CRISPuRe-seq: pooled screening of barcoded ribonucleoprotein reporters reveals regulation of RNA polymerase III transcription by the integrated stress response via mTOR

CRISPuRe library construction

Each library plasmid contains the following relevant parts: third-generation lentiviral expression components, a single-guide RNA (sgRNA) driven by the human U6 promoter on the positive strand, an MS2 hairpin-containing small BarRNA driven by the human 7SK promoter on the negative strand, and a puromycin-resistance gene driven by the EF-1α promoter (see sample supplemental file Nontarget_ci_MS2_7sk.gb). The BarRNA contains a single MS2 hairpin, a 25-nucleotide random barcode, and primer binding sites for reverse transcription and subsequent polymerase chain reaction (PCR) amplification.

We generated CRISPuRe plasmid libraries by incorporating a BarRNA expression cassette into an already existing CRISPRi plasmid library (see sample supplemental file Initial_CrisprI_Library.gb). An existing library of CRISPRi plasmids was digested with BamHI and EciI to liberate a library of fragments containing the sgRNAs and a portion of the ampicillin-resistance gene. A plasmid containing a generic BarRNA expression cassette was digested with EcoRI and FspI to liberate a second fragment containing half of a complete BarRNA (minus the barcode), the puromycin-resistance gene, and an overlapping piece of the ampicillin-resistance gene. A third fragment containing a random N₂₅ barcode was generated using Klenow-based overlap extension of two oligonucleotide Ultramers from Integrated DNA Technologies (IDT) [16]. These three pieces were pooled together and combined using Gibson Assembly (NEB Gibson Assembly Master Mix-E2611L).

A whole genome library targeting a total of 18946 genes was divided into nine sublibraries, including an inflammation/innate immunity sublibrary containing 5930 sgRNA covering ∼1100 genes. Each sublibrary was electroporated separately into MegaX electrocompetent cells (Thermo Fisher Scientific, C640003) and was plated on large agar plates. Colonies from the plates were scraped and plasmid was purified (ZymoPure II Plasmid Maxiprep-Zymogen-D4203). Each library was cloned at a depth such that each protospacer is paired with ∼30 unique barcodes. While there are many barcodes per protospacer, there are ∼1% of barcodes shared between protospacers. Once constructed, the plasmid pool was amplified by PCR and the inserts were sequenced at ∼600× coverage relative to the estimated number of colonies to generate a lookup table of protospacer–barcode pairings.

Lookup table construction

Each sublibrary containing BarRNA and sgRNA for coding genes was sequenced using a HiSeq 4000. Raw sequence reads were demultiplexed and putative barcode/protospacer pairs were detected using string matching for the sequence surrounding each barcode and protospacer. Putative barcode/protospacer pairs with fewer than three sequence reads were removed. The remaining list was then compared with the list of known protospacers and only those barcode/protospacers with an identical protospacer match were kept. Barcodes that were shared between multiple protospacers were removed.

Construction of ISRE and ATF4 reporter cell lines

Two reporters were cloned containing either five or seven copies of an Interferon Stimulated Response Element (ISRE) [17] and were inserted into a lentiviral expression construct upstream of a TATA-box containing minimal promoter and the MCP:mClover3:Flag(3X) reporter coding sequence (see supplemental files pCDH_ISRE_5X_CPuRe.gb and pCDH_ISRE_7X_CPuRe.gb).

The ATF4 translational CRISPuRe reporter was created by cloning the uORFs from the endogenous ATF4 locus between the CMV promoter and a Clover-MCP-Flag(3X) sequence (see sample supplemental file DH_CMV_atf4uORFClover_MCPFlag_EF1_Blast.gb).

Lentivirus was generated by transfection into HEK293T cells and viral supernatant was subsequently used to infect CRISPRi-enabled K562 cells carrying KRAB-dCas9. Cells were selected with blasticidin and clonal populations were generated. A 7X ISRE reporter line was selected based on strong mClover3 fluorescence. A 5X ISRE reporter line was subsequently selected based on lower mClover3 fluorescence in response to interferon alpha (IFN-α) relative to the 7X ISRE cell line. An ATF4 reporter was selected based on an expression profile similar to the 5X ISRE reporter when treated with various ISR activators.

ISRE screen procedure

Fifty million K562 cells stably expressing KRAB-dCas9 and carrying an ISRE reporter were transferred to a T175 flask containing 100 ml of RPMI 1640 (Roswell Park Medical Institute), 10% Fetal Bovine Serum (FBS), GlutaMAX, and 8 µg/ml polybrene media. Six milliliters of lentivirus-containing supernatant (titrated for a multiplicity-of-infection (MOI) of ∼0.3) were added (AM Day 0). After 24 h of infection, cells were spun down and virus-containing medium was replaced with fresh RPMI, 10% FBS, and GlutaMAX and transferred back to a 250-ml spinner flask (AM Day 1). Puromycin selection at 1 µg/ml was initiated 24 h later (AM Day 2). Selection was continued for 2 days (Days 2 and 3). Cells were spun down and puromycin-containing medium was removed and replaced with medium without puromycin (AM Day 4). Cells were maintained in RPMI media at a concentration below 1 million cells/ml for four more days.

On treatment day (PM Day 9), cells were split into single T75 flask (25 million cells) at 0.5 × 10⁶ cells/ml for each technical replicate. Cells were treated with IFN-α (PBL Assay Science; 11200) overnight. On screening day (AM Day 10), cells were spun down for 10 min at 500 rcf at 4°C. Cell pellets were resuspended in 1 ml of cold phosphate-buffered saline (PBS) and then spun down again. PBS was removed prior to cell lysis.

ATF4 reporter screen procedure

Five hundred million K562 cells stably expressing KRAB-dCas9 and carrying the ATF4 reporter were split into 10 × T175 flasks containing 80 ml of RPMI 1640, 10% FBS, GlutaMAX, and 8 µg/ml polybrene media. To each flask, 6 ml of lentivirus-containing supernatant (corresponding to an MOI of 0.3) were added (AM Day 0). After 24 h of infection, cells were combined and spun down in two 500 ml centrifuge tubes. Virus-containing medium was replaced with fresh RPMI, 10% FBS, and GlutaMAX and cells were transferred back to a 2-l spinner flask (AM Day 1). Puromycin selection at 1 µg/ml was initiated 24 h later (AM Day 2). Selection was continued for 2 days (Days 2 and 3). Cells were spun down and puromycin-containing medium was removed and replaced with medium without puromycin (AM Day 4). Cells were allowed to grow for 24 h in the absence of puromycin prior to any screening.

