Metabolic and transcriptional plasticity supports CD8⁺ T cell resilience and anti-tumor immunity under nutrient stress

Summary

CD8+ T cells need to function in complex environments with varied nutrient availability, including the tumor microenvironment and inflamed tissues. The mechanisms allowing CD8+ T cells to maintain immune function in these perturbed settings are poorly understood. Here, we show that CD8⁺ T cells adapt to nutrient stresses over time, reconfiguring gene-regulatory and metabolic networks to license functional recovery. Under acute stress, T cells reoriented translational programming, which limited nutrient demand and prioritized stress-sensitive metabolic and transcriptional responses. Within these responses, the transcription factors ATF4 and CEBPG jointly established an adaptive metabolic program, promoting amino acid synthesis and uptake while maintaining mitochondrial metabolism. Despite diminished energetic capacity under environmental stress, this program sustained central carbon metabolism. This subsequently mitigated cellular dysfunction and potentiated anti-tumor immunity. Altogether, we demonstrate that biosynthetic plasticity via translational and metabolic reprioritization confers T cell resilience in unfavorable environments, offering potential strategies to enhance immunotherapies.

eTOC

CD8+ T cells must retain functional resilience as they traverse diverse environments. Scaglione et al., reveal how the integrated stress response enables CD8+ T cells to engage alternate biosynthetic modes under environmental stress that prevent dysfunction and promote anti-tumor immunity.

Graphical Abstract

Introduction

Cells within multicellular organisms require nutrients to support growth and viability, as well as lineage-specific functions that promote organismal health. To do this, cells acquire environmental nutrients and coordinate their utilization across competing biochemical pathways^1,2. As both cell homeostasis and lineage-specific functionality depend on a common store of nutrients, cells must coordinate nutrient consumption and allocation, integrating information from growth signaling, functional programing, and nutrient availability.

To ensure that metabolic demand and nutrient supply are balanced, cells must couple information on their intracellular and environmental metabolic state to the regulation of global biosynthesis^3–5. For example, nutrient-sensitive signaling pathways like the mechanistic target of rapamycin (mTOR) and the integrated stress response (ISR) regulate protein translation, a primary consumer of energy and amino acids^6–10. These pathways also influence metabolic activity, nutrient acquisition, and regulatory responses, linking cell biology to environmental state. Although these pathways control stereotyped cellular processes, how they intersect with lineage-specific programming in response to changing or limiting environmental conditions is underexplored^11–15.

CD8⁺ T cells travel to perturbed peripheral sites to perform protective effector functions such as cytokine production or cytotoxicity. These sites are biochemically distinct from blood, containing varying levels of nutrients as well as inhibitory or immunosuppressive metabolites^15–18. Indeed, nutrient deprivation and metabolic dysregulation drive cellular dysfunction and limit cellular persistence^19–29. Despite this, there is limited insight into factors that mitigate stress-driven dysfunction and support functional resilience in changing or limiting environments. Additionally, though there is a growing appreciation that metabolic perturbation and sensing regulate gene expression and cell function, the global regulatory impact, temporal dynamics, and specificity of nutrient-stress-induced changes in T cell programming are poorly understood^30–39.

Here, we investigated the global regulatory underpinnings of nutrient stress adaptation in CD8⁺ T cells. Despite an initial loss of function during nutrient stress, CD8⁺ T cells exhibited metabolic adaptation and functional recovery over time. Using multi-omic analyses of a diverse set of nutrient stresses, we comprehensively resolved stress-sensitive programming in CD8+ T cells, and identified nutrient-sensitive molecular targets within the transcriptome, translatome, and proteome. We found that early during periods of nutrient stress, loss of mTORC1 signaling translationally repressed mitochondrial and ribosomal mRNAs, reducing metabolic demand and preserving amino acid pools. Concurrently, activation of ISR signaling promoted metabolic adaptation and functional recovery through a stress-sensitive transcriptional network. We highlight two stress-induced transcription factors, ATF4 and CEBPG, as key mediators of this adaptive response. Despite diminished energetic capacity under environmental stress, these factors cooperatively promoted amino acid synthesis and uptake and maintained mitochondrial anaplerosis. This program sustained T-cell function required for anti-tumor immunity, revealing a mechanism mitigating stress-driven cellular dysfunction. These data illustrate how CD8⁺ T cells concordantly reprogram global gene regulation and metabolism to maintain immune function during stress, providing insight into environmental control of immune function for immunotherapeutic applications as well as a paradigm for broader inquiry into resilience of lineage-specific cell functions across biochemical environments.

Results

Nutrient stress drives metabolic and gene-regulatory adaptation in CD8⁺ T cells to support functional recovery

To determine whether CD8⁺ T cells adapt to nutrient stress over time, we assayed the metabolic content and function of activated CD8⁺ T cells during an acute (6 hour) or long-term (24 hour) exposure to MC38 colorectal cancer cell line conditioned medium (“tumor supernatant”) (Fig 1A). Conditioned medium showed multiple alterations in amino acid composition relative to non-conditioned medium (Fig 1B). After 6 hours, T cells exhibited intracellular amino acid profiles mirroring the medium and showed decreased cytokine production (Fig 1 C and D, Fig S1A). However, by 24 hours, the intracellular levels of many previously-depleted amino acids recovered, and CD8+ T cells showed rescued or even increased function relative to controls with no loss in viability (Fig 1 C and D, Fig S1A). Moreover, we observed an increase in the surface expression of CD98, a common heavy chain subunit of System L and System Xc- amino acid transporters when dimerized with light chains including SLC7A5 (LAT-1) and SLC7A11 (xCT) (Fig 1E)⁴¹. Consistent with this, we observed an increase in Slc3a2, Slc7a5, and Slc7a11 mRNAs within 6 hours of tumor supernatant culture (Fig S1B).

Figure 1: CD8⁺ T cells exhibit genetic and metabolic adaptation during nutrient stress.

A) Schematic of nutrient stress models. Activated CD8+ T cells were treated with tumor supernatant or nutrient-depleted media for a total of 6 or 24 hours and re-stimulated during the final 6 hours of culture.

B) Relative difference in medium amino acid abundance in tumor-conditioned vs. control medium.

C) Relative difference in cellular amino acid abundance in tumor-conditioned vs. control treated CD8⁺ T cells, after 6 or 24 hours of culture. Those increasing over time are shown in color.

D-E) Intracellular cytokine production (D) and CD98 expression (E) in CD8⁺ T cells treated with tumor-conditioned medium for 6 (top) or 24 hours (bottom)

F-G) Intracellular cytokine production (F) and surface CD98 expression (G) in nutrient stress media cultured CD8+ T cells after 6 (top) or 24 hours (bottom)

H) Amino acid transporter transcripts expression in nutrient stress media cultured CD8⁺ T cells

Data points represent technical replicates. Error bars represent SEM. Individual experiments performed at least twice. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by Welch’s t test (D and E) and Brown-Forsythe and Welch ANOVA tests (F and G).

To test the specificity of this stress-adaptation, we assayed CD8⁺ T cell function over time using an array of single nutrient-depleted media formulated without glucose, glutamine, branched-chain amino acids (BCAAs – leucine, isoleucine, valine), arginine, or methionine (Fig 1A). These metabolites have broad differences in essentiality, utilization by cellular catabolism and biosynthesis pathways, and sensing by signaling pathways⁴². We confirmed that amino acid depletion from media led to specific and rapid intracellular depletion (Fig S1C). These nutrient-stress media drove diverse acute and long-term outcomes: cells cultured without glutamine completely recovered function over time, cells cultured without BCAAs retained full viability but failed to recover function, cells cultured without glucose lost function and viability, and cells cultured without arginine or methionine displayed minimal functional changes over time (Fig 1F, Fig S1D). In contrast, CD98 protein and transcript expression was rapidly increased in all conditions (Fig 1G, Fig 1H). (Add summary statement/take-home message?)

Acute nutrient stress drives translational reprioritization of pro-adaptive programming over growth programming

Next, we aimed to better understand the acute response to nutrient stress at a global level. As gene expression can be modified through transcriptional, translational, and post-translational mechanisms, we tested which mechanisms were most sensitive to stress, which targets were most dynamic, and whether these responses were unique or conserved across conditions by profiling the transcriptome, translatome, and proteome after 3 hours in five nutrient stress conditions (Fig 2A)⁴³. Each nutrient stress affected global ribosomal activity; all except arginine exhibited a significantly elevated monosome abundance, with glucose, arginine, and methionine restriction yielding diminished polysome abundance, resulting in depressed polysome/monosome ratios for all conditions (Fig 2B and C). Analysis of the transcriptome and translatome composition revealed various protein coding and non-coding transcripts, including several long-non-coding RNAs (Fig S2A–C). While patterns of relative transcript abundance between the transcriptome and translatome were distinct, acute nutrient stress drove rapid remodeling of both pools, with the greatest number and magnitude of differentially-expressed transcripts in the polysome-bound RNA fraction (Fig 2D, Fig S2D–F). Conversely, changes in the proteome were limited (Fig 2D).

Figure 2: Acute nutrient stress drives translational reprioritization of pro-adaptive programming over growth programming.

A) Schematic of comparative analysis of stress-sensitive programming in CD8⁺ T cells. Cells were treated with nutrient stress media for 3 hours before total RNA sequencing of input and polysome-bound RNA and whole-cell proteomics

B) Polysome profiles of nutrient stress cultured CD8⁺ T cells after 3 hours. Inset shows polysome region.

C) Quantification of monosome area, polysome area, and polysome / monosome ratio.

D) Plot of control-treatment sample-sample Euclidean distances in input RNA, polysome RNA, and protein datasets.

E) Number of differentially expressed genes within the top 100 DEGs (up or down) uniquely differential in conditions (lightest), or differential in the same direction across 2 (middle) or 3+ (dark) nutrient stress conditions

F) Global heatmap of differentially expressed transcripts in input RNA (left) and polysome-associated RNA (right) in nutrient stress conditions relative to control. Left annotation: k-means cluster membership and relative expression in control condition. Right annotation: selected pathways enriched in indicated clusters

G) Change in expression of growth/biosynthesis-related transcripts in input RNA (left) and polysome-associated RNA (right) in nutrient stress conditions

H) Change in expression of all 5’ TOP motif containing transcripts from Thoreen et al. (2012) in input (left) and polysome-associated RNA (right)

I) Change in expression of amino acid handling transcripts in input RNA (left) and polysome-associated RNA (right) in nutrient stress conditions.

J) Change in expression of all differential and selected stress-related transcription factors in input RNA (left) and polysome-associated RNA (right) in nutrient stress conditions.

K) Schematic for translational reprioritization during acute nutrient stress

Data points represent technical replicates. Error bars represent SEM. Individual experiments performed at least twice. *p < 0.05, **p < 0.01, ***p < 0.001 by Brown-Forsythe and Welch ANOVA tests (C).

We observed that stress-driven changes in both the transcriptome and translatome were largely conserved across conditions (Fig 2E and F, Fig S2G). However, we also found nutrient-specific modules in each response, with glucose depletion and methionine depletion the most unique (Fig 2E and F, Fig S2G). Lastly, we saw that the magnitude of each stress-response differed between conditions. Whereas arginine depletion produced the fewest differentially-expressed transcripts and weakest changes in transcript abundance, methionine depletion produced the greatest number and magnitude of differentially-expressed transcripts (Fig 2F, Fig S2G). Taken together, these data suggest that nutrient depletion drives rapid remodeling of the transcriptome and translatome. Furthermore, while some nutrient-specificity is observed, these acute RNA responses are largely conserved, highlighting a core stress-response program.

Next, we aimed to identify the major stress-sensitive regulatory networks altered upon acute nutrient stress. We evaluated how CD8⁺ T cells “prioritize” transcripts during environmental stress, quantifying the fraction polysome-associated RNAs related to given pathways across conditions. We noted that the relative polysome enrichment of mRNAs involved in T cell function were maintained after nutrient stress, while other abundant mRNAs like cytosolic ribosome transcripts were selectively deprioritized (Fig S2F). To identify pathways that exhibited translational prioritization or repression under stress, we clustered stress-driven changes in input and polysome-associated RNA enrichment and performed pathway analysis (Fig 2F). While some clusters exhibited similar stress-driven responses within input and polysome-bound fractions, other clusters displayed stronger polysome-bound RNA responses suggestive of translational regulation (Fig 2F, Fig S3A and B). Of note, Cluster 6 contained translationally repressed transcripts related to ribosomal and mitochondrial biology (Fig 2F and G). These changes were robust at the pathway level and common to all five nutrient stresses (Fig S2H). Many of these transcripts contain a 5’ TOP motif, an mTOR pathway-sensitive mRNA translational regulatory motif implicated in effector T cell responses (Fig 2H)^44–49. Additionally, transcripts related to cell growth, including those involved in cholesterol biosynthesis (Hmgcs1, Mvk, Dhcr24, Hsd17b7 and Scd5) and growth factor signaling (Il2ra), were depleted from both fractions across stresses (Fig 2F and G).

Multiple clusters also displayed elevated RNA abundance under stress, often with disproportionate enrichment within polysome-bound fractions. Here, many transcripts related to amino acid handling: tRNA charging (Lars1, Cars1, Aars1), amino acid transport (Slc7a5, Slc3a2, Slc6a9, Slc1a4, Slc7a11 etc.) and synthesis, particularly for alanine (Gpt2), proline (Pycr1 and Aldh18a1), serine/glycine (Phgdh, Psat1, Psph, Shmt2 etc.), and asparagine (Asns) (Fig 2F and I). Stressed T cells also displayed increased Havcr2 (Tim3) and Lag3 transcript, suggesting a link between nutrient stress and inhibitory receptor expression (Fig 2F). Additionally, transcripts containing internal ribosomal entry sites (IRES) were enriched in polysome-bound fractions, suggesting that stress-prioritized transcripts may contain unique regulatory features that sustain polysome association (Fig S2I).

Lastly, we observed approximately 90 transcription factors that were acutely sensitive to nutrient stress. Most of these factors were upregulated, including members of the ATF (Atf3, Atf4, Atf5, Atf6), CEBP (Cebpb, Cebpg), NRF (Nfe2l1, Nfe2l2), and AP-1 (Fos, Jun) families (Fig 2J). We also observed that putative binding sites for ATF4 and CEBPG, which share a similar motif, were enriched near the transcriptional start sites of genes upregulated across stresses, suggesting that these transcription factor families may play a key role in early nutrient stress responses (Fig S2J).

While the proteome was less dynamic, we observed several differentially expressed proteins upon nutrient stress. Some targets displayed a correlation between changes in protein abundance and polysome-bound RNA levels, including Egr2, Ifrd1, Txnip, Irf8, Slc6a9, Fos, and Il2ra (Fig S3C). Moreover, there were significant associations between the proteome and polysome-associated transcripts (Fig S3D). This suggests that although stress-driven RNA dynamics occur rapidly, alterations are also detectable within the proteome during this acute response.

Overall, these data suggest that environmentally stressed CD8⁺ T cells rapidly remodel gene expression networks throughout the transcriptome, translatome, and proteome, with marked conservation across conditions and selectivity across targets. Under acute stress, cells deprioritize translation of growth-related programming in favor of stress-adaptive programming (Fig 2L). Here, transcription factors like ATF4 and CEBPG and metabolic genes are associated with acute responses to stress, highlighting a potential link between gene regulation, metabolic reprogramming, and functional recovery of CD8⁺ T cells.

Loss of mTORC1 and/or translation limits metabolic demand, while ISR signaling drives amino acid programming to restore function

Both mTOR and ISR signaling broadly regulate metabolism, gene expression, and cap-dependent translation in response to nutrient availability or cellular stressors^47,50. Accordingly, we observed that nutrient stress dampened mTORC1 activity and activated the ISR (Fig 3A, Fig S4A). We further observed that nutrient stress rapidly induced multiple ISR-related transcripts, including Atf3, Nupr1, Ppp1r15a, and Sesn2 (Fig 3B, Fig S4B)⁵¹. This gene signature was sensitive to nutrient concentrations, suggesting that the magnitude of ISR signaling scales with nutrient abundance (Fig S4C). Similarly, we observed dampened mTOR and elevated ISR signaling in cells cultured in MC38 supernatant (Fig S4D, Fig S4E).

Figure 3: The ISR, mTOR, and global translational capacity govern discrete metabolic and regulatory adaptations.

A) mTORC1 (p-S6) and ISR signaling (ATF4) in nutrient stress cultured CD8⁺ T cells over time. Numbers represent quantification of band density relative to actin control.

B) Heatmap of change in ISR-related transcripts from Han et al. 2023 in nutrient stress media vs control.

C) Schematic of experimental approach. Left: Nutrient stress models concurrently alter nutrient-sensitive signaling, translation, and metabolite availability. Right: Pharmacological tools isolate stress-driven changes in signaling, translation, and metabolism without confounding effects of substrate availability.

D) Effect of drugs on mTORC1 (p-S6) and ISR signaling (ATF4) over time. Numbers represent quantification of band density relative to actin control.

E) Intracellular cytokine production and CD98 expression in Torin or 4Egi1 treated CD8⁺ T cells after 6 (top) or 24 hours (bottom).

F) Intracellular cytokine production and CD98 expression in halofuginone treated CD8⁺ T cells after 6 (top) or 24 hours (bottom).

G) Polysome profiles of drug treated CD8+ T cells after 3 hours. Inset shows polysome region.

H) Change in ISR-related transcripts and ribosomal/mitochondrial pathways in drug treatments (3 hours) vs control for input (left), and polysome-associated RNA (right)

I) Oxygen consumption rate of drug treated cells over time via Resipher assay

J) Relative cellular amino acid abundance under drug treatments at 6 hours (vs. DMSO).

K) Model of the contributions of ISR / mTOR / global translation to stress adaptation

Data points represent technical replicates. Error bars represent SEM. Individual experiments performed at least twice. **p < 0.01, ****p < 0.0001 by Brown-Forsythe and Welch ANOVA tests (E) and Welch’s t test (F).

To isolate the contribution of these pathways without the confounding effects of reducing substrate availability for translation or energy production, we employed a pharmacological approach, culturing CD8⁺ T cells with either Torin1 (an mTOR inhibitor), halofuginone (an ISR agonist), or 4EGi-1 (an eIF4F-dependent translation inhibitor) (Fig 3C)^52–54. As expected, Torin1 treatment dampened p-S6, while halofuginone increased ATF4 levels and stress-related transcripts (Fig 3D, Fig S5A and B). Consistent with mTORC1 activity driving ATF4-dependent gene expression in other cell types, Torin treatment decreased stress-related transcript expression^55–59. Treatment with 4EGi-1 was also sufficient to elevate ATF4 protein and stress-related transcripts, suggesting translational insults can drive ISR induction (Fig 3D, Fig S5A and B). Over time, we observed that mTOR signaling or cap-dependent translation inhibition acutely impaired cytokine production, which failed to recover (Fig 3E, Fig S5C). Conversely, ISR activation had minimal functional effects acutely, but yielded increased function long-term and drove the rapid induction of Slc7a5 and Slc3a2 transcripts and CD98 protein expression, highlighting ISR signaling as a mediator of stress-induced CD98 expression in CD8⁺ T cells, consistent with cancer cell studies (Fig 3F, Fig S5D, E, and F)⁶⁰.

We next performed RNA-seq and polysome profiling on halofuginone, Torin1, or 4EGi-1 treated cells. Whereas halofuginone mildly impacted polysome abundance, Torin and 4EGi-1 treatment robustly decreased polysome abundance and increased monosome abundance (Fig 3G). Halofuginone also induced ISR-associated programming including genes involved in amino acid transport and synthesis including Slc7a11, Slc6a9, Slc1a4, Slc38a2, and Gpt2 (Fig 3H, Fig. S5G). In contrast, Torin and 4EGi-1 treatment depleted polysome-associated cytosolic ribosome, mitochondrial ribosome, and mitochondrial respiratory chain transcripts relative to input RNA, indicating translational repression (Fig 3H). Accordingly, Torin or 4EGi-1 treatment led to a rapid decrease in mitochondrial oxygen consumption rate, an effect only slightly apparent under halofuginone treatment. (Fig 3I). Together, these data suggest that diminished mTOR signaling and/or cap-dependent translation underlies the repression of biosynthetic and growth-related programming during nutrient stress, while ISR induction drives adaptive metabolic reprogramming and amino acid handling to support stress resolution and functional recovery.

Lastly, we examined if ISR induction, loss of mTOR signaling, or dampening of 4EGi-sensitive translation impacted the amino acid profile of CD8⁺ T cells. Torin and 4EGi-1 treatment both led to broadly elevated intracellular amino acid levels, while halofuginone drove more selective changes in serine, alanine, asparagine, proline, and glutamate, suggesting diminished translational demand globally impacts amino acid availability while ISR activation can drive selective changes in amino acid abundance (Fig 3J).

Altogether, these findings support a model where dynamic control of mTOR signaling, cap-dependent translation, and ISR activation drive complementary metabolic and regulatory responses that shape functional capacity during T cell nutrient stress adaptation (Fig 3K). Acute loss of mTOR signaling and translational capacity reduces global biosynthetic and mitochondrial activity, limiting function and reducing amino acid demand. In parallel, ISR activation induces a selective metabolic response, permitting long-term functional recovery without wholesale loss of translational or mitochondrial capacity.

The ISR is required for CD8⁺ T cell functional resilience during anti-tumor immunity

We next asked whether ISR-driven responses were necessary to support CD8⁺ T cells under nutrient stress, using the general control nonderepressible 2 (GCN2) inhibitor GCN2iB to inhibit GCN2 signaling (Figure 4A).⁶¹ GCN2 inhibition impaired induction of ATF4 by halofuginone and selectively abrogated CD8⁺ T cell function and survival in halofuginone, suggesting GCN2 is important for stress adaptation (Figures 4B and 4C). GCN2iB-treatment also decreased cell viability and CD98 expression after BCAA, arginine, or methionine depletion, but not glucose depletion or control conditions (Figure 4D).

Figure 4: The ISR kinases GCN2 and HRI prevent CD8⁺ T cell dysfunction during anti-tumor immunity.

A) Schematic of the integrated stress response downstream of amino acid stress or halofuginone, highlighting the contribution of GCN2-mediated sensing, translational responses, and transcriptional response downstream of stress sensitive transcription factors.

B) Validation of inhibition of halofuginone-induced ISR signaling by GCN2iB

C) Viability and intracellular cytokine production of CD8+ T cells treated with halofuginone in the presence of DMSO (closed) or GCN2iB (open) for 24 hours.

D) Viability and CD98 expression of CD8+ T cells treated with nutrient stress conditions in the presence of DMSO (closed) or GCN2iB (open) for 24 hours.

E) Experimental schematic: CD45.1 Rosa26 (control) P14 CD8⁺ T cells were co-transferred 1:1 with CD45.1.2 GCN2 KO or HRI KO P14 CD8+ T cells into gp33-MC38 tumor-bearing mice to assess anti-tumor responses. n = 17 mice/group from 3 independent replicates.

F) Chimerism of tumor infiltrating P14 CD8⁺ T cells from Rosa (control), GCN2 KO or HRI KO donors.

G) Intracellular IFNg production in dLN P14 CD8⁺ T cells from Rosa (control), GCN2 KO or HRI KO donors.

H) Representative flow cytometry plots of intracellular IFNg production in dLN P14 CD8+ T cells from CD45.1 Rosa (control) and CD45.1.2 GCN2 KO or HRI KO donors.

I) Exhausted T cell (T_EX) differentiation in tumor infiltrating P14 CD8⁺ T cells Rosa (control), GCN2 KO or HRI KO donors.

J) Representative flow cytometry plots of CD69 and Ly108 expression in tumor infiltrating P14 CD8⁺ T cells.

K) Distribution of exhausted T cell subsets among tumor infiltrating P14 CD8+ T cells from CD45.1 Rosa (control) and CD45.1.2 GCN2 KO or HRI KO donors, as assessed by CD69 and Ly108 expression.

Data points represent technical replicates (C and D) or individual mice (F-K). Error bars represent SEM. Individual experiments performed at least twice. **p < 0.01, ***p < 0.001, ****p < 0.0001 by multiple comparisons paired T-test with Holm-Šídák method (F, G, and I).

To extend these findings in vivo, we investigated the requirement for the ISR during anti-tumor immunity. While GCN2 is the canonical ISR kinase during amino acid stress, the complex environmental stressors present in the tumor microenvironment (TME) can engage other ISR kinases like heme regulated inhibitor (HRI). We co-transferred congenically marked GCN2- or HRI-deficient P14 TCR-transgenic CD8⁺ T cells into mice bearing subcutaneous gp33-expressing MC38 tumors, along with Rosa26-deficient controls (Fig 4E). Loss of either stress kinase resulted in reduced P14 CD8⁺ T cell tumor infiltration and cytokine production (Fig 4F–H). Concurrently, GCN2 and HRI KO cells exhibited accelerated exhausted T cell differentiation, with a loss in early progenitor exhausted T cells (Fig 4I–K). These data indicate that CD8⁺ T cells require multiple ISR kinases to maintain functional resilience within the TME.

ISR-activating kinases converge on eIF2α phosphorylation, promoting the selective expression of transcription factors (Fig 4A)^7–9. While ATF4 is the best characterized factor, other basic leucine zipper (bZIP) transcription factors like C/EBP family members can dimerize with ATF4 and modulate transcriptional response^56,62,63. As we observed the association of ATF4 and CEBPG motifs with loci for nutrient stress-induced transcripts, we tested whether each factor was required for CD8⁺ T cell functional resilience. We found both ATF4 and CEBPG were selectively required to sustain maximal cytokine production during halofuginone treatment, yet had little functional impact in control conditions (Fig 5A and B). Moreover, loss of either ATF4 or CEBPG dampened stress-induced CD98 (Fig 5C). We further evaluated whether ATF4 or CEPBG govern the recovery of CD8⁺ T cells once environmental