATF4 protein expression was
still induced, an intermediate dose (0.2 nM) that caused a partial
(50%) reduction in protein synthesis, and a high dose (3 nM) that led
to the strongest (75%) reduction in protein synthesis and accumulation
of SGs (Fig. [100]1d, Supplementary Fig. [101]1a). To analyze the gene
expression time course data, we first normalized the expression of each
gene to its average expression at t = 0, filtered for genes that showed
a log[2] fold change of 1.5 or greater at any time point, and then
identified genes with a statistically significant time-dependent
response to any dose of dimerizer treatment. Genes with a significant
time-dependent response were grouped using hierarchical clustering,
revealing four classes of genes with distinct dynamic behaviors
(Fig. [102]2a, b, Supplementary Data [103]2). A representative response
pattern for each class is illustrated in Fig. [104]2c.
Fig. 2. Varying the strength and duration of ISR activation reveals distinct
transcriptional responses.
[105]Fig. 2
[106]Open in a new tab
a Heatmap of protein coding genes with time-dependent expression
changes in Dmr-PERK cells treated with the indicated concentration of
dimerizer for the indicated amount of time. Columns show the mean
expression of three technical replicates relative to the mean
expression at t = 0 on a log[2] scale. Classes describe groups of genes
clustered by their dose- and time-dependent responses b Selected GO
terms with significant enrichment by ORA of the genes within each
class. There were no significant enriched pathways among Class 3 genes,
but the PERK-mediated unfolded protein response term identified
transcripts exhibiting unique behavior. c Plot of log[2] transcripts
per million (TPM) over time at the indicated doses of dimerizer for
genes representative of the classes described in a. GOT1 is a
representative Class 1 gene, IL4I1 is a representative Class 2 gene,
IL1RL1 is a representative Class 3 gene, and PCDH18 is a representative
Class 4 gene.
The first class (Class 1) was characterized by genes with a rapid
(2–4 h) and sustained upregulation at all 3 doses of dimerizer
(Fig. [107]2b, c). Significantly enriched GO pathways included genes
related to non-coding RNA processing (ncRNA metabolic process, ncRNA
processing), ribosome biogenesis (ribosome biogenesis, rRNA metabolic
process), and tRNA metabolism (tRNA modification, tRNA metabolic
process). The ISR target genes in amino acid metabolism^[108]8,[109]30
were also among the significantly induced GO pathways including “amino
acid import” (ATP1A2, SLC47A1, SLC7A2, SLC1A6, SLC7A5, SLC6A14, SLC3A2,
SLC7A1, SLC25A38, SLC6A9, SLC1A4, SLC1A5, SLC7A11) and “carboxylic acid
biosynthetic process” which included enzymes involved in synthesis of
non-essential amino acids including serine (PSAT1, PSPH, PHGDH),
asparagine (ASNS), cysteine (CBS, CTH, CBSL), proline (PYCR1), and
alanine (GPT2) and transaminases (GOT1, GPT2, BCAT1) (Supplementary
Data [110]2). Class 1 comprises the largest number of genes that are
upregulated by this minimal ISR-inducing system and represents the
earliest and most sensitive ISR gene-expression outputs. Class I gene
expression was inhibited when Dmr-PERK was activated in the presence of
the ISR inhibitor ISRIB, which allosterically antagonizes phospho-eIF2ɑ
and restores translation initiation^[111]34–[112]36, confirming that
this transcriptional response is a consequence of reduced ternary
complex (Supplementary Fig. [113]2a).
A second and smaller class (Class 2) of upregulated genes showed a
pulse-like behavior. Gene expression in this cluster peaked between 4
and 8 h but returned to baseline by 24 h (Fig. [114]2b, c). The gene
sets overrepresented in this class related to immune signaling,
including interleukin-4 mediated signaling (IL2RG, JAK3, IL4R) and TNF
signaling pathways (VCAM1, NFKB1, CLDN1, SPHK1, TRAF3, TNFRSF9,
EDARADD, BIRC3, TNFRSF21, TRAF1, GBP3), and response to osmotic stress
(CLDN1, ICOSLG, ABCB1, RELB, EGFR, RCSD1) (Supplementary Data [115]2).
These acute responders were also upregulated at low levels of pathway
input, but their rapid downregulation suggests that their prolonged
expression may not be adaptive.
In contrast to Classes 1 and 2, genes in a third class (Class 3) were
induced only at higher doses of dimerizer (Fig. [116]2b, c). While
there were no significantly enriched pathways among this class, we
found transcripts encoding two major effectors of the ISR: PPP1R15A,
encoding for GADD34, and DDIT3, encoding CHOP (Supplementary
Fig. [117]2b, c, Supplementary Data [118]2). Although these transcripts
are only induced when protein synthesis is strongly reduced, both
PPP1R15A and DDIT3 contain a uORF that allows the rapid translation of
their baseline pre-existing mRNA, even at low levels of Dmr-PERK
activation^[119]2,[120]37.
Finally, the fourth and largest class (Class 4) consisted of genes that
were downregulated over time proportionally to dimerizer dose
(Fig. [121]2b, c). The timing of the downregulation of Class 4 targets
varied but was generally initiated at later time points compared to the
upregulation of genes in Class 1. Associated GO terms included those
related to cell adhesion and extracellular matrix pathways, as well as
several pathways related to developmental signaling and differentiation
and may reflect the slowing of cell growth processes with prolonged ISR
activation. Together, this analysis revealed that the gene expression
changes downstream of ISR activation exhibit distinct profiles, both in
terms of their dynamics and in response to the strength of the input.
These modules suggest that multiple control mechanisms coordinate the
transcriptional network downstream of the ISR and that the dynamics of
these outputs is important in defining downstream cellular responses.
ATF4-dependent transcriptional targets are early responders and enriched for
metabolic regulators
Transcriptional profiling revealed an ISR gene-expression program that
reflects the duration and strength of stress input. Because ATF4 is a
central ISR transcription factor that quickly accumulates even at very
low levels of pathway input (Supplementary Fig. [122]1a) we assessed
the contribution of ATF4 to the transcriptional response elicited by
Dmr-PERK. We used CRISPR to knock out the ATF4 gene from Dmr-PERK cells
(ATF4 KO) and isolated homozygous mutant clones (Supplemental
Fig. [123]3a). As expected, dimerizer induction of p-eIF2ɑ and
inhibition of puromycin incorporation, which are upstream events, was
not compromised in these cells (Supplementary Fig. [124]3b, c). We also
observed the basal gene expression changes reported in other ATF4-null
cell types^[125]8,[126]38 (Supplementary Data [127]3) and rescued the
resulting impaired cell growth by supplementation with nonessential
amino acids and β-mercaptoethanol^[128]8. Accordingly, all comparisons
between WT and ATF4 KO cells were performed in supplemented medium.
We compared the transcriptional response of WT and ATF4 KO cells after
2, 4, 8, 16, or 24 h of dimerizer treatment at the intermediate
concentration (0.2 nM) that resulted in a partial (50%) inhibition of
protein synthesis and elicited induction of all four classes of genes
in WT cells (Figs. [129]1d, [130]2). Surprisingly, the majority of
transcriptional changes still occurred in ATF4 KO cells, including
upregulation of ncRNA processing pathway genes in Class 1, and the
majority of genes in Classes 2, 3, and 4. Of the 3487 genes that were
significantly changed over time, only 94 were significantly altered in
ATF4 KO cells. Clustering of these ATF4-dependent genes showed that the
majority were upregulated with dimerizer treatment in WT cells
(Fig. [131]3a) and were enriched for GO terms related to amino acid
synthesis and import, including “cellular amino acid biosynthetic
process”, “serine family amino acid metabolic process,” and “amino acid
import” (Fig. [132]3b and Supplementary Data [133]3). The majority
(75%) of the ATF4-dependent genes were from Class 1, which we confirmed
in a second independently derived clone (Supplementary Fig. [134]3d).
Thus, ATF4 is responsible for only a very small subset of the
transcriptional changes that ensued upon ISR activation but drives one
of the earliest and most sensitive transcriptional modules (8% of all
upregulated genes) that is characterized by amino acid metabolism.
Fig. 3. ATF4 drives a metabolic gene expression signature.
[135]Fig. 3
[136]Open in a new tab
a Heatmap of protein coding gene expression changes in WT and ATF4 KO
Dmr-PERK cells at the indicated time points following treatment with
dimerizer (0.2 nM). Genes shown have a significantly altered time
course between the two genotypes. Columns show the mean expression of
three technical replicates relative to the mean expression at t = 0 on
a log[2] scale. Color bar indicates the class of each transcript based
on the dose response study in Fig. [137]2. Gray boxes indicate genes
that did not meet statistical significance in the dose response study
from Fig. [138]2. b Selected GO terms with significant enrichment by
ORA for the upregulated ATF4-dependent genes. Top 20 GO terms are in
Supplementary Data [139]3.
ATF4 diverts carbon utilization away from energy production and towards amino
acid metabolism
Gene-expression profiling showed that ISR activation led to a rapid and
sustained upregulation of several metabolic enzymes and transporters
(Class 1, Fig. [140]2), many of which were ATF4 dependent
(Fig. [141]3). To determine how this transcriptional program impacts
cellular metabolism, we performed a time course of untargeted
metabolomic profiling of polar metabolites by tandem liquid
chromatography-mass spectrometry (LC-MS) in both WT and ATF4 KO
Dmr-PERK cells. Measurements were performed at 0.5, 4, 8, and 24 h of
dimerizer treatment, which allowed us to capture any metabolic changes
that occurred rapidly following Dmr-PERK activation, as well as at
timepoints that correspond with the transcriptional analysis. To
minimize gross metabolic alterations caused by reduced protein
synthesis, we chose a low concentration of dimerizer (0.05 nM) that
only mildly (<15%) inhibited bulk protein synthesis (Fig. [142]1d) but
still induced ATF4 protein expression (Supplementary Fig. [143]1c), the
transcription of Class 1 genes, and the ATF4-dependent metabolic module
(Figs. [144]2, [145]3). We identified metabolites that changed
significantly over time following Dmr-PERK activation and clustered
these to reveal distinct responses between the two genotypes
(Fig. [146]4a and Supplementary Data [147]4).
Fig. 4. ISR activation rewires central carbon metabolism.
[148]Fig. 4
[149]Open in a new tab
a Heatmap of significant metabolites (FDR q < 0.01) as measured by
LC-MS at the indicated time points following addition of dimerizer
(0.05 nM) to WT or ATF4 KO Dmr-PERK cells. Relative abundances are
shown as row mean centered z-scores. b Heatmap of TCA cycle metabolites
from a. c Schematic of incorporation of ^13C-label from U-^13C6-glucose
into the TCA cycle and in the synthesis of fatty acids. d Bar plots
representing fractional labeling of TCA cycle metabolites from
U-^13C6-glucose in WT and ATF4 KO cells treated with vehicle (Veh.) or
dimerizer (Dmr, 0.05 nM) for 24 h. Error bars depict mean ± SD of three
technical replicates. e Schematic of incorporation of ^13C-label from
U-^13C5-glutamine from oxidation (blue) vs. reductive carboxylation
(red). Bar plots representing fraction labeling of malate, Asn or Asp
from oxidative (M + 4) (f) or reductive (M + 3) (g) metabolism of
U-^13C5-glutamine in WT and ATF4 KO cells treated with vehicle (Veh.)
or dimerizer (Dmr, 0.05 nM) for 24 h. Error bars depict mean ± SD of
three technical replicates. Bar plots representing the fraction of
palmitate carbons labeled from U-^13C5-glutamine (h) or U-^13C6-glucose
(i) in lipids from WT or ATF4 KO Dmr-PERK cells treated with vehicle
(Veh.) or dimerizer (Dmr, 0.05 nM) for 24 h. Error bars depict
mean ± SD of three technical replicates. Statistical significance was
evaluated by two-way ANOVA followed by Tukey’s HSD test. Source data
are provided as a [150]Source Data file.
A first cluster of metabolites showed a rapid (4 h) and sustained
increase over time (Fig. [151]4a, marked in red) and included most
amino acids, both essential and non-essential, as well as intermediates
in glutathione synthesis. The timing and content of this cluster
suggests that it is driven by the upregulation of Class 1 genes, which
includes both amino acid transporters and biosynthetic enzymes,
including the transsulfuration pathway enzymes (Fig. [152]2a, b). This
transcriptional module was ATF4-dependent and, as anticipated,
accumulation of metabolites in this cluster was absent in ATF4 KO
cells.
Amino acids in this cluster accumulated in an ATF4-dependent manner and
were low in ATF4 KO cells, even at baseline (Supplementary
Fig. [153]4a). A notable exception to the trend of increasing amino
acid concentrations was aspartate, whose lack of accumulation is
consistent with ATF4 transcriptional induction of asparagine synthetase
(ASNS), which converts aspartate to asparagine^[154]39 (Fig. [155]3b).
Because several enzymes in amino acid synthesis pathways were
identified as ATF4-dependent Class 1 genes, including serine synthesis
enzymes (Fig. [156]3b), we considered whether upregulation of these
pathways may be contributing to the cellular accumulation of amino
acids. To assess pathway activity, we cultured cells in media
containing isotopically labeled glucose (U-^13C6-glucose) and
quantified the steady state incorporation of heavy carbon into
metabolic intermediates by LC-MS after 24 h of dimerizer (0.05 nM)
treatment (Supplementary Fig. [157]4b). Despite a culture media replete
with serine, glucose labeling of serine was increased with Dmr-PERK
activation in WT cells but remained nearly undetectable in ATF4 KO
cells (Supplementary Fig. [158]4c).
ISR induction also led to an ATF4-dependent increase in the abundance
of γ-glutamylcysteine in this cluster (Supplementary Fig. [159]4d).
This is indicative of enhanced synthesis of glutathione, the most
abundant antioxidant in cells, and is consistent with the regulation of
the transsulfuration pathway enzymes CBS and CTH by ATF4 (Supplementary
Data [160]3)^[161]8,[162]25. We determined whether glutathione
synthesis was also upregulated by Dmr-PERK activation by culturing
cells with isotopically labeled glutamine (^13C5-glutamine), a primary
precursor to glutathione (Supplementary Fig. [163]4e). We found that,
indeed, metabolic flux from glutamate to glutathione was increased ~50%
(F[Dmr]/F[Veh] = 1.46) with ISR activation in WT cells but was
unaffected in ATF4 KO cells (Supplementary Fig. [164]4f). Together,
these results demonstrate that the ISR is sufficient to drive an early
ATF4-dependent increase in amino acid and glutathione synthesis, a
result in agreement with the gene expression analysis and with previous
studies using pleiotropic stressors^[165]8,[166]26.
In parallel to the accumulation of amino acids, a second cluster of
metabolites rapidly (4 h) decreased upon ISR activation (Fig. [167]4a,
marked in green). This class was composed primarily of TCA cycle
intermediates (Fig. [168]4b), suggesting a possible slowing of the TCA
cycle with ISR induction. Surprisingly, this cluster was also
ATF4-dependent, even though we did not detect any corresponding
transcriptional changes. To test whether depletion of these metabolites
was due to reduced oxidation of glucose, we measured U-^13C6-glucose
labeling of TCA cycle intermediates following Dmr-PERK activation
(Fig. [169]4c) and observed a reduced fraction of α-ketoglutarate,
malate, citrate and acetyl-CoA labeling in WT but not ATF4 KO cells
(Fig. [170]4d). Furthermore, glucose labeling of malate and
α-ketoglutarate was higher in ATF4 KO cells than WT cells, even without
ISR activation (Fig. [171]4d, white bars), suggesting that, like what
we observed with amino acid synthesis, the ATF4-dependent mechanisms
that contribute to the slowing of the TCA cycle are also active
basally.
In addition to glucose, cultured cells rely heavily on glutamine to
fuel the TCA cycle^[172]40. This was reflected in the relatively low
fraction of malate and α-ketoglutarate (<2%) that was labeled by
glucose in WT cells (Fig. [173]4d). We assessed whether ISR activation
also inhibited glutamine oxidation by measuring ^13C5-glutamine
labeling of oxidative TCA cycle intermediates, which can be identified
by their M + 4 mass (Fig. [174]4e, blue schematic). As expected, the
fractional labeling of TCA cycle metabolites from glutamine was higher
than from glucose (Fig. [175]4f). However, labeling of malate derived
from glutamine oxidation (M + 4) was also decreased significantly more
in WT cells after dimerizer treatment than in ATF4 KO (Fig. [176]4f).
Thus, activation of the ISR reduces oxidation of both glucose and
glutamine by the TCA cycle in an ATF4-dependent manner.
This slowing of the TCA cycle would be expected to impact mitochondrial
respiration by limiting the amount of NADH available for oxidative
phosphorylation. We used a Seahorse bioanalyzer to assess oxygen
consumption rate (OCR), a measurement of oxidative phosphorylation,
after 24 h of dimerizer treatment, consistent with the timing of the
tracing experiments. Basal OCR was decreased with Dmr-PERK induction in
both WT and ATF4 KO cells, which is consistent with the decreased
energy demand caused by comparable inhibition of protein synthesis in
both genotypes (Supplementary Fig. [177]4g, h). However, the decrease
in basal OCR was more pronounced in WT cells, as was the reduction in
mitochondrial spare respiratory capacity (Supplementary Fig. [178]4g,
i), consistent with ATF4-dependent TCA cycle inhibition. Addition of
ISRIB to WT cells had no effect on OCR in the absence of dimerizer but
prevented the decrease in basal OCR and spare respiratory capacity
caused by Dmr-PERK activation (Supplementary Fig. [179]4j, k),
indicating that, as expected, slowing of the TCA cycle and of oxidative
phosphorylation are a consequence of reduced eIF2B activity downstream
of eIF2 phosphorylation.
When oxidative metabolism in the mitochondria is impaired,
α-ketoglutarate derived from glutamine can be reductively carboxylated
to citrate to support amino acid and fatty acid biosynthesis
(Fig. [180]4e, red schematic)^[181]41,[182]42. To assess whether this
pathway was engaged with Dmr-PERK activation, we traced ^13C5-glutamine
labeling of reductive carboxylation products, identified by their M + 3
mass. Malate, aspartate and asparagine labeling through this pathway
all increased after Dmr-PERK activation (Fig. [183]4g), indicating that
reductive carboxylation of glutamine was mobilized to support the
synthesis of these amino acids. Consistent with the TCA cycle
measurements, this was genotype dependent as glutamine labeling of
reductive carboxylation products was constitutively low in ATF4 KO
cells (Fig. [184]4g).
Reductive carboxylation-derived citrate can be used to produce
cytosolic acetyl-CoA for de novo synthesis of fatty acids, which we
assessed by measuring incorporation of U-^13C5-glutamine into the acyl
chains lipids. We saw that glutamine incorporation into palmitate
(C16:0) was unchanged with Dmr-PERK activation in either genotype
(Fig. [185]4h), indicating that fatty acid synthesis via reductive
carboxylation of alpha-ketoglutarate was maintained upon ISR induction.
However, we observed a significant ATF4-dependent reduction in
glucose-derived palmitate with dimerizer treatment (Fig. [186]4i). This
amounted to an overall ATF4-dependent decrease in cellular de novo
lipid synthesis, as measured by reduced deuterium labeling of palmitate
in cells cultured in heavy water (^2H2O) (Supplementary Fig. [187]4l).
Taken together, these results indicate that the ISR induces an
ATF4-dependent slowing of the TCA cycle and inhibition of de novo fatty
acid synthesis, while maintaining amino acid synthesis derived from
glucose or reductive carboxylation of glutamine.
To test whether this metabolic response occurs in other cellular
contexts and modes of ISR induction, we used primary embryonic
fibroblasts (MEFs) derived from mice that carry a homozygous point
mutation (N208Y) in the alpha subunit of eIF2B^[188]43. This causes a
dramatic reduction of eIF2Bα levels and activation of the ISR without
phosphorylation of eIF2 (Supplementary Fig. [189]5a). This constitutive
ISR is attenuated by maintaining MEFs in the presence of the ISRIB
analog 2BAct^[190]30. Withdrawal of 2BAct leads to upregulation of ATF4
(Supplementary Fig. [191]5a) and increased ISR target gene expression
(Supplementary Fig. [192]5b) in N208Y MEFs and, to a lesser extent, in
WT MEFs. Untargeted metabolomics profiling was performed on N208Y cells
maintained in the presence or absence of 2BAct for 24 h. Overall, the
change in metabolite levels in N208Y MEFs upon 2BAct withdrawal
positively correlated (r = 0.575) with the changes observed with
Dmr-PERK activation in U2OS cells (Supplementary Fig. [193]5c). Further
similarities between the metabolic responses of N208Y MEFs and Dmr-PERK
cells to ISR activation were observed with ^13C6-glucose and
^13C5-glutamine tracing. As with Dmr-PERK activation, oxidation of
glucose by the TCA cycle was significantly reduced in N208Y MEFs in the
absence of 2BAct (Supplementary Fig. [194]5d–f). 2BAct withdrawal also
decreased glucose labeling of TCA cycle intermediates in WT MEFs
(Supplementary Fig. [195]5 e, f), consistent with the basal ISR
detected in these cells. We also measured a significant increase in the
synthesis of glutathione (Supplementary Fig. [196]5g) and asparagine
(Supplementary Fig. [197]5h) from glutamine in N208Y MEFs upon 2BAct
withdrawal. Intriguingly, N208Y MEFs also synthesize more proline from
glutamine (Supplementary Fig. [198]5i), an effect that was not observed
with ISR activation in the Dmr-PERK cells. Fibroblasts have been shown
to synthesize high levels of proline to support collagen
synthesis^[199]44, so this discrepancy may be due to differences in
cell lineage. Together, the concordance of metabolic phenotypes in
these two very different cell systems suggests that the observed
reduction in glucose oxidation in the TCA cycle, as well as the
increase in synthesis of glutathione and non-essential amino acids are
generalizable responses to ISR induction.
The ISR inhibits glycolysis and lipid synthesis pathways independently of
ATF4
In addition to the ATF4-dependent changes described above, we
identified two clusters of metabolites that either increased
(Fig. [200]4a, marked yellow) or decreased (marked purple) in both
genotypes during ISR induction. Early glycolytic intermediates and
pentose phosphate pathway metabolites accumulated with prolonged
activation of the ISR in both genotypes while lactate levels were
significantly reduced (Supplementary Fig. [201]6a), suggesting that ISR
induction leads to an ATF4-independent decrease in the rate of
glycolysis, consistent with Seahorse measurements showing reduced
extracellular acidification rates (ECAR), a measurement of lactate
production by glycolysis, independent of genotype (Supplementary
Fig. [202]4m). In agreement with the lack of ATF4 effect on glycolysis,
there was no defect in basal ECAR in ATF4 KO cells. Interestingly, the
ATP/ADP ratio increased with prolonged ISR activation in both genotypes
(Supplementary Fig. [203]6b, c). This suggests that, although
activation of the ISR slowed both oxidative phosphorylation and
glycolysis, less ATP is being consumed in these cells, possibly as a
result of slowed protein synthesis and other anabolic processes.
Finally, several polar metabolites of phospholipid metabolism
(glycerol-3-phosphate, choline and phosphoethanolamine) accumulated
with ISR activation (Supplementary Fig. [204]6d), suggesting that, in
addition to ATF4-dependent inhibition of de novo fatty acid synthesis,
the ISR also alters lipid metabolism pathways through ATF4-independent
mechanisms.
ISR activation modifies cellular lipid composition and induces lipid droplet
formation
The observation that intermediates of phospholipid synthesis were
altered with Dmr-PERK activation suggested that the ISR could be
modifying cellular lipid composition. To investigate whether this was
the case, we measured the lipidome composition at the same time points
and dimerizer concentration (0.05 nM) as in the metabolomics analysis.
We found that 73% of identified lipid species showed a significant
change in abundance over time (Supplementary Data [205]5). Clustering
analysis identified two clear trends which had contrasting dynamic
behaviors (Fig. [206]5a, Supplementary Fig. [207]7a–c). The largest
cluster showed a time-dependent decrease in abundance with Dmr-PERK
activation and was composed mostly of diacylglycerol (DG) and
phospholipids (Supplementary Fig. [208]7a,d). A second cluster,
consisting of TG and CE species showed a striking accumulation over
time (Supplementary Fig. [209]7b,e) and was enriched in lipids
containing polyunsaturated fatty acyl chains (PUFAs) (Supplementary
Fig. [210]7f). Pathway enrichment analysis further demonstrated a
significant accumulation of TG and CE with long chain PUFA acyl chains
containing at least three double bonds (20:4, 22:4, 22:5, 24:6) after
24 h of Dmr-PERK activation, whereas unsaturated and monounsaturated
fatty acids (16:1, 12:0, 14:1, 14:0, 24:0) were underrepresented
(Supplementary Fig. [211]7g). Although there was strong enrichment in
18:0 acyl chains, this is most likely the result of it being typically
found in combination with PUFA side chains in TG species (Supplementary
Fig. [212]7h).
Fig. 5. The ISR alters the cellular lipidome and drives LD formation to
promote cell survival during stress.
[213]Fig. 5
[214]Open in a new tab
a Heatmap of lipid abundances in Dmr-PERK cells treated for the
indicated amount of time with dimerizer (0.05 nM). Columns show three
technical replicate measurements for each time point, grouped by lipid
class. Measurements are normalized to the mean abundance at t = 0 on a
log[2] scale. Color bars indicate triglycerides (TG), cholesterol
esters (CE), diacylglycerides (DG), glycerophospholipids (GPL),
plasmalogens, sphingolipids and other lipid species. b Schematic of the
lipid synthesis pathways that supply the constituents of lipid
droplets. Enzymes mediating each step are labeled in gray. c BODIPY
labeled Dmr-PERK cells treated for 8 h with vehicle (Veh) or dimerizer
(Dmr, 1 nM). Scale bar = 5 μm d Quantification of the number of LD per
cell in Dmr-PERK cells treated for 8 h with vehicle (Veh.) or the
indicated concentration of dimerizer (Dmr). Median with interquartile
range is indicated. 240 cells were quantified per condition.
Statistical significance was evaluated by one-way ANOVA followed by
Dunnett’s multiple comparison test. e Quantification of the number of
LD in Dmr-PERK cells at the indicated time points following treatment
with the indicated concentration of dimerizer. Error bars show
mean ± SD per well across three independent experiments. f, g Bar plots
representing the mean number of LD per cell per well in Dmr-PERK cells
treated for 8 h with dimerizer (Dmr, 1 nM), ACATi (10 μM), DGAT1/2i
(20 μM each) or ACCi (2 μg/mL) as indicated. Error bars show mean ± SD
across three independent experiments. Statistical significance was
evaluated by one-way ANOVA followed by Dunnett’s multiple comparison
test. h Quantification of the percent confluence of Dmr-PERK cells
treated with vehicle, dimerizer (Dmr, 0.024 nM), ACATi (10 μM),
DGAT1/2i (20 μM each) or ACCi (2 μg/mL), as indicated. Error bars show
mean ± SD of two technical replicates and is representative of three
independent experiments. Source data are provided as a [215]Source Data
file.
Because cells deposit TGs and CEs in LDs, ER-derived organelles
composed of a phospholipid monolayer membrane encapsulating a
hydrophobic lipid core (Fig. [216]5b)^[217]45,[218]46, we assessed
whether Dmr-PERK activation of the ISR was accompanied by the formation
of these structures. Dimerizer treatment led to a dose and
time-dependent increase in the number of LDs per cell, as visualized
with BODIPY 493/503 (Fig. [219]5c, d), with timing that was consistent
with our measurement of neutral lipid accumulation (Fig. [220]5e). LD
formation has been reported to coincide with the formation of SGs in
response to arsenite treatment^[221]47. Although Dmr-PERK activation
induced the formation of both SGs and LDs, they accumulated with
different kinetics (Supplementary Fig. [222]8a). SGs abundance peaked
at 4 h and resolved by 8 h, whereas LDs steadily increased over 8 h.
Furthermore, we observed no colocalization between SGs and LDs in these
conditions (Supplementary Fig. [223]8b), indicating that these are
distinct structures whose ISR-induced formation is not correlated.
Previous studies have shown that insults such as hypoxia, nutrient
deprivation and ER stress cause LDs to accumulate^[224]48,[225]49. To
test whether the ISR contributes to LD formation in response to
pleiotropic stressors, we measured LDs across a time course in cells
treated with oligomycin or thapsigargin in the presence or absence of
ISRIB^[226]50. Both stressors induced LD formation with kinetics
similar to that of Dmr-PERK activation (Supplementary Fig. [227]8c). In
both cases, addition of ISRIB reduced the number of LDs per cell,
indicating that the ISR is necessary for cell stress-induced LD
biogenesis (Supplementary Fig. [228]8c).
DGAT activity is required for ISR-driven lipid droplet formation
To determine the source of these neutral lipids, we tested whether LD
formation depended on increased lipid uptake by comparing LD induction
in Dmr-PERK cells grown in standard or delipidated media. ISR induction
with either dimerizer or thapsigargin treatment still upregulated LDs
in delipidated media (Supplementary Fig. [229]8d), indicating that this
process does not depend on extracellular fatty acids and suggesting
that cell-intrinsic remodeling underlies ISR-dependent LD formation.
This change is independent of ATF4, as ATF4 KO cells were still able to
induce LD formation (Supplementary Fig. [230]8e), a result consistent
with the ATF4-independent accumulation of lipid related molecules
glycerol-3-phosphate, choline and phosphoethanolamine (Fig. [231]4a,
Supplementary Fig. [232]6d).
The ISR has been reported to inhibit signaling by the
nutrient-sensitive mechanistic target of rapamycin complex 1
(mTORC1)^[233]51, a master regulator of cellular metabolism whose
inhibition is sufficient to increase cellular LD abundance and is
required for LD biogenesis during nutrient deprivation^[234]48,[235]52.
Although activation of Dmr-PERK induced expression of DDIT4 and SESN2
(Supplementary Data [236]3), known regulators of mTORC1
signaling^[237]51,[238]53,[239]54, we saw no evidence of decreased
mTORC1 activity over a time course of Dmr-PERK activation, as measured
by either western blot (Supplementary Fig. [240]8f, g) or using a
fluorescent immunoassay to monitor phosphorylation of the mTORC1 target
S6K (Supplementary Fig. [241]8h). This indicates that mTORC1 is not
modulated by the ISR in this system and is therefore not required for
ISR-driven LD formation.
Finally, we assessed whether transcriptional changes to pathways
involved in TG and CE synthesis could account for their accumulation.
Consistent with the reduced deuterium labeling of palmitate
(Supplementary Fig. [242]4l), fatty acid synthesis enzymes were
unchanged or reduced in expression with dimerizer treatment
(Supplementary Fig. [243]9a), indicating that increased de novo fatty
acid synthesis was not responsible for increased TG content.
Conversely, several enzymes involved in TG synthesis and LD biogenesis
were induced by the ISR (Supplementary Fig. [244]9a). Most notably, the
transcript for the diacylglycerol acyltransferase DGAT2 was elevated
over time with dimerizer treatment as a part of the Class 1 responders.
DGAT1 and DGAT2 catalyze the final, rate-limiting step in TG synthesis
by esterifying DG to yield TG which is then packaged into
LDs^[245]46,[246]55. DGAT1 showed no change in expression over time
with ISR induction but was constitutively expressed at a high level
relative to DGAT2 (Supplementary Fig. [247]9b). Consistent with LD
formation being independent of ATF4, the induction of DGAT2 transcript
was also observed in ATF4 KO cells (Supplementary Fig. [248]9c).
To investigate the role of DGATs in mediating the ISR-driven
upregulation of LDs, we activated Dmr-PERK in the presence of the
pharmacological inhibitors of DGAT1, T863, and DGAT2, PG-06424439
(combined, DGAT1/2i). We compared the effects of DGAT1/2i to inhibiting
CE synthesis with the pan-cholesterol acyltransferases (ACAT) inhibitor
Avasimibe (ACATi), or to inhibiting the rate limiting step in de novo
fatty acid synthesis with the acetyl-CoA carboxylase (ACC) inhibitor
TOFA (ACCi). In the absence of dimerizer, cellular LD content at 8 h
was decreased by all inhibitors, a result consistent with the role of
these pathways in regulating LD biogenesis
(Fig. [249]5f)^[250]46,[251]55. However, with the addition of
dimerizer, only DGAT1/2i abolished the induction of LDs, while ACATi
and ACCi did not prevent the relative increase in cellular LD content
by Dmr-PERK activation (Fig. [252]5f, g). Together, these findings
demonstrate that ISR-driven LD formation depends critically and
specifically on TG synthesis by DGAT activity through an
ATF4-independent mechanism.
Inhibition of LD formation upon activation of the ISR impairs cell survival
LDs are increasingly recognized as serving a protective role during
cell stress by providing a source of fatty acids for energy production,
regulating membrane composition, and sequestering deleterious proteins
and lipid species^[253]45,[254]48,[255]49. We therefore considered
whether LD induction protected against cell death during a prolonged
ISR^[256]1. We performed time-lapse microscopy on Dmr-PERK cells and
measured their confluence over 96 h of ISR activation in the presence
or absence of lipid synthesis inhibitors. Dimerizer treatment alone had
a dose-dependent effect on cell growth and viability, with extensive
cell death occurring at the highest doses (15 nM, 3 nM) and only
moderate effects on confluency measured at the lowest dose (0.024 nM)
(Supplementary Fig. [257]9d). Despite their effects on basal LD content
(Fig. [258]5f), ACATi, DGAT1/2i or ACCi alone had no effect on cell
confluence (Supplementary Fig. [259]9e). However, when a sublethal
dimerizer treatment was combined with DGAT1/2i, cell viability was
severely compromised, an effect not observed with ACATi or ACCi
co-treatment (Fig. [260]5h). Collectively, our findings indicate that
the ISR leads to a reorganization of cellular lipid content, favoring
DGAT-dependent TG production and storage in LDs, as part of a
protective response to prolonged stress signaling.
DGAT1 and 2 have distinct functions in ISR-induced LD formation and cell
survival
Although DGAT1 and DGAT2 catalyze the same reaction, they are
evolutionarily unrelated proteins that have overlapping but distinct
functions^[261]55. To test the relative contribution of DGAT1 and DGAT2
to ISR-induced LD formation and cell viability, we activated Dmr-PERK
in the presence of each DGAT inhibitor individually. Following 8 h of
dimerizer treatment, LD formation was fully prevented by DGAT1i but not
by DGAT2i which, instead, further increased LD content (Fig. [262]6a,
b), indicating that TG synthesis by DGAT1 is sufficient to drive LD
formation after Dmr-PERK activation. When DGAT1i was combined with a
sublethal dose of dimerizer (0.024 nM), there was no effect on cell
confluency (Fig. [263]6c, Supplementary Fig. [264]10a), despite its
ability to prevent LD formation. However, inhibition of DGAT2 was
sufficient to reduce cell viability (Fig. [265]6c) and induce apoptotic
caspase activation (Fig. [266]6d, Supplementary Fig. [267]10b) to the
same extent as DGAT1/2i. Similarly, siRNA knockdown of DGAT1, but not
DGAT2, blunted LD formation when Dmr-PERK was activated (Supplementary
Fig. [268]10c, d), whereas only knockdown of DGAT2 was sufficient to
impair cell growth during prolonged treatment with a low dose of
dimerizer (0.12 nM) (Supplementary Fig. [269]10e, f). Therefore,
although ISR-driven LD formation is DGAT1-dependent, cell viability
depends on the enzymatic activity of DGAT2.
Fig. 6. DGAT1 and DGAT2 play distinct roles in ISR-induced LD formation and
cell survival.
[270]Fig. 6
[271]Open in a new tab
a, b Bar plots representing the mean number of LD per cell per well in
Dmr-PERK cells treated for 8 h with dimerizer (Dmr, 1 nM), DGAT1i
(20 μM) or DGAT2i (20 μM) as indicated. Error bars show mean ± SD
across three independent experiments. Statistical significance was
evaluated by one-way ANOVA followed by Dunnett’s multiple comparison
test. c Quantification of the percent confluence of Dmr-PERK cells
treated with vehicle, dimerizer (Dmr, 0.024 nM), DGAT1i (20 μM) or
DGAT2i (20 μM), as indicated. Error bars show mean ± SD of two
replicates and is representative of three independent experiments. d
Western blot for cleaved caspase-3 in lysates from Dmr-PERK cells
treated with the indicated dose of dimerizer (Dmr), DGAT1i (20 μM) or
DGAT2i (20 μM), as indicated. Quantification of the number of LD (e)
and cell number (f) simultaneously imaged by high-content live-cell
microscopy in Dmr-PERK cells treated with vehicle, dimerizer (Dmr),
DGAT1i (20 μM) or DGAT2i (20 μM), as indicated. Error bars show
mean ± SD of three independent experiments. g Western blot for ATF4 in
lysates from Dmr-PERK cells treated for 24 h with dimerizer (Dmr,
0.05 nM), DGAT1i (20 μM) or DGAT2i (20 μM), as indicated. h Average
z-core of ISR GeneCLIC genes calculated from nCounter gene expression
profiling of Dmr-PERK cells treated for 24 h with dimerizer (Dmr,
0.05 nM), DGAT1i (20 μM) or DGAT2i (20 μM), as indicated. Error bars
show mean ± SD of three technical replicates. Statistical significance
was evaluated by one-way ANOVA followed by Dunnett’s multiple
comparison test. Source data are provided as a [272]Source Data file.
Because DGAT enzymes have been reported to be modulated by
posttranslational modifications, including
phosphorylation^[273]55–[274]57, we considered whether the downstream
functions of DGAT1 and DGAT2 depended on the activity of the eIF2
kinase that initiated the ISR. We transiently knocked down expression
of eIF2Bα in U2OS cells to trigger the ISR independently of eIF2
phosphorylation, in a manner analogous to the N208Y mutation
(Supplementary Fig. [275]11a)^[276]43,[277]58. As with Dmr-PERK
activation, this resulted in a DGAT1-dependent increase in LD content
(Supplementary Fig. [278]11b, c) and increased cell death when DGAT2
was inhibited (Supplementary Fig. [279]11d). We also observed an
accumulation of LDs in N208Y MEFs compared to WT after 8 h of
withdrawal from 2BAct, which was prevented by DGAT1i but not DGAT2i
(Supplementary Fig. [280]11e, f). The distinctive roles of DGAT1 and
DGAT2 in mediating LD formation and cell survival during ISR activation
are therefore not a consequence of eIF2 kinase-dependent
posttranslational modifications.
Finally, we examined whether DGAT1 and DGAT2 played similar roles in
response to pleiotropic stress conditions. In both U2OS and HCT116
cells, LD buildup in response to oligomycin was only DGAT1-dependent
(Supplementary Fig. [281]11g, h), whereas only inhibition of DGAT2
impacted proliferation of both cell types in the presence of a
sub-lethal dose of thapsigargin (3.3 nM) (Supplementary Fig. [282]11i,
j). Together, our results indicate that DGAT1 and DGAT2 have unique
contributions downstream of ISR activation across multiple stress
conditions and cell types.
Inhibition of DGAT2 activates the ISR
To better understand the relationship between DGAT activity, LD
formation, and cell growth, we used high content live cell imaging to
simultaneously track LD content and cell number over 24 h of dimerizer
treatment (0.0048 nM, 0.024 nM, 0.12 nM) in the presence or absence of
DGAT inhibitors. By imaging a large number of cells at higher time
resolution, we found that the rapid accumulation of LDs over 8 h of
Dmr-PERK activation was followed by a second, slower phase
(Fig. [283]6e). This was accompanied by a decrease in cell number, with
onset of cell death at the highest dimerizer doses (0.024, 0.12 nM)
(Fig. [284]6f). Simultaneous inhibition of DGAT1 and DGAT2 ablated
cellular LD content throughout the time course (Fig. [285]6e, f),
whereas DGAT1i alone prevented the early peak of LD formation but did
not completely inhibit the second phase (Fig. [286]6e, f, Supplementary
Fig. [287]12a). This suggests that DGAT2 drives a later phase of LD
biogenesis following ISR activation, which is consistent with the
timing of its transcriptional induction (Supplementary Fig. [288]9a,
b). DGAT2i treatment led to a striking amplification of the early peak
of LD formation, followed by the expected loss of cell growth and
viability at all dimerizer doses, which made it impossible to directly
assess the effects of DGAT2 inhibition on the later phase of LD
formation (Fig. [289]6e, f, Supplementary Fig. [290]12a). We found that
the ISR-induced LDs formed in the presence of DGAT2i were on average
smaller than with Dmr-PERK activation alone (Supplementary
Fig. [291]12b, c), which reflects DGAT2’s role in promoting LD
growth^[292]55,[293]59. Together, these results suggest that TG
synthesis and LD formation downstream of the ISR are dynamic processes
that are controlled by time-dependent mechanisms to which DGAT1 and
DGAT2 contribute differently and define distinct downstream cellular
outputs.
We noted that the effect of DGAT2i on LD content and cell growth at low
doses of dimerizer mimicked what was observed at higher dimerizer doses
with no DGAT inhibition. We therefore examined the status of ISR
activity when DGAT inhibitors were present. At a low dimerizer dose
(0.05 nM), treatment with DGAT2i, but not DGAT1i, enhanced ATF4 protein
levels (Fig. [294]6g) and elevated Dmr-PERK-induced expression of ISR
GeneCLIC target genes, (Fig. [295]6h, Supplementary Fig. [296]12d, e),
indicating that loss of DGAT2 activity amplifies ISR signaling.
Strikingly, even in the absence of dimerizer, inhibition of DGAT2 was
sufficient to drive ATF4 protein accumulation (Fig. [297]6g), promote
LD formation (Fig. [298]6e), and induce an ISR gene expression profile
comparable to that of Dmr-PERK (Fig. [299]6h, Supplementary
Fig. [300]12e), suggesting that, even at baseline, DGAT2 is important
for maintaining lipid homeostasis and that loss of its enzymatic
activity triggers a cell stress response. Our findings are consistent
with this endogenous response having additive effects on Dmr-PERK
activation, leading to the observed increase in ISR signaling, enhanced
DGAT1-dependent LD formation, and cell death. This suggests that the
transcriptional induction of DGAT2 is a protective ISR mechanism that
prevents additional cell stress and illustrates how coordination of
transcriptional and metabolic ISR outputs is important in shaping
downstream cellular outcomes.
Discussion
The use of a minimal pharmacogenomic system to activate the ISR allowed
us to explore how the dose and timing of input into the pathway affects
its various outputs. This approach uncovered a previously unappreciated
metabolic remodeling that, in addition to the known effects on amino
acid and glutathione synthesis, had profound effects on cellular
bioenergetics and lipid content (Fig. [301]7). These metabolic
responses were already evident at early time points and low levels of
pathway input, in a regime when protein synthesis inhibition was
minimal and SG formation was not yet detectable. Thus, we propose that
this metabolic state is a primary output of the ISR that is critical
for adaptation to physiological levels of stress.
Fig. 7. Model of ATF4-dependent and ATF4-independent ISR metabolic outputs.
[302]Fig. 7
[303]Open in a new tab
The ISR increases glutathione and serine synthesis, while diverting
glucose and glutamine carbon away from oxidation in mitochondria.
Synthesis of aspartate, asparagine and acyl-CoA is maintained via
reductive carboxylation of glutamine. These ATF4-dependent changes are
accompanied by ATF4-independent inhibition of glycolysis, increased
pentose phosphate pathway activity, and increased TG production and
accumulation in lipid droplets. Changes in pathway utilization,
metabolite abundance and enzyme expression are indicated by color (red
= up, blue = down).
Induction of amino acid synthesis and uptake by ATF4 has been well
described downstream of pleiotropic ISR activating conditions such as
ER stress^[304]8,[305]60, mitochondrial
dysfunction^[306]23,[307]25,[308]61, and amino acid depletion^[309]26.
Here, we demonstrate that ISR activation is sufficient to drive this
response, even in the absence of any cellular insult. The accumulation
of amino acids may therefore be a hard-wired response preparing cells
to resist adverse conditions where amino acid accumulation would be
beneficial, such as ER stress and nutrient deprivation. It’s also
notable that this metabolic response is initiated even at minimal
levels of ISR activity and at very early time points. This may reflect
a physiological condition where stress can still be resolved and these
pools of amino acids would be available to support recovery of protein
synthesis and to fuel cell growth^[310]62,[311]63. The possibility that
this ATF4-dependent process is supporting growth is consistent with
reports of ATF4 activation downstream of mTORC1 activation by
pro-growth signals^[312]9,[313]64,[314]65. Induction of ATF4 by low
doses of insulin initiates a transcriptional response that mirrors the
ATF4-dependent metabolic module defined here and is required to
maintain increased protein synthesis^[315]9. This overlap also suggests
that this transcriptional module represents a core ATF4 transcriptional
program that may promote recovery when stress has subsided.
To our knowledge, this is the first report of ATF4-dependent slowing of
oxidative metabolism. As we did not observe any gene expression changes
that directly explain this effect, it appears to be a secondary
consequence of the ATF4 transcriptional metabolic module. The
accompanying upregulation of the reductive carboxylation of glutamine
is a compensatory response typically seen in cells with impaired
mitochondrial respiration^[316]42 and, in this case, is likely a direct
response to the slowed TCA cycle. Mitochondrial dysfunction through
inhibition of oxidative phosphorylation^[317]25,[318]61, disruption of
the TCA cycle^[319]23, or loss of membrane potential^[320]25 are major
ISR triggers that induce the same ATF4-dependent changes in amino acid
and glutathione metabolism as Dmr-PERK activation. Reductive
carboxylation allows cells to route glutamine utilization away from
oxidative pathways and maintain amino acid synthesis in the context of
dysfunctional mitochondria. This would also alleviate metabolic demand
and stress on already malfunctioning respiratory pathways. Thus, we
propose that an additional central role for ATF4 during the ISR is to
reroute biosynthetic pathways to enable amino acid synthesis during
mitochondrial dysfunction.
This ATF4-dependent metabolic switch may also assist with matching
cellular bioenergetics with reduced energy demand during stress. The
maintenance of ATP levels despite slowing of the TCA cycle and
glycolysis suggests that overall cellular anabolism is repressed by the
ISR. This is also supported by the later-stage downregulation of genes
that support growth processes (Class 4). It is possible that slowed
growth also explains the decrease in cellular phospholipid content, as
inhibition of anabolic processes would reduce the demand for membrane
biogenesis^[321]66. Another possibility is that, in the context of
inhibition of global protein synthesis, phospholipid levels may
decrease to maintain the balance between membrane lipid and protein
content^[322]67. Phospholipids and TG share a common precursor, DG, and
both synthesis pathways draw from the same pool^[323]55,[324]68. It’s
possible that, in an effort to balance membrane phospholipid content,
the ISR drives a conversion of phospholipids to TG. The predominance of
PUFAs in the ISR-induced TG pool suggests that this process could have
additional protective effects, such as resistance to oxidative stress.
Sequestration of PUFAs in LDs can limit phospholipid peroxidation at
cell membranes and protect from ferroptosis^[325]69.
The mechanisms underlying this ISR-induced shift towards TG
accumulation have yet to be defined, as it was not a consequence of the
ATF4-induced metabolic changes and was not dependent on fatty acid
uptake or de novo synthesis. DGAT1-dependent conversion of
phospholipids to TG has been reported downstream of mTORC1 inhibition,
through a lysosome-dependent but autophagy-independent hydrolysis of
phospholipid fatty acids^[326]52. However, this process does not seem
to be occurring in this Dmr-PERK system, which did not inhibit mTORC1,
nor demonstrate the reported accumulation in lyso-phospholipids.
DGAT1-dependent TG accumulation is also seen in response to high levels
of free fatty acids derived from mTORC1-induced autophagy during acute
nutrient deprivation^[327]48. While we do not see any transcriptional
evidence that autophagy is induced by Dmr-PERK activation, we cannot
rule this out as a potential source of fatty acids contributing to TG
synthesis by the ISR. Regardless of the mechanisms, they will
undoubtedly be time-dependent and coordinated with baseline lipid
homeostasis pathways, including TG synthesis by the DGATs. Similar to
our findings with DGAT2, basal DGAT1 activity has been shown to protect
the ER from lipotoxic stress in adipocytes by re-esterifying released
fatty acids, a process whose upregulation becomes critical in
conditions of enhanced lipolysis^[328]70. It’s possible that DGAT2
activity may be similarly necessary to prevent the accumulation of
toxic, stress-inducing lipid species. Although the basal
stress-protective function of DGAT2 is not yet known, nor is the nature
of the stress sensors that are engaged in its absence, our findings
highlight the extent to which the ISR is integrated with the state of
the cellular metabolome and lipidome. Our findings add to a growing
number of reports that LD formation, through various mechanisms, is a
protective response to cell stress^[329]49 by preventing mitochondrial
dysfunction from fatty acid overload during starvation^[330]48,
sequestering misfolded proteins from the ER^[331]71, avoiding lipid
peroxidation and oxidative stress^[332]72,[333]73, maintaining membrane
saturation^[334]74 or serving as a long-term supply of fatty acids for
energy generation by β-oxidation^[335]48,[336]52,[337]75.
This pharmacogenomic approach establishes a footprint of ISR-sufficient
transcriptional, metabolic and cellular events that promote early
responses to cell stress. Although we used a construct derived from the
kinase domain of PERK, the specificity of PERK, PKR, HRI, and GCN2 in
targeting eIF2ɑ as their substrate, and the similarity of our
transcriptomic signature with one obtained from a PKR-derived
optogenetic tool^[338]76, together suggest that these downstream
outputs are independent of the specific ISR kinase domain. Indeed, that
many of the metabolic outputs downstream of Dmr-PERK activation were
also observed in N208Y-EIF2α MEFs, where the is ISR activated without
eIF2 phosphorylation, suggests that these responses are agnostic to the
nature of the activating kinase.
Understanding the core response of the ISR will allow us to better
understand the cellular state downstream of pleiotropic insults, where
these core outputs are modulated by inputs via different signaling
pathways. Although this footprint is derived from an immortalized human
cell line, we found that the major ISR-induced changes were conserved
in primary mouse fibroblasts, suggesting this may be a conserved
response. Indeed, the sensitive ATF4-dependent transcriptional module
is active in the astrocytes of Vanishing White Matter mice, where a
mutation in eIF2B causes chronic engagement of the ISR and
leukodystrophy, and is also represented in the ISR-associated GeneCLIC
signature^[339]30,[340]43. Data from the minimal cell line system
generates a hypothesis for what transcriptional and metabolic changes
to expect in vivo during ISR activation, including the newly uncovered
role of the ISR in LD biogenesis, an important finding to help dissect
the mechanisms of LD formation and function in stressed
tissues^[341]27,[342]77,[343]78. Intriguingly, many of the
neuropathologies associated with ISR activation are also associated
with the accumulation of LDs^[344]79,[345]80 and it will be critical to
parse the contribution of the ISR to their formation and whether this
is a protective or maladaptive response to these diseases. Thus, our
results using a minimal pharmacogenomic system will help us resolve the
mechanistic basis for the ISR’s contribution to cell- and
tissue-specific outcomes to injury, aging and disease.
Methods
Cloning
The Dmr-PERK expression vector was generated using the Lenti-X
iDimerizer Inducible Homodimerizer System (Takara). The mouse PERK
kinase domain (amino acids 537-1114) was obtained as a custom gBlock
from Integrated DNA Technologies. The pLVX-Hom-1 vector (Takara)
contains a CMV promoter driving expression of a dimerization domain,
and a multiple cloning site followed by an IRES sequence driving
expression of a Puromycin resistance cassette. pLVX-Hom-1 was digested
with BamHI and ligated with the mouse PERK kinase domain amplified with
corresponding BamHI overhangs. Ligation and transformation was
performed using the In-Fusion Cloning System with Stellar competent
cells (Takara). Clones expressing the correct vector sequence were
confirmed by Sanger sequencing.
Cell culture
U2OS cells were grown in Dulbecco’s modified Eagle’s medium (DMEM,
Gibco) supplemented with 10% fetal bovine serum (FBS, Sigma) and 1X
antibiotic-antimycotic solution (Gibco). Cells were housed in an
incubator at 37 °C, 5% CO[2], 20% O[2] and passaged every 2–3 days with
trypsin. In experiments comparing WT and ATF4 KO cells, cells of both
genotypes were grown in media supplemented with 1X NEAA (Sigma) and
55 nM beta mercaptoethanol (Gibco) for at least 24 h prior to
experimentation. For experiments in delipidated media, cells were
washed 3× and treated in DMEM supplemented with 10% lipid-depleted FBS
(Biowest LLC) and 1X antibiotic-antimycotic solution. Where indicated,
cells were treated with AP20187 dimerizer (Sigma Aldrich), 100 nM
oligomycin (Sigma Aldrich), 100 nM thapsigargin (Sigma Aldrich), 500 nM
ISRIB (Sigma Aldrich), 500 nM 2BAct (synthesized in house), 100 nM
Torin (Tocris).
Wild-type and N208Y MEFs were cultured in Dulbecco’s modified Eagle’s
medium (DMEM, Gibco) supplemented with 10% fetal bovine serum (FBS,
Sigma) and 1X antibiotic-antimycotic solution (Gibco) and 500 nM 2BAct.
Cells were housed in an incubator at 37 °C, 5% CO[2], 3% O[2] and
passaged every 2–3 days with trypsin. For withdrawal studies, MEFs were
acclimatized to 20% O[2] for 24 h then washed 3× with PBS and
replenished with fresh media with or without 2BAct. For metabolomics,
media was supplemented with 1X NEAA (Sigma) and 55 nM beta
mercaptoethanol (Gibco) for at least 24 h to mimic U2OS media
conditions in these experiments. All downstream assays were performed
in normoxic conditions.
Confluence measurements were performed by plating 0.012 × 10^6 cells
per well, in duplicates, in 96 well plates and treatment started the
next day. As indicated, cells were incubated with 10 μM Avasimibe
(Selleckchem), 20 μM T863 (Sigma Aldrich), 20 μM PF-06424439 (Sigma
Aldrich) or 2 μg/mL TOFA (Sigma Aldrich) immediately before
transferring cells to the Incucyte SX-5 Live-Cell Analysis (Sartorius).
Cells were imaged every 2 h for 96 h and confluence was quantified with
the Incucyte software using label-free segmentation methods.
Cell line generation
The U2OS Dmr-PERK cell line was generated by lentiviral transduction.
Dmr-PERK lentivirus was generated by delivering the
pLVX-Hom-1-Dmr-PERK-IRES-Puro vector to HEK293T using the Takara
Lenti-X Packaging Single Shot (VSV-G) system. Supernatant was collected
after 48 h and concentrated using Lenti-X Concentrator (Takara). U2OS
cells were transduced with varying concentrations of viral supernatant
and stable integrants were selected with media containing 3 µg/mL
puromycin. Stable clones were confirmed by Western blot analysis of
PERK at the expected protein size and by induction of ISR outputs after
treatment with dimerizer, including p-eIF2 AlphaLISA, and ATF4, GADD34,
and CHOP protein expression (Fig. [346]1 and Supplementary
Fig. [347]1).
U2OS Dmr-PERK ATF4 KO cells were generated using Alt-R CRISPR-Cas9
reagents from Integrated DNA Technologies according to the
manufacturer’s protocol. Predesigned crRNAs targeting ATF4 were
obtained from IDT (Hs.Cas9.ATF4.1.AA) and annealed with ATTO
550-labeled tracrRNA. RNPs were generated by combining with Cas9
protein (IDT) and reverse transfected into U2OS Dmr-PERK cells with
Lipofectamine RNAiMAX reagent. 24 h after transfection, cells were
trypsinized and sorted for ATTO 550-positive cells into single wells of
a 96-well plate. ATF4KO cells were grown in supplemented DMEM that
included 1X NEAA (Sigma) and 55 uM beta-mercaptoethanol (Gibco).
Individual clones were verified to be ATF4-null by sequencing and
confirmed by Western blot for ATF4 after ISR induction.
siRNA knockdown
U2OS Dmr-PERK cells were transfected with Silencer Select siRNA
(ThermoFisher) using the reverse transfection method. As indicated,
10 pmol of siRNA for DGAT1 (s16569), DGAT2 (112269) or eIF2B1 (s4558 or
s4559) was diluted in 100 μL Opti-MEM (Gibco) and added to the wells of
a 24 well plate followed by 1 μL of Lipofectamine RNAiMAX
(ThermoFisher) and incubated for 20 min at room temperature. 0.5 mL of
suspension of 100,000 cells/mL in complete DMEM was added to each well,
incubated for 48 h, then split into vessels for various downstream
assays.
AlphaLISA assay
Dmr-PERK U2OS cells were plated on 96-well plates and left to recover
overnight. Cells were treated in triplicate with indicated compounds
for 2 h and lysed in 50 μL 1X AlphaLisa lysis buffer. Lysates were
assayed in technical duplicates to measure the level of eIF2α
phosphorylation using the Total eIF2α AlphaLISA SureFire Ultra
Detection Kit (Perkin Elmer) following the manufacturer’s instructions.
For p70 S6K phosphorylation, lysates were assayed with the Alpha
SureFire Ultra Multiplex phospho/total p70 S6K Assay (Perkin
Elmer). Plates were read using a CLARIOstar Plus plate reader (BMG
Labtech) with standard AlphaLISA settings.
Immunofluorescence
Cells were plated in CellCarrier 96 well plates (PerkinElmer) to
achieve 80% confluence at the time of fixation. After treatment, cells
were washed once with PBS and fixed with 4% paraformaldehyde for
15 min. Cells were washed 3 times with PBS then permeabilized and
blocked in blocking buffer (PBS, 5% goat serum and 0.5% Triton X-100)
for 1 h. The following primary antibodies were diluted in labeling
buffer (PBS, 0.1% Tween 20 and 5% goat serum) (1:1000) and incubated
overnight at 4 °C with gentle rocking: anti-G3BP1 (BD), anti-CHOP
(1:1000, Cell Signaling L63F7). Following three 5-min PBST washes,
secondary antibodies were applied (1:2500) in blocking buffer and
incubated for 1 h. After three PBST washes for 5 min each, cells were
stained with DAPI (1:10,000) in PBS for 5 min. When co-staining with
BODIPY, Triton X-100 and Tween 20 were replaced with 0.05% saponin and
0.01% saponin, respectively. All steps were performed at room
temperature unless otherwise noted. For lipid droplet imaging, fixed
cells were labeled with 50 ng/mL BODIPY 493/503 and DAPI (1:5000) in
PBS for 15 min followed by three PBS washes. All labeling steps were
performed at room temperature unless otherwise noted and protected from
light throughout.
Cells were imaged in fresh PBS. Imaging was performed on the Opera
Phenix High Content Screening system (Perkin Elmer). Plates were imaged
with the 20× or 40× water objective with 9–25 fields of view imaged per
well. Nucleus, cell, and spot segmentation was performed using Harmony
software (Perkin Elmer). For quantification of LD content, eight random
fields were selected and a total of >100 cells were scored per
conditions for each of >3 independent experiments Percentage of cells
with SGs, mean number of spots per cell was calculated in R. Data
plotting, EC50 determination, and statistics were performed on GraphPad
Prism software.
High-content microscopy
U2OS Dmr-PERK cells were seeded in a 384-well PhenoPlate (PerkinElmer)
at a density of 4000 cells/well in DMEM supplemented with 10% FBS, 1X
antibiotic-antimycotic solution and 1X GlutaMAX (Gibco). The following
morning cells were washed using a BioTek EL406 Washer and labeled with
114 nM BODIPY 493/503 and 1X SPY650-DNA (Cytoskeleton, Inc) in
FluoroBrite DMEM (Gibco) supplemented as above. 3 h later cells were
dosed, as indicated, with AP20187 dimerizer (Sigma Aldrich), 20 μM T863
(Sigma Aldrich) or 20 μM PF-06424439 (Sigma Aldrich) using an Echo 555
or 655 liquid handler (Thermo Scientific) and all volumes normalized
with vehicle (DMSO). Six replicates per condition were arranged
randomly across 2 plates performed on different days.
Following compound dosing, cells were immediately transferred to a
pre-warmed Opera Phenix High Content Screening System running Harmony
version 5.1 for imaging. Cells were imaged every 20 min for 12 h and
then every 2 h up to 24 h using a 40× water objective with four fields
of view imaged per well. The imaging chamber was kept at 37 °C and 5%
CO[2] for the duration of imaging. Nuclei, cytoplasm and spot
segmentation were performed in the Harmony software on maximum
intensity projections of three slices for each field. Data plotting and
statistical analysis were performed in Python.
LD radius quantification was performed in Python 3.10. LDs were first
identified by creating maximum intensity projections of three slices
for each imaging field. We then ran a multi-scale Laplacian of a
Gaussian (LoG) filter over five sigma values ranging from 1 to 2.67
pixels in size to enhance spherical structures, keeping only those
pixels with a post-filter value above 0. The first LoG filtered image
corresponding to a sigma of 1 underwent a triangle threshold to create
a definitive LD semantic segmentation mask. A maximum filter was then
applied across the LoG stack, and local peak maxima were identified as
those pixels in the maximum filtered image whose intensity value
matched those same pixels in the LoG stack, corresponding to the
midpoint of individual lipid droplets. A border mask was then created
by performing a binary dilation of the LD semantic segmentation mask,
followed by a binary NOT operation on the original mask, leaving behind
only the outline of the LD mask. These border mask pixel coordinates
were then queried using a k-d tree to find each local max peaks’
nearest border pixel neighbor, giving the effective radius for each LD
in a computationally efficient and accurate way, as previously
described^[348]81. Any lipid droplet whose local max peak coordinates’
pixel was below a threshold value of 4000 was removed from the
analysis. The code is open-source and publicly available^[349]82.
Puromycin incorporation assay
To measure puromycin incorporation, the Click-iT Plus OPP Alexa Fluor
488 Protein Synthesis assay kit was used and manufacturer’s
instructions were followed, with a 30-min puromycin labeling. For
puromycin incorporation measurements by Western blot, cells were
labeled with 10 µg/mL puromycin for 30 min prior to harvesting and
protein analysis was performed as described below.
Protein expression analysis
Lysates were collected by washing cells once with PBS and harvesting in
RIPA buffer (ThermoScientific) with phosphatase and protease inhibitors
(Pierce). 4X Laemmli buffer was added to samples and boiled at 95 °C
for 5 min. For signaling time courses, cells were directly lysed in 2X
Laemmli sample buffer (BioRad) and boiled for 10 min.
Samples were run on a precast PAGE gel (Mini-PROTEAN TGX, BioRad) using
the Mini-PROTEAN Tetra Cell and then transferred to a nitrocellulose
membrane at 100 volts for 1 h using a Mini Trans-Blot Cell (BioRad).
Membranes were incubated in blocking buffer (5% milk or 5% BSA in TBST)
for 1 h at room temperature. The following primary antibodies was
diluted in blocking buffer and added to the membrane, incubating
overnight at 4 °C with gentle rocking: anti-p70S6K (Cell Signaling
9202) (1:2000), anti-phospho-p70S6K (T389) (Cell Signaling 9205)
(1:1000), anti-cleaved caspase-3 (Cell Signaling 9662s) (1:2000),
anti-cleaved caspase-8 (Cell signaling 9746t) (1:2000), anti-eIF2B1
(Proteintech 18010-1-AP), anti-actin (Cell signaling 3700) (1:2000).
Membranes were washed with TBST three times for 5 min each, and then
secondary antibody diluted in blocking buffer was added for 1 h at room
temperature. Membrane was washed three times with TBST and then
visualized with the Odyssey Infrared Imaging System (LI-COR
Biosciences) or detection reagents were applied, and imaging was
performed on a ChemiDoc imager (BioRad).
For automated Western blot analysis (Wes), samples were diluted to
0.4 mg/mL and 3 μL of sample was run on a ProteinSimple Wes capillary
system using a 12–230 kDa separation module with the Chemiluminescent
Detection module and using following primary antibodies: anti-ATF4
(1:100, Cell Signaling 11815), anti-GADD34 (1:100, Proteintech
104491-1-AP), anti-phospho-eIF2a (S51) (1:200, Cell Signaling 3398),
anti-eIF2a (1:200, Cell Signaling 5324).
nCounter gene expression analysis
Multiplex transcript expression levels were measured by nCounter
(Nanostring Technologies, Seattle, WA) with a custom panel containing
the 95 ISR GeneCLIC genes. 100–500 ng of purified total RNA was used
for nCounter gene expression analysis as instructed by the
manufacturer. Briefly, reporter and capture probes to the genes of
interest were hybridized to total RNA at 65 °C for 16 h. Hybridized
probes were then captured to the nCounter cartridge prior to imaging
and quantification.
Raw counts were background subtracted against negative control probes
then normalized against housekeeping genes (B2m, Gapdh, Hprt, Rpl19)
and positive control samples. The counts were log[2]-transformed and a
Z-score for each gene was calculated by subtracting the overall mean of
the control group from the sample and dividing that result by the SD of
all of the measured intensities. ISR pathway activation was measured by
taking the averages of the Z-score within each sample. GraphPad Prism
(La Jolla, CA) was used to perform statistical analyses utilizing one
way ANOVA with Dunnett’s Test performed post-hoc to correct for
multiple comparisons. Log[2] fold changes of ISR GeneCLIC genes were
clustered by correlation-based distance using the heatmap.2 function
from the gplots package.
Bioenergetics
1 × 10^4 WT or ATF4 KO U2OS cells were seeded per well in a Seahorse
96-well assay plate in DMEM with 10% FBS, 1% NEAA and 55 µM
β-mercaptoethanol. Medium was exchanged the following day with either
0.05 nM AP20187 dimerizer or ethanol vehicle. 16 h later, media was
exchanged to Seahorse XF DMEM with 2% dialyzed FBS, 2 mM glutamine,
25 mM glucose, 1X NEAA, and 55 µM β-mercaptoethanol ±0.05 nM dimerizer.
Oligomycin, FCCP, and Rotenone/Antimycin A were diluted in Seahorse XF
DMEM and added to the calibrant plate according to manufacturer
protocol. Mito Stress tests were run on a Seahorse Bioanalyzer with
Wave software (Agilent). Data was exported to GraphPad prism for
visualization.
RNA extraction and library preparation
Cells were seeded at 2.5 × 10^5 cells/mL in 24-well plates one day
prior to treatment. Treatments added to wells at indicated time points
prior to sample collection. Each condition and time point were measured
in triplicate. RNA was extracted using the MagMAX mirVana Total RNA
Isolation Kit in 96-well plate format and processed on the KingFisher
Flex Magnetic Particle Processor using the A27828_FLEX_Tissue_Cells
protocol. RNA quality and concentration was determined with the
Fragment Analyzer Standard Sense RNA kit. RNA-seq libraries were
prepared using the NEBNextUltra II Directional RNA Library Prep Kit for
Illumina with 1000 ng of input material. Samples were pooled for
sequencing on the NovaSeq instrument.
Transcriptomics processing and quantification
RNA-seq analysis was executed and visualized using an in-house,
web-based platform, consisting of the following steps. Sequencing
quality control was performed using FastQC (v0.11.5). Transcript
expression was then quantified using Salmon^[350]83 (v0.9.1) in
pseudo-alignment mode, without adapter trimming, producing
transcript-per-million estimates, using the Ensembl human GRCh38
transcriptome. For the dose response study, samples were corrected by
removing the second principal component that captured variation due to
batch. For visualization and time course analyses, low count genes
(<25) were excluded. Differential expression analysis was performed
using the DESeq2 package in R producing q values and effect sizes for
the comparison between basal states of WT and ATF4 KO Dmr-PERK cells.
Transcriptomics time course analysis
To identify genes that were significantly changing over time in
response to ISR induction, we used the method described in ref.
^[351]84 First, each gene measurement was normalized to its mean
expression level at t = 0. Then, genes were selected if they showed a
log[2] fold change greater than 1.5 in any condition. Of these, the
expression profile for each gene was fit to a quadratic model:
[MATH:
Y(t)=
β1t+β2t2 :MATH]
Time-dependent genes required a p value of the F statistic of the model
to be <0.001. To perform pairwise comparisons of gene expression time
courses between genotypes (WT vs. ATF4 KO), a multiple regression
analysis was performed. We fit an additional model that took into
account the genotype with the variable
[MATH: DG
:MATH]
:
[MATH:
Y(t)=
β1t+β2t2+<
msub>DG<
mi>β3t+DGβ4t
2 :MATH]
where
[MATH:
DG=0 :MATH]
for WT and
[MATH: DG
:MATH]
= 1 for ATF4 KO. We computed the p value of the F statistic between the
original and genotype-aware models. Genes were classified as
significantly different if the Benjamini and Yekutieli FDR-corrected p
value < 0.05. The same procedure was performed to identify genes
differentially expressed in two treatment conditions, with a p value
cutoff of 1e-5 (dimerizer vs. thapsigargin conditions in Fig. [352]1f
and the dimerizer vs. arsenite conditions in Fig. [353]1g).
To generate heatmaps, protein-coding genes were included (non-protein
coding genes are included in raw data files and Supplementary Data) and
hierarchical clustering and visualization was performed using the
heatmap.2 function in R. Overrepresentation analysis of Gene
Ontology-Biological Process terms for the genes in each cluster was
performed using the R function enrichGO from the ClusterProfiler
library^[354]85.
LC-MS/MS for polar metabolites
Cells were washed with room temperature PBS followed by quenching in
80% methanol containing internal standards (D4-L-tyrosine,
15N4-L-arginine and D5-benzoic acid; Sigma-Aldrich) maintained at
−20 °C. Cells were scraped, transferred to microcentrifuge tubes, then
centrifuged at 16,000 × g at 4 °C for 15 min. Supernatants were dried
down under nitrogen at 4 °C. Dried extracts were resuspended at 5×
concentration in 40% methanol/40% acetonitrile/20% water containing
additional standards (D5-L-phenylalanine, D4-L-lysine and D5-succinate;
Sigma-Aldrich). Metabolomics samples were analyzed in both positive and
negative ESI-LC-MS methods on Vanquish UPLCs coupled to Q-Exactive Plus
mass spectrometers. Metabolites were separated using a SeQuant®
ZIC-pHILIC column (5 μm, 200 Å, 150 × 2.1 mm) where the mobile phase A
was 20 mM ammonium carbonate in water (pH 9.2) and mobile phase B was
95:5 (v/v) acetonitrile/mobile A at a flow rate of 150 μL/min and the
gradient was t = −6, 84.2% B; t = 0, 84.2% B; t = 2.5, 76.8% B; t = 5,
68.4% B, t = 7.5, 60% B; t = 10, 52.6% B; t = 15, 36.8% B; t = 20; 21%
B; t = 22, 15.8% B; t = 22.5, 84.2% B; t = 24; 84.2% B. Data were
acquired using data-dependent acquisition (DDA) mode with the following
parameters: resolution = 70,000, AGC target = 3.00 × 10 maximum IT
(ms) = 100, scan range = 70–1050. The MS2 parameters were as follows:
resolution = 17,500, AGC target = 1.00 × 10^5, maximum IT (ms) = 50,
loop count = 6, isolation window (m/z) = 1, (N)CE = 20, 40, 80;
underfill ratio = 1.00%, Apex trigger(s) = 3–10, dynamic
exclusion(s) = 25. For negative mode, (N)CE = 20, 50, 100.
For ^13C-glucose and ^13C-glutamine tracing studies, media was
exchanged to glucose-free DMEM (Gibco) with addition of 25 mM
U-^13C6-glucose (Cambridge Isotopes) or glutamine-free DMEM (Gibco)
with addition of 2 mM U-^13C5-glutamine (Cambridge Isotopes),
respectively, at different timepoints prior to harvest to determine
steady state labeling of intracellular metabolites. Dialyzed fetal
bovine serum (Corning) was utilized to reduce unlabeled glucose and
glutamine derived from serum.
Relative fatty acid synthesis rates
For ^2H[2]O tracing into lipid palmitate (C16:0), 8% deuterium oxide
(Sigma) or 8% water was added to standard culture medium and media was
exchanged 24 h prior to harvest. ^13C-glucose and ^13C-glutamine
tracing were performed by replacing standard medium with medium
containing 100% U-13C-glucose or U-13C-glutamine in separate cultures,
24 h before harvest. Lipids were extracted with isopropanol, and
saponified by incubating in 0.6 M 75% KOH for 45 mins at 60 °C,
neutralized with 25% acetic acid, dried under nitrogen and resuspended
in 200 µL of ethanol.
FA analysis was performed in negative ion mode using LC-MS method of a
Vanquish UPLC-Q-Exactive Plus. Compounds were separated using a
Phenomenex Kinetex C8 (2.6 µm, 100 ×2.1 mm) with a column guard UHPLC
C8 2.1 mm. Method was adapted from ref. ^[355]86. Label incorporation
of label was determined using MAVEN2^[356]87, which identified isotopic
envelopes using expected MS1 m/z shifts, and quantified isotopic peaks
either by standard peak quantitation approaches or by full integration
of observed intensity within the [M + 0] RT peak bounds applied to each
isotopic m/z. Natural ^13C isotope abundance was corrected using
MAVEN2’s natural isotopic abundance correction functionality and the
IsoCorrectoR R package^[357]88.
MTBE-LLE extraction for lipidomics
5e5 500,000 cells per well was washed with warm PBS and quenched with
0.8 mL of 50% MeOH/H[2]O containing internal standards (100 µL of
LipidSplash standards (Avanti Polar Lipids)) for 15 min at −20 °C.
Samples were scraped and collected to a 2 mL glass vial and extracted
for lipidomics analysis using methyl tert-butyl ether liquid-liquid
extraction (MTBE-LLE) adopted from Matyash et al.^[358]89 In short,
800 μL of MTBE was added to the mixture, vortexed for 30 s and
incubated on ice for another 15 min and centrifuged at 3000 g for
10 min at 4 °C. Lipids partitioned in the top layer were collected into
a separate vial, and the extraction process was repeated with an
addition of 600 µL of MTBE. After incubating and centrifuging, the
second organic layer was collected and combined in the first lipid
vial, dried under nitrogen at 4 °C, resuspended in 200 µL of
ButOH/MeOH/H[2]O (2:1:1, v/v/v) for analysis.
LC-MS for lipidomics
Cells were cultured in six-well plates, and ~500,000 cells per well
were washed with warm PBS and quenched with 0.8 mL of 50/50 MeOH/H[2]O
containing 100 µL of LipidSplash internal standards (Avanti Polar
Lipids), followed by incubation at −20 °C for 15 min. Samples were
scraped and collected to a 2 mL glass vial and extracted for lipidomics
analysis using an MTBE-LLE extraction adopted from Matyash et
al.^[359]89. In brief, 800 μL of MTBE was added to the mixture,
vortexed for 30 s and incubated on ice for another 15 min and
centrifuged at 3000 g for 10 min at 4 °C. Lipids partitioned in the top
layer were collected into a separate vial, and the extraction process
was repeated with an addition of 600 µL of MTBE. After incubating and
centrifuging, the second organic layer was collected and combined in
the first lipid vial, dried under nitrogen at 4 °C, resuspended in
200 µL of ButOH/MeOH/H[2]O (2:1:1, v/v/v) for lipid analysis.
Lipidomics samples were analyzed in both positive and negative ion
modes using the same LC-MS method consisting of a Vanquish UPLC coupled
to a Q-Exactive Plus mass spectrometer. Lipids were separated using a
Thermo Scientific Accucore C30 column (2.6 μm, 150 Å, 2.1 × 250 mm) at
a flow rate of 200 μL/min. Mobile phase A was 20 mM ammonium formate in
60:40 acetonitrile:water +0.25 μM medronic acid, mobile phase B was
20 mM ammonium formate in 90:10 isopropanol:acetonitrile + 0.25 μM
medronic acid. The gradient was t = −7, 30% B, t = 0, 30% B, t = 7, 43%
B, t = 12, 65% B, t = 30, 70% B, t = 31, 88% B, t = 51, 95% B, t = 53,
100% B, t = 55, 100% B, t = 55.1, 30% B, t = 60, 30% B. The mass
spectrometer settings were as follows: DDA was performed with the
following parameters: resolution = 140,000, AGC target = 3.00 × 10^6,
maximum IT (ms) = 100, scan range = 200–2000. The MS2 parameters were
as follows: resolution = 17,500, AGC target = 3.00 × 10^6, maximum IT
(ms) = 150, loop count = 8, isolation window (m/z) = 1, (N)CE = 20, 30,
40; underfill ratio = 1.00%, Apex trigger(s) = 5–30, dynamic
exclusion(s) = 15 s.
Raw files were converted to mzML files using msconvert from
ProteoWizard^[360]90, using vendor centroiding on all scans, and
analyzed using MAVEN2 software^[361]87. Identification was performed by
spectral matching using in-house spectral libraries^[362]87.
Statistical analysis for metabolomics and lipidomics
Analysis of the lipidomics and metabolomics data was performed using
Calico Lipidomics and Metabolomics Analysis (clamanR) package
([363]https://github.com/calico/claman) in R. Raw files were converted
to the mzML format using ProteoWizard Ver 3^[364]90. Compound
identifications were detected and grouped using the OpenCLaM R package
([365]https://github.com/calico/open_clam) and manually inspected using
the MAVEN2 peak analysis program
([366]https://github.com/eugenemel/maven)^[367]87 with the criteria of
a precursor ion tolerance of 10 ppm and a product ion tolerance of 20
ppm, comparing fragmentation and retention time to an in-house library
generated from authentic standards for metabolomics, and in-house
generated in-silico libraries for lipidomics
([368]https://github.com/calico/CalicoLipidLibrary). The sum of the the
raw scan intensity of the smoothed maximum, as well as the scan
immediately preceding and following the smoothed maximum scan, was
taken as the peak quantity, hereafter referred to as “peak area top”.
Peak area top of each feature underwent log[2] transformation, peaks
below the limit of detection were assigned a value of 12. Each compound
was normalized to the average signal of that compound in the reference
condition at t0. Heatmaps were generated using the pheatmap package in
R. Relative abundance is shown as log[2] fold change (lipids) or raw z
scores (metabolites) as calculated using the row scaling function in
pheatmap.
A linear model was fit to the data and a pairwise comparison between
treatments at individual time points was performed for each identified
feature using this model:
[MATH:
yik=α+β1g
i+β2<
/mn>,ktk+β
3,ktkg
i :MATH]
, where
[MATH:
yij
:MATH]
are the normalized log[2] to the reference condition for the annotated
lipids and metabolites. Observation indenced by i = 1 are from WT U2OS,
and i = 2 from ATF4-KO, so that
[MATH: β1
:MATH]
are expected to be the difference between genotypes. k = 0, 1, 2, 4, 8
or 24 indexes the time-course study of cells in dimerizer. For lipids,
the model excludes out the genotype differences since the experiment
was done only in WT-U2OS. P values from these regressions were False
Discovery Rate controlled on a term-by-term basis using the qvalue R
package adapted from the Storey lab^[369]91. Significant changes were
reported for q values < 0.01.
Absolute quantification of lipid species and lipid classes was done by
normalizing peak area top of each species to the lipid standards
included in LipidSplash spiked into samples prior to LLE. For lipid
enrichment analysis, we ranked all identified species using the
estimated values calculated from the regression model. Identified lipid
species were grouped by lipid class and acyl chain composition.
Relative enrichment of differentially abundant lipid species was
measured using the R package FGSEA^[370]92,[371]93. We report the
−log[10] Benjamini-Hochberg corrected p value and use the normalized
enrichment score to assign the direction of the change.
Relative metabolic flux to glutathione biosynthesis was calculated from
^13C-tracing data by determining the first order rate constants for
disappearance of unlabeled gamma-Glu-Cys and glutathione, as previously
described^[372]94, but using the relative, mass spectrometry-derived,
peak sizes in place of the absolute pool sizes.
Reporting summary
Further information on research design is available in the [373]Nature
Portfolio Reporting Summary linked to this article.
Supplementary information
[374]Supplementary Information^ (16.3MB, pdf)
[375]Peer Review File^ (893KB, pdf)
[376]41467_2024_52538_MOESM3_ESM.pdf^ (33.3KB, pdf)
Description of Additional Supplementary Files
[377]Supplementary Data 1^ (16.3MB, xlsx)
[378]Supplementary Data 2^ (3.9MB, xlsx)
[379]Supplementary Data 3^ (5.3MB, xlsx)
[380]Supplementary Data 4^ (44KB, xlsx)
[381]Supplementary Data 5^ (321.5KB, xlsx)
[382]Reporting Summary^ (3.4MB, pdf)
Source data
[383]Source Data^ (2.9MB, xlsx)
Acknowledgements