On a screening day (AM Day 5), 100 million library-carrying cells were split into two T175 flasks at 0.5 × 10⁶ cells/ml. Cells were treated with drug for 4 h, split into two sets of 2 × 50 ml conical tubes, and spun down for 10 min at 500 rcf at 4°C. Cell pellets were resuspended in 1 ml of cold PBS and then the cells originating from a common flask were recombined into a single conical tube. Both tubes (one representing each treatment flask) were spun down again and the PBS was removed.

Cell lysis for screening

Fifty million cells were resuspended in 1 ml of cold (4°C) hypotonic lysis buffer [20 mM HEPES (pH 7.9), 2 mM MgCl₂, 10% glycerol, 0.1% IGEPAL, 0.5 mM DTT, 150 mM NaCl, protease/phosphatase inhibitor (Halt Protease and Phosphatase Inhibitor Cocktail; Thermo Fisher Scientific, P178444) with “blocking” hairpin RNA (IDT)] at a concentration of 10 µg/ml. Cells were left on ice for 30 min for complete lysis. Trituration beyond complete resuspension is not needed. Cells were then spun at 7500 rcf at 4°C for 5 min to pellet nuclei and cell debris.

Total RNA isolation

One hundred and fifty microliters of supernatant were transferred to 1 ml of TRIzol (TRIzol Reagent; Life Technologies, 15596018). Two hundred microliters of chloroform were added to each 1 ml tube of TRIzol/sample. Samples were vortexed for 15 s and incubated at room temperature for 2–3 min. Samples were transferred to Phasemaker tubes (Life Technologies, A33248) for phase separation. Phasemaker tubes were spun for 5 min at 12 000 rcf. The aqueous phase was transferred to a new tube. We then added 1 μl of GlycoBlue (Life Technologies, AM9516) and 500 μl of isopropanol to each tube. We then incubated the tubes at −80°C for 30 min or until the sample was frozen. We then thawed the sample and spun it for 10 min at 12 000 rcf at 4°C. Supernatant was removed and the pellet was washed twice with 1 ml of 75% ethanol, briefly vortexing and spinning at 7500 rcf for 5 min for each wash. Ethanol was removed by pipetting and the pellet was allowed to air dry. The total RNA pellet was resuspended in 20 μl of water.

Small RNA enrichment

To improve the amplification efficiency of the BarRNA from the total RNA fraction, we performed a small RNA enrichment step prior to complementary DNA (cDNA) synthesis. For each sample, we performed two enrichments in parallel. Fifteen micrograms of total RNA were diluted to volume of 100 μl in water. We then added 70 μl of SpriSelect magnetic beads (Beckman Coulter, Inc., B23317) and mixed by pipetting. We let this sit for 5 min at room temperature. We then moved the sample to a magnetic rack and allowed the beads to fully stick to the side of the tube. We then transferred the supernatant containing the enriched smaller RNAs to a new tube.

We then performed a second isopropanol precipitation on the enriched fraction by adding 600 μl of water, 1 μl of GlycoBlue, and 750 μl of isopropanol. We incubated the sample at −80°C for 1 h and continued with the RNA precipitation as above. The final small RNA-enriched fraction was resuspended in 10 μl of water.

Immunoprecipitation

After cell lysis, supernatant was incubated with 25 μl of M2 anti-FLAG magnetic beads (Sigma–Aldrich, IPFL00005) rotating at 4°C for 90 min. Beads were washed (inverted three times) with 1 ml of cold wash buffer (lysis buffer without IGEPAL) three times and were then resuspended in 20 μl of RNAse-free water.

cDNA synthesis for screen analysis

cDNA synthesis was performed following the Superscript II Reverse Transcriptase (Thermo Fisher Scientific, 18064014) or Maxima H Minus Reverse Transcriptase (Thermo Fisher Scientific, EP0752) protocol, with minor modification. For cDNA synthesis from the immunoprecipitation (IP) fraction, either we used the resulting bead solution (including the beads) as template in a 40-μl reaction or we eluted the BarRNA-bound reporter protein from the beads using 3X Flag Peptide following the manufacturer’s batch protocol (optimized protocol). RNA from the eluate was purified using TRIzol and isopropanol precipitated in the presence of Glycoblue. RNA was then resuspended in 10 μl of H₂O and used as template for reverse transcription (optimized protocol).

For the total RNA fraction, we used all 10 μl of the small RNA-enriched fraction as template. Similarly, we used all 10 μl of the eluted IP fraction. The amount of added reverse transcription (RT) primer, which contained a unique molecular identifier (UMI) sequence, was 5 pmol instead of 2 pmol per 20 μl of reaction. Following incubation at 42°C for 50 min, 1 μl of EXO-SAP-IT Express (Life Technologies, 75001.1.ML) was added to the reaction. This reaction was incubated at 37°C for 5 min before being inactivated at 80°C for 5 min.

cDNA reactions containing beads were raised to 95°C for 2 min following EXO-SAP-IT inactivation and then quickly moved to a bar magnet. The supernatant containing the cDNA reaction was then transferred to a new tube and subsequently used for sequencing library preparation.

Sequencing library construction

Sequencing library construction was carried out in two rounds of PCR. Round 1 amplified the cDNA and attached common adapters to each product. Round 2 further amplified and added specific indices for Illumina sequencing. Five microliters of cDNA reaction were used in each of 4 × 50 μl Round 1 PCRs per sample using Phusion polymerase in HF buffer (Life Technologies, F531L). The four Round 1 PCRs were then pooled and 5 μl of that mix was used as template for a 50-μl Round 2 PCR. The Round 2 PCR underwent 12 amplification cycles. Each reaction was then either SPRI-purified or run on a 1.8% agarose gel and gel-purified. Primers used for reverse transcription and library construction can be found in Supplemental_File_1.xls.

Screen barcode processing

Barcode and UMI sequences were extracted from each demultiplexed fastq.gz file using a custom bash script (CRISPuRe_mageck2.sh). For each condition, barcodes were matched to their corresponding protospacer sequence and the number of unique UMI/barcode pairs for each protospacer was counted.

For barcode distribution analysis, log₂ ratios were calculated for each barcode. This is the log₂ of the percentage of a given barcode in the IP fraction divided by the percentage of the same barcode in the total RNA fraction. Adjusted P-values were assigned to each sgRNA using the Wilcoxon rank sum test based on a comparison of the barcode log₂ ratio distribution for that sgRNA and a null distribution. The null distribution in this case was generated using barcodes associated with nontargeting negative control sgRNA. A hypergeometric test was performed on the list of statistically significant sgRNA (adjusted P-value < 0.05) to identify gene enrichment.

For protospacer-level analysis, unique barcode counts were summed for each corresponding protospacer sequence. Protospacer counts from paired samples (total RNA versus IP RNA) were compiled into a count table. This count table was then processed using MAGeCK [18] to determine which genes/perturbations were enriched for each sample. All screens were processed as individual replicates in MAGeCK unless otherwise stated, in which case multiple replicates were processed using MAGeCK’s “paired” function. MAGeCK output from all screen processing can be found in Supplemental_File_1.xls.

Protein–barcode association

HEK293T cells that stably expressed either MCP-mClover3-Flag(3X) (see supplemental file pCDH_CMV_CPuRe.gb) or MCP-mClover3 (see supplemental file pCDH_CMV_CPuRe_NoFlag.gb) were generated by lentiviral transduction. MCP-mClover3-Flag(3X) cells were transduced with lentivirus for the expression of a single sgRNA–barcode (target) pair. MCP-mClover3 cells were transduced with lentivirus for the expression of a pool of many sgRNA–barcode (competitor) pairs. Target barcode- and competitor barcode-expressing cells were mixed at an intended ratio of 1:100 in a total of ∼10 million cells. Cells were lysed in a hypotonic lysis buffer in the presence or absence of “blocking RNA”, cytosolic fractions were isolated by centrifugation, and FLAG IP was performed for 1 h. A 50-μl aliquot of cytosolic fraction was used to quantify total “target” and “nontarget” barcode levels in the total RNA.

For the protein–barcode stability analysis, FLAG IP was performed on the cytosolic total RNA of 1 million MCP-mClover3-Flag(3X) target barcode-expressing cells. After 1 h of binding to the IP beads, the target total RNA was replaced with cytosolic total RNA from 10 million MCP-mClover3 nontarget barcode-expressing cells. The nontarget total RNA and bead mixture was rotated at 4°C for increasing amounts of time. At each final timepoint, the beads were washed three times and were then further processed for cDNA sequencing and subsequent analysis. All sequencing for initial protein–barcode associations were performed on an Illumina MiSeq.

Flow cytometry

For each sample, 500 000 cells were spun down in a 1.5-ml tube. Each sample was washed once in PBS and then resuspended in 100 μl of PBS and 1 μl of anti-CD47 PerCP-Cy5.5 conjugated antibody. Cells were incubated at 4°C for 1 h and then washed once with 1 ml of PBS. Cells were spun down and resuspended in 100 μl of PBS and 2% FBS. Cells were then analyzed on a BD Fortessa using the PerCP-CY5.5 filter.

FACS screen

Twenty million 5X ISRE reporter K562 cells that had been treated for 24 h with 36 U/ml IFN-α were resuspended in 4 ml of PBS and 2% FBS at 4°C. Approximately 10 000 cells were used to draw sorting gates based on fluorescence in the Fluorescein isothiocyanate (FITC) channel and normalized by cell size via forward scatter (FSC). The highest and lowest expressing 25% (each) of cells were sorted separately for further processing [19].

Total RNA isolation for qPCR

Total 500 000 cells in 1 ml of RPMI media were aliquoted into one well of a 12-well dish. Cells were treated for 4 h with either 20 nM oligomycin (Sigma–Aldrich, 75351-5MG), 300 nM thapsigargin (Sigma–Aldrich, T9033-1MG), or 100 µM ML-60218 (MedChem Express, HY-122122). Medium was removed and cells were resuspended in 500 μl of TRIzol reagent for 10 min at room temperature. A total of 100 μl of chloroform was added and each tube was vortexed for 15 s. The mixture was transferred to Phasemaker Tubes and spun for 5 min at 12 000 rcf at 4°C. The aqueous solution was transferred to a new 1.5-ml tube and isopropanol-based RNA precipitation was performed following the TRIzol manufacturer’s protocol. The resulting pellet was resuspended in 22 μl of molecular biology grade water.

cDNA synthesis for qPCR

A total of 500 ng of RNA was combined with 1 μl of 10 mM dNTP mix and 1 μl of custom RT primer mix, and this mixture was diluted to a total volume of 15.75 μl with water. Sample was heated at 65°C for 5 min and then brought to 4°C in a thermal cycler. Then, 4 μl of 5× RT buffer and 0.25 μl of Maxima H Minus Reverse Transcriptase were added to bring the total volume to 20 μl. The sample was then incubated at 50°C for 30 min and then brought to 37°C in a thermocycler. A total of 0.5 μl of EXO-SAP-IT Express was then added to each sample which was then incubated at 37°C for 5 min, followed by a 15 min incubation at 80°C.

Each sample was diluted 1:100 prior to quantitative PCR (qPCR) analysis. A total of 4 μl of this 1:100 cDNA dilution was then added to 6 μl of pre-allocated primer-containing PowerUp SYBR Green Master Mix (Life Technologies, A25776) that was previously added to each well of a 384-well plate (MicroAMP Optical 384-Well Reaction Plate; Life Technologies, 4309849). qPCR was performed on a QuantStudio 6 Flex qPCR machine (Applied Biosystems).

Protein sample preparation for western blots

A total of 500 000 cells in 1 ml of media were treated in a single well of a 12-well dish for 4 h.

Cells were washed once in cold PBS and the cell pellet was resuspended in 50 μl of RIPA buffer (Life Technologies, 89901) with HALT Protease/Phosphatase Inhibitor and 0.05 μl of benzonase (MilliporeSigma, 70664-3). Cells were left on ice for 15 min. Samples were spun at 12 000 rcf for 10 min at 4°C and 50 μl of supernatant was transferred to a new tube.

Phos-tag western blots

A total of 30 µg of protein (∼1.5 μl) was diluted with 4× Laemmli buffer (with Beta-mercaptoethanol) (Bio-Rad, #1610747) and water to a total of 8 μl. We then loaded the sample onto a 12.7% SuperSep Phos-tag gel (VWR, 103258-494 or Thermo Fisher Scientific, NC1140699) such that every lane of the gel contained either sample or Laemmli buffer only. The gel was run at 100 V for 13 min until the dye front formed a thin line and then for an additional 75 min at 150 V. The gel was removed from the glass plates and placed in standard Tris/glycine transfer buffer (Bio-Rad) with 20% methanol and 10 mM ethylenediaminetetraacetic acid (EDTA). The gel was incubated on a rocker for 15 min at room temperature. This buffer was then replaced and the gel was incubated again for another 15 min in the same buffer. The buffer was removed and replaced with Tris/glycine buffer with 20% methanol (no EDTA) and incubated for 10 min. Then, the buffer was replaced with fresh Tris/glycine buffer with methanol and incubated for another 10 min. An overnight (16 h at 30 V, 4°C) wet-tank transfer onto Immobilon-FL PVDF membrane (Sigma–Aldrich, IPFL00005) was then performed. After transfer, the membrane was allowed to dry for 1 h. The membrane was then re-wet by soaking in methanol for 30 s and a standard western blot was performed.

Western blots

Approximately 50 µg of protein (3 μl) in 10 μl of sample buffer was loaded into each lane of a 4%–20% SDS–PAGE gel (Bio-Rad, 4561096). Samples were loaded in biological triplicate. Gels were run for 40 min at 200 V. Protein was then transferred to a PVDF membrane using the Trans-Blot Turbo device (Bio-Rad) using the 7 min mixed molecular weight program. After transfer, blots were allowed to dry for 1 h. Blots were then re-wet using methanol for ∼30 s and then transferred to LI-COR Intercept Blocking Buffer containing Tris-Buffered Saline (TBS) (LI-COR-927-60003). After blocking for 1 h, fresh blocking buffer with primary antibody at the appropriate dilution was added to the membrane following LI-COR’s recommendations. Blots were incubated in primary antibody overnight at 4°C. Primary antibody was removed followed by a brief wash with TBS + .1% Tween 20 (TBST) and then three 5-min TBST washes. Blots were then incubated with LI-COR infrared fluorescent secondary antibodies for 1 h. The blots were then washed as before. After the final wash, TBST was replaced with TBS. Blots were then imaged on a LI-COR Odyssey fluorescent imager.

Western blots were quantified using densitometry in ImageJ. All western blots apart from MAF1 were quantified by comparison to an appropriate loading control. MAF1 Phos-tag gel westerns did not require a loading control as relative signals between bands within a single lane were being compared. Phosphorylation of p70 S6K was quantified by performing parallel western blots with either anti-phospho p70 S6K and anti-B-actin or anti-total p70 S6K and anti-B-actin. The phospho and total bands were normalized to their corresponding B-actin loading controls. Phosphorylation was then represented as a fraction of total. For SESN2 and SESN2; DDIT4 cells, anti-p70 S6K CST #9205S was used instead of CST #97596 as replacement lots no longer showed immunoreactivity.

Growth assay

Control, DDIT4 or MAF1 knockout cells were placed in the wells of a six-well dish at a concentration of 2 × 10⁵ cells/ml. Cells were either left untreated or exposed to a given concentration of oligomycin. A 250-μl aliquot of cells was counted on a Vi-cell XR at 24 and 48 h. Cell counts and cell viability were recorded.

Cas12a knockout cell lines

Two Alt-R Cas12a guide RNAs (IDT) were designed for each targeted gene. We pooled both guide RNAs and generated Cas12a-containing RNPs. K562 cells stably expressing KRAB-dCas9 were transfected with the Cas12a RNPs using TransIT-LT1 (Thermo Fisher Scientific, MIR2304) following manufacturer’s guidelines and were allowed to propagate for 3 days. Genomic DNA was isolated from an aliquot of cells and PCR was performed using primers that flank the putative deletion site. Following successful amplification of the deletion PCR product, clonal populations of putative knockout cells were generated by limiting dilution. Knockout cell lines were verified by western blot.

Gene ontology analysis

Gene lists were used as input to ShinyGO [20]. Lists of enriched pathways were shortened to remove redundancy and to maximize gene coverage. Graphs were generated using Python.

Screening by protein purification

CRISPuRe-seq comprises two main components: a phenotypic reporter consisting of a protein-of-interest (POI) fused to the RNA-binding MS2 coat protein (MCP) and a perturbation library expressing sgRNAs paired with BarRNAs (Fig. 1A, left panel). Each barcode is co-expressed in cis with a CRISPR sgRNA. The CRISPuRe-seq phenotypic reporter forms a barcoded ribonucleoprotein complex (BarRNP) with the genotype-identifying BarRNA through the specific and stable MCP/MS2 hairpin interaction (Supplementary Fig. S1A–C) [21]. Thus the POI carries information about the genotype of the cell that expressed it. This linkage can then be used to interrogate genetic perturbations that modify various properties of the POI.

A clonal cell line expressing the reporter and CRISPR machinery is infected with an sgRNA/BarRNA library to create a pool of perturbed, barcoded cells (Fig. 1A, middle panel). Perturbations that alter the abundance of the POI in the cell will correspondingly alter the abundance of the BarRNP. Relative POI levels are inferred by comparing the distribution of specific BarRNAs recovered after IP of the BarRNP to the distribution of BarRNAs in a total RNA fraction from the same cell lysate as measured by high-throughput sequencing (Fig. 1A, right panel). This system is highly amenable to both serialization and parallelization, thus allowing for dense time sampling and comparison of multiple environmental perturbations.

Here, we initially focus on screens of protein abundance. However, we note that the BarRNP can be isolated by most standard purification strategies, including IP, co-immunoprecipitation, sedimentation by density, and subcellular fractionation. We anticipate that future screens will leverage this flexibility to identify novel regulators of post-translational modifications, protein–protein interactions, and many other biochemically accessible protein attributes.

Development of an sgRNA/BarRNA co-expression vector

CRISPuRe requires co-expression of a unique BarRNA with a corresponding sgRNA. BarRNA expression cassettes were cloned into existing lentiviral sgRNA expression plasmids using a convergent transcriptional orientation where the sgRNA is driven by the U6 promoter and the BarRNA is expressed from either the 7SK or H1 Pol III promoters to minimize library recombination (Supplementary Fig. S1D). BarRNA levels are measured by Illumina sequencing after RNA isolation and cDNA synthesis with the incorporation of UMIs during reverse transcription. De-duplicated barcode counts are a proxy for the actual number of BarRNAs present in the isolated fraction.

We employed a “shotgun” cloning approach where synthesized barcodes were incorporated at random with respect to protospacers (Supplementary Fig. S1E). Using a 25-nucleotide barcode ensured that the majority of barcode/protospacer pairings would be unique and that each protospacer would be associated with multiple unique barcodes. Barcodes were assigned to corresponding sgRNA by Illumina sequencing the cloned plasmid pool.

The convergent transcriptional orientation of sgRNA and BarRNA expression cassettes limited lentiviral packaging-mediated recombination events between the protospacer and barcode to ∼11% (Supplementary Fig. S1F) and was compatible with both CRISPRi and CRISPR nuclease (Supplementary Fig. S1G).

A proof-of-concept transcriptional reporter screen

To benchmark CRISPuRe-seq screening for regulators of reporter protein abundance, we turned to the well-characterized IFN-α signaling pathway. IFN signaling offers a highly tunable transcriptional response achieved through changes in the dose of IFN-α and the number of ISREs in the promoter of the reporter.

We generated clonal K562 (chronic myelogenous leukemia) CRISPRi reporter cell lines that express an mClover3 [22] tandem-dimer MCP-3X Flag CRISPuRe reporter ORF under transcriptional control of an array of either 5X or 7X ISREs (Supplementary Fig. S1H). After validating IFN-inducible reporter expression for both constructs using flow cytometry (Supplementary Fig. S1I), we performed an analogous test using pools of nontargeting control guides and CRISPuRe-seq. While we could easily distinguish untreated cells from treated cells using the 7X ISRE reporter (Supplementary Fig. S1J), the dynamic range between low and high doses of IFN-α was limited. We hypothesized that the copy number of BarRNA limits the maximum amount of BarRNP and upper end of the dynamic range of CRISPuRe-seq. For this reason, we continued with the 5X ISRE reporter which had lower expression across all IFN doses tested.

We performed six parallel replicate CRISPuRe screens using 37 U/ml IFN-α and a focused library targeting genes involved in inflammation and innate immunity. We processed the screening data using either (i) a barcode-level analysis that compares the distribution of barcode log₂ ratios of BarRNAs associated with each sgRNA to the null population using the Wilcoxon rank sum test (Supplementary Fig. S2A) or (ii) a protospacer-level analysis that aggregates all BarRNAs for a given sgRNA and uses MAGeCK for further processing (Supplementary Fig. S2B).

To determine optimal read depth we sequenced our libraries up to 800× the number of library elements and then simulated lower read depths to establish baseline levels of BarRNA variance across multiple read depths (Supplementary Fig. S2C–E). Barcode-level analysis reproducibly identified the known positive-regulators of type 1 IFN signaling (Fig. 1B and C), but required high read depths to achieve accurate estimates of barcode abundance (Supplementary Fig. S2F and G). Protospacer-level analysis generated comparable log₂ ratios (Fig. 1D) but was more stable across all read depths (Fig. 1E), reducing the amount of sequencing required 10-fold and making the approach more amenable to larger scale screens.

Using 37 U/ml of IFN-α, CRISPuRe screening identified the same positive and negative regulators of IFN-α signaling as compared with a FACS screen performed in parallel (Fig. 1F and Supplementary Data S1–S4). Treatment with 1000 U/ml of IFN-α generated stronger gene scores for the positive regulators, but a much weaker gene score for the negative regulator USP18 (Supplementary Fig. S2H, Supplementary Data S5, and S6), consistent with the limited upper end dynamic range seen with higher reporter protein expression (Supplementary Fig. S2I). Thus, CRISPuRe-seq could identify regulators of reporter abundance at the sublibrary scale with comparable sensitivity to a FACS-based approach.

A novel ATF4 reporter screen accurately reflects induction of the ISR

The ISR coordinates protein synthesis rates with a multifaceted gene expression program to allow cells to adapt to a variety of conditions, including genetic abnormalities and exogenous stressors. CRISPuRe-seq’s ability to both parallelize and serialize made it ideally suited to (i) sample ISR regulators as a function of knockdown duration and (ii) identify modifiers of the ISR in response to exogenous stressors. We therefore applied CRISPuRe-seq to perform whole genome screens to identify regulators of the ISR.

To create a reporter of the ISR, the CRISPuRe reporter was placed under the translational control of the uORFs in the ATF4 5′ untranslated region (UTR) (Fig. 2A). We introduced this reporter into the K562 CRISPRi cell line and selected a clone with modest basal expression that increased in response to stress (Supplementary Fig. S3A). Perturbations that induce the ISR will upregulate reporter translation and fraction of BarRNAs in the BarRNP. Conversely, inhibitors of the ISR should decrease BarRNP abundance relative to the total expressed BarRNA (Fig. 2B).

To validate that the IP strategy can detect barcode enrichment upon stress induction, we performed the following cell mixing experiment (Supplementary Fig. S3B). Three populations of K562 cells stably expressing non-targeting sgRNA/BarRNA pools 1, 2, and 3 were generated by lentiviral transduction. We then treated these pools separately as follows: untreated (pool 1), 20 nM oligomycin (a mitochondrial F_OF₁-ATPase inhibitor, oligomycin pool 2), or 300 nM thapsigargin (a sarco/endoplasmic reticulum Ca²⁺ ATPase inhibitor, thapsigargin pool 3). After 3 h of treatment, we combined these pools in an 8:1:1 ratio, lysed the cells, and performed CRISPuRE-seq by Flag IP, comparing the log₂ ratio in barcode abundance between the IP fraction and the total RNA fraction for each pool (Supplementary Fig. S3C). Both CRISPuRe and flow cytometry demonstrated reporter induction in response to stress, with thapsigargin eliciting a stronger response than oligomycin.

Whole-genome CRISPuRe screens identify known and novel regulators of the ISR

Having validated the reporter system, we next conducted a series of genome-scale screens in K562 cells using a library targeting 18 946 genes (Supplementary Data S7). These screens aimed to identify genetic perturbations that elevate ATF4 levels under basal conditions in the absence of a known stressor. To determine the optimal timepoint for examining the ISR, we performed CRISPRuRe screens over five consecutive days, starting on Day 5 post-transduction. We hypothesized that certain gene classes, particularly those crucial for cell fitness, could exhibit phenotypic changes at early timepoints but not later due to cell dropout and survivorship bias against efficient knockdown. Conversely, there might exist genes whose knockdown effects take longer to manifest due to half-life considerations or ISR induction due to compensatory effects that would become evident at later timepoints. We binned BarRNA abundance at the protospacer-level and compared the total RNA fraction to the IP fraction on each day of screening, processing two separate samples of 50 million cells from a total population of 500 million cells per timepoint. Candidate genes with elevated reporter expression had protospacer distributions that were significantly and reproducibly shifted relative to either the negative control population or the total protospacer population (Supplementary Fig. S3D).

To perform a gene-level analysis, we processed both samples from each day as paired replicates using MAGeCK [18]. For each gene, we calculated a gene score, defined as the product of the negative log₁₀P-value and the effect size (log₂ IP:total ratio). We then performed K-means time-series cluster analysis [23] to identify the highest scoring genes that behaved similarly across the 5 days of screening (Fig. 2C, Supplementary Fig. S3E, and Supplementary Data S8–S12).

We identified a cluster of genes that generated high reporter levels on Days 5 and 6 when knocked down, but with nearly undetectable effects by Day 9. Gene Ontology (GO) analysis revealed that this group was highly enriched for known regulators of the ISR, including members of the eIF2 and eIF2B complexes [24] and multiple tRNA aminoacyl synthetases [25]. Perturbation of several regulators of mitochondrial function, including ATP synthesis and mitochondrial protein import also increased reporter abundance (Fig. 2D and Supplementary Data S13). Genes associated with the U2/U12 spliceosome made up the largest class of hits that were strong early but did not score at later timepoints.

As further validation of our approach, we performed a replicate screen, processing two Day 5 samples drawn from independent cell populations. These samples were similarly processed in MAGeCK and compared with the original Day 5 screen results. Gene scores between the two replicates were correlated with a Pearson correlation coefficient of 0.67 (Supplementary Fig. S3F and Supplementary Data S14) which is within the range of what would be expected from a screen yielding this number of hits [26].

To benchmark our screening approach against a direct readout of endogenous ISR gene expression, we compared CRISPuRe results with recently published "genome-scale" CRISPRi Perturb-seq atlases [27] that targeted up to 9867 expressed genes. Perturb-seq measures the transcriptomes of genetically perturbed cells. Each perturbation was assigned an ISR score by taking the sum of the normalized expression value from a set of genes known to be transcriptionally induced by the ISR [27] (Supplementary Data S15). As CRISPuRe screening generated the strongest results on Days 5–7, we aggregated the top hits and compared their gene score to the Perturb-seq ISR score. For the genome-scale Perturb-seq screens in K562, we found a strong overlap between genes with the highest gene score and highest ISR score. These hits included the eIF2 and eIF2B complex members, the tRNA synthetases, and many of the mitochondrial genes (Fig. 2E). Interestingly, splicing factors that were strong hits by CRISPuRe screening were undetected in K562 by Perturb-seq. Because the K562 Perturb-seq screen was performed on Day 8 post-transduction and our data demonstrated a day-dependent effect, we compared the CRISPuRe screening data to a second K562 Perturb-seq screen performed on Day 6 (“essential genes”) and found similar overlap but were still unable to identify splicing factors (data not shown). In contrast, Perturb-seq performed in epithelial RPE1 cells revealed that splicing factors were significantly enriched for genes with elevated ISR scores [27] (Fig. 2F).

Given that up to 98% of human transcripts undergo splicing [28], perturbing this process provides many potential routes to an ISR. Whether this is the result of intron retention leading to aberrant protein production or the loss of specific proteins is not currently known. However, it was recently shown that a mutation in the splicing factor SF3B1, commonly found in myelodysplastic syndrome, was sufficient to trigger an ISR in K562 by disrupting heme synthesis and activating EIF2AK1/HRI [29]. Furthermore, treatment with isoginkgetin, a small molecule inhibitor of pre-messenger RNA (pre-mRNA) splicing, has been shown to increase ATF4 translation and activate an ISR gene expression signature in HeLa and HCT116 cells [30].

It is possible that the translational CRISPuRe reporter can more readily detect perturbations that affect splicing, as it requires no splicing itself and is a more direct readout of ATF4 translation. Perturb-seq requires accurate splicing of ATF4 pre-mRNA and the proper functioning of Pol II machinery for a robust transcriptional response. Furthermore, as each element of the Perturb-seq library contains two targeting guides selected for maximal knockdown, individual cells may experience a stronger splicing defect, possibly changing the dynamics of the transcriptional response.

We also found that Perturb-seq identified several mitochondrial genes that generated little-to-no gene score via CRISPuRe-seq. To determine if this discrepancy could also result from differences in knockdown efficiency between the two libraries, we analyzed protospacer-level log₂ ratios for two Day 5 replicates and grouped the protospacers based on their presence in the Perturb-seq library (Supplementary Fig. S3G). Protospacers that were present in the Perturb-seq library showed an elevated log₂ ratio compared with protospacers that were present only in the CRISPuRe library. This is consistent with the idea that the effects of the strongest protospacers are detectable by CRISPuRe-seq but the remaining weaker protospacers blunted these effects at the gene level.

Oligomycin and thapsigargin inhibit RNA Pol III

Having identified known ISR regulators from the time-course screens, we next sought to identify genes required for ISR induction in response to the exogenous stressors oligomycin and thapsigargin. We performed parallel whole genome CRISPuRe screens using CRISPRi in Day 6 cells treated with oligomycin or thapsigargin for 4 h or left untreated. Top hit genes required for ISR induction in response to oligomycin included KIAA0141/DELE1, EIF2AK1/HRI, and OMA1, all of which were recently shown to be required to activate the ISR in response to mitochondrial stress (Fig. 3A and Supplementary Data S16) [31, 32]. Similarly, the strongest hit gene in the thapsigargin screen was EIF2AK3/PERK, the ER transmembrane kinase responsible for activating the ISR in response to ER stress [33] (Supplementary Fig. S4A and Supplementary Data S19).

GO analysis of the top 250 hit genes that reduced reporter abundance (Supplementary Data S17 and S18) in the oligomycin screen showed strong enrichment for HRI-mediated signaling, negative regulators of ATP synthesis, regulators of translation initiation, maturation of the ribosomal small subunit, and, surprisingly, regulators of RNA Pol III activity (Fig. 3B). The Pol III regulators include three components of the TFIIIC complex [34], which is required for expression of tRNAs, and the Pol III negative regulator MAF1 [35, 36].

The identification of Pol III regulators was unexpected since induction of the ISR is not known to require Pol III. By contrast, our analysis of the Perturb-seq data from RPE1 cells suggested that perturbation of Pol IIl components, including the TFIIIC complex, promoted an ISR rather than attenuated the response (Supplementary Fig. S3E). Because the CRISPuRe system itself requires the Pol III-mediated expression of both the sgRNA and the BarRNA, we investigated whether the ISR might regulate Pol III and, consequently, reduce BarRNA expression.

To determine if ISR stressors were affecting BarRNA expression, we compared log₂ ratio of BarRNA counts in oligomycin-treated versus untreated cells separately in the total RNA and the IP pools (Fig. 3C). Given that induction of the ISR increased reporter expression during stress, we anticipated increased BarRNA abundance in the IP fraction of cells competent to initiate an ISR, but not in cells with the ISR blocked. Unexpectedly, we found no relative increase in BarRNA abundance in the IP fraction from cells that have an active ISR in response to oligomycin (EIF2AK2, EIF2AK3, and EIF2AK4) (Fig. 3C). By contrast, we found increased abundance of BarRNA in the total RNA fraction for ISR-blocking perturbations (DELE1, OMA1, and EIF2AK1). This suggested that the ISR reduced total BarRNA levels, potentially blunting the increase in BarRNP levels despite increased reporter translation.

To better isolate the effects on BarRNA expression, we re-analyzed the screens using total BarRNA levels as the phenotype (Fig. 3D, Supplementary Fig. S4B, and Supplementary Data S20, S23, and S24). This analysis identified the screen hits with the most elevated BarRNA expression under stressed conditions as validated ISR regulators and members of the TFIIIC complex. Furthermore, GO analysis on the top 250 screen hits for high BarRNA expression during mitochondrial stress (Supplementary Data S21 and S22) showed significant enrichment for mitochondrial RNA processing, mitochondrial translation, mitochondrial gene expression, and mitochondrial respiratory chain complex assembly (Fig. 3E), suggesting that BarRNA expression accurately reflects ISR status.

Does the ISR repress endogenous Pol III gene expression? We employed a qPCR approach to quantify specific pre-tRNAs as proxies for nascent Pol III activity. We measured two tRNAs with typical internal type 2 tRNA promoters [37] (pre-Ile and pre-Tyr) and the selenocysteine (SelCys) tRNA [38], which has a promoter with external control elements similar to the type 3 promoter used to express the BarRNA. DDIT4, a known ATF4 target gene, was used to monitor ISR induction [39]. Cells were treated with either oligomycin, thapsigargin, or ML-60218 as a positive control for Pol III inhibition [40].

We found that treating cells with either oligomycin or thapsigargin for 4 h induced the ISR as monitored by DDIT4 (Fig. 3F–1). ML-60218 treatment dramatically increased DDIT4 expression. Treatment with oligomycin or thapsigargin caused a significant reduction in pre-lle and pre-Tyr levels (60%–70%) (Fig. 3F2 and 3). Similarly, BarRNA and pre-SelCys levels were reduced by ∼40% in response to oligomycin and thapsigargin, and 70% in response to ML-60218 (Fig. 3F4 and 5). These findings demonstrate that treatments that activate the ISR also suppress Pol III transcription from type 2, type 3, and hybrid Pol III promoters.

Knockdown of GTF3C1 prevents Pol III inhibition specifically at type 3 promoters

We sought to determine the relevance of screen hits on the regulation of Pol III transcription at endogenous loci. The top hits from either the oligomycin-based or thapsigargin-based BarRNA screens were components of the TFIIIC complex and the known regulators of the ISR. To determine if the TFIIIC complex prevented ISR activation in response to oligomycin or thapsigargin, or otherwise regulated RNA Pol III inhibition of endogenous genes, we generated a stable GTF3C1 CRISPRi knockdown cell line and performed the same qPCR experiments.

GTF3C1 knockdown cells exhibited an elevated baseline level of DDIT4, consistent with Perturb-seq data in RPE1 cells (Supplementary Fig. S4C1). Treatment with oligomycin or thapsigargin further increased DDIT4 expression, indicating that the cells maintain a functional ISR. GTF3C1 knockdown dramatically reduced baseline type 2 tRNA expression, in line with its role in promoting Pol III activity at type 2 promoters (Supplementary Fig. S4C2 and 3 and D). However, expression of the type 3-like selenocysteine tRNA was increased nearly three-fold (Supplementary Fig. S4C5). Inhibition of type 2 tRNA transcription by drug treatment was comparable to control cells (Supplementary Fig. S4C2 and 3 and E1 and 2), while inhibition of the BarRNA and selenocysteine tRNA transcription was greatly reduced, consistent with screening results (Supplementary Fig. S4C4 and 5 and E3 and 4). Together, these findings suggest that GTF3C1 knockdown cells maintain a weak chronic ISR that can be augmented by exogenous stress.

Stress induced Pol III inhibition is dependent on the canonical ISR pathway

We next interrogated whether Pol III repression by exogenous stress was mediated by eIF2α phosphorylation and a canonical ISR. As expected, knockdown of EIF2AK1/HRI prevented induction of the ISR specifically in response to oligomycin as measured by DDIT4 expression (Fig. 4A1). When treated with oligomycin, pre-tRNA expression of Ile and Tyr was partially restored (Fig. 4A2 and 3). Additionally, BarRNA and SelCys tRNA expression were almost fully restored (Fig. 4A4 and 5). Mature Met tRNA levels were higher in HRI knockdown cells compared with control cells under all conditions (Supplementary Fig. S5A1). However, baseline expression of pre-tRNAs appeared unaffected (Supplementary Fig. S5A2–5).

Similarly, EIF2AK3/PERK knockdown eliminated ISR induction specifically in response to thapsigargin (Fig. 4B1). PERK knockdown significantly restored expression of all the tested Pol III transcripts in the presence of thapsigargin but seemed to sensitize cells to oligomycin (Fig. 4B2–5). In contrast to HRI knockdown, PERK knockdown led to elevated baseline expression of both type 2 pre-tRNAs in untreated cells (Supplementary Fig. S5B2 and 3) but had little effect on baseline levels of Met tRNA (Supplementary Fig. S5B1).

To determine if Pol III repression was mediated by ATF4, these experiments were repeated in an ATF4 knockout cell line. In the absence of ATF4, no induction of DDIT4 was detected in response to any treatment (Fig. 4C1). Partial rescue was observed for all Pol III transcripts when treated with oligomycin and near-complete rescue for cells treated with thapsigargin (Fig. 4C2–5). The effect of the Pol III inhibitor itself was significantly rescued in the ATF4 knockout background in all cases, suggesting that a portion of the Pol III inhibitory activity of ML-60218 is mediated by the ISR. While mature Met tRNA levels in ATF4 knockout cells were comparable to negative controls, baseline levels of pre-tRNAs and BarRNA were significantly reduced (Supplementary Fig. S5C1–5), suggesting that basal ATF4 functions to maintain normal Pol III activity in K562 cells. Taken together, these data demonstrate that activation of the ISR in response to mitochondrial or ER stress triggers an ATF4-dependent inhibition of RNA Pol III activity, likely via the expression of one or more ATF4 target genes.

Activation of the ISR inhibits mTORC1 and decreases phosphorylation of the Pol III repressor MAF1

Having identified the RNA Pol III inhibitor MAF1 in the oligomycin screens and given that mTORC1 has been demonstrated to regulate RNA Pol III activity through the phosphorylation of MAF1 [41], we hypothesized that the ISR represses mTORC1 to inhibit RNA Pol III activity. To examine this, we treated cells with either oligomycin or thapsigargin for 4 h and measured the phosphorylation of p70 S6K, as a readout of mTORC1 activity. We found that oligomycin decreased p70 S6K phosphorylation to 10% of the levels observed in untreated cells, while thapsigargin decreased it to 40% (Fig. 4D and Supplementary Fig. S5D).

Knocking down HRI partially rescued the decreased phosphorylation in response to oligomycin but not thapsigargin (Fig. 4E). Conversely, knocking down PERK almost completely rescued the decreased phosphorylation in response to thapsigargin, but not oligomycin (Fig. 4F). Furthermore, in ATF4 knockout cells treated with oligomycin and thapsigargin, p70 S6K phosphorylation was restored to ∼80% and 100% of the levels seen in untreated cells, respectively (Fig. 4G).

During exponential growth, mTORC1 is active, leading to MAF1 phosphorylation and subsequent inhibition. However, under nutrient stress, mTORC1 activity decreases, allowing unphosphorylated MAF1 to inhibit Pol III activity [41]. We investigated MAF1 phosphorylation in response to oligomycin and thapsigargin treatment in control and ISR-defective cells.

MAF1 protein resolves as three bands on Phos-tag gels: an upper band that represents hyperphosphorylation at multiple sites, a middle band that represents phosphorylation at one site, and a lower band that represents the unphosphorylated active form [41]. In untreated cells, MAF1 is predominantly in the upper band. Upon treatment with oligomycin or thapsigargin, there is a marked redistribution from the upper band to the unphosphorylated active form of MAF1. Likewise, treatment with Torin 1, an mTORC1 kinase inhibitor [42], shifts the majority of MAF1 protein into the active lower band (Fig. 4D and Supplementary Fig. S5E).

In HRI knockdown cells, the oligomycin-induced shift of MAF1 protein from the upper to the lower band is specifically blocked (Fig. 4E and Supplementary Fig. S5F). Conversely, PERK knockdown prevents the thapsigargin-induced shift but not the oligomycin-induced shift (Fig. 4F and Supplementary Fig. S5G). Knockout of ATF4 blocks the shift of MAF1 in response to both oligomycin and thapsigargin (Fig. 4G and Supplementary Fig. S5H). These findings indicate that the ISR inhibits mTORC1 activity, leading to a decrease in the inhibitory phosphorylation of MAF1, a key RNA Pol III inhibitor (Fig. 4H).

DDIT4 and SESN2 act downstream of ATF4 to inhibit Pol III activity and to modulate MAF1 phosphorylation

The mTORC1 inhibitors DDIT4 and SESN2 are regulated by ATF4 [43–47]. Recent studies also suggest that DDIT4 and SESN2 contribute to mTORC1 inhibition following oligomycin treatment in HEK293T cells [48]. To determine if the ISR repressed Pol IIl activity through DDIT4 or SESN2, we generated a DDIT4 knockout line (Supplementary Fig. S6A), a SESN2 knockdown line (Supplementary Fig. S6B), and knocked down SESN2 in the DDIT4 knockout background. We then assayed Pol III activity by qPCR and assessed mTORC1 activity and MAF1 phosphorylation in untreated cells and cells treated with oligomycin or thapsigargin.

To assess ISR activation in the DDIT4 knockout background, we measured expression of the known ATF4 target gene ASNS [49]. ASNS expression was increased upon treatment with oligomycin or thapsigargin (Fig. 5A1). Knockout of DDIT4 significantly restored Pol III activity in response to oligomycin and thapsigargin for type 2 pre-tRNAs (Fig. 5A2 and 3) but not for the BarRNA or for the SelCys pre-tRNA (Fig. 5A4 and 5). Like the ATF4 knockout, DDIT4 knockout also partially rescued the inhibitory effects of ML-60218 across all genes. Loss of DDIT4 had no effect on the levels of mature Met tRNA or on the baseline levels of any of the pre-tRNA (Supplementary Fig. S6C1–3 and 5), but baseline BarRNA levels were reduced by almost 60% (Supplementary Fig. S6C4).

SESN2 knockdown cells showed a normal baseline level of ASNS that was increased similarly to control cells when treated with oligomycin or thapsigargin (Fig. 5B1). DDIT4 was elevated three-fold in SESN2 knockdown cells and treatment with oligomycin or thapsigargin was able to further increase DDIT4 expression beyond what was observed in control cells (Supplementary Fig. S6F). Knockdown of SESN2 on its own did not rescue Pol III activity in response to any of the treatments (Fig. 5B) and did not dramatically alter baseline Pol III activity (Supplementary Fig. S6D). SESN2 knockdown in the DDIT4 knockout background showed nearly identical rescue as DDIT4 knockout on its own (Fig. 5C), with a slightly enhanced rescue for the BarRNA and the SelCys pre-tRNA in response to thapsigargin (Fig. 5C4 and 5). While the overall level of baseline expression for most of the transcripts was similar between the DDIT4 knockout and the DDIT4; SESN2 double, the double did have significantly reduced baseline expression of the Ile pre-tRNA and ASNS (Fig. 5C1 and Supplementary Fig. S6E2).

Consistent with the qPCR results, we observed strong rescue of mTORC1 activity in the DDIT4 knockout background, with high levels of S6K and MAF1 phosphorylation even in the presence of oligomycin or thapsigargin (Fig. 5E and Supplementary Fig. S6G and I). Interestingly, while basal S6K phosphorylation in SESN2 knockdown cells was difficult to detect, there appeared to be no further reduction in response to thapsigargin. Furthermore, SESN2 knockdown weakly rescued MAF1 phosphorylation (Fig. 5F and Supplementary Fig. S6G and J). The DDIT4; SESN2 double background showed strong rescue with respect to S6K phosphorylation in response to thapsigargin and partial rescue with respect to oligomycin. MAF1 phosphorylation however, was better maintained in response to stress in the DDIT4; SESN2 double background than in the DDIT4 knockout alone (Fig. 5G and Supplementary Fig. S6G and K).

These data demonstrate that ATF4-dependent upregulation of DDIT4 and SESN2 contributes to mTORC1 inhibition and decreased phosphorylation of MAF1 in response to mitochondrial and ER stress (Fig. 5H). While DDIT4 and SESN2 activities account for the majority of mTORC1 and RNA Pol III inhibition in thapsigargin-induced ER stress, an additional inhibitory mechanism likely functions independently of the ISR in the context of oligomycin treatment.

MAF1 is required for ISR-dependent Pol III inhibition

To determine if MAF1 was necessary for ISR-dependent Pol III inhibition, we repeated our experiments in a MAF1 knockout cell line. We observed that MAF1 was completely required for this inhibition (Fig. 6A). Interestingly, in the absence of MAF1 we found that treatment with oligomycin, thapsigargin, or the Pol III inhibitor ML-60218 itself, instead of decreasing, slightly increased Pol III activity over the already elevated baseline Pol III activity (Supplementary Fig. S7A). Taken together, these findings indicate that ISR activation triggers MAF1 activity through the DDIT4/SESN2-mediated inhibition of mTORC1, thereby limiting RNA Pol III output under stress conditions.

Pol III inhibition is required for sustained growth in the presence of oligomycin

Given the energy cost of maintaining Pol III transcription during periods of growth arrest, we investigated the impact of unrestrained Pol III activity on cell growth under normal and stressed conditions. Under optimal conditions, cells lacking DDIT4 or MAF1 trended nonsignificantly towards a slower doubling time compared with control cells. However, when cells were challenged with oligomycin, control cells maintained growth at a reduced rate, while cells lacking DDIT4 or MAF1 were significantly compromised (Fig. 6B). This result suggests that Pol III inhibition is required for sustained growth when energy reserves are depleted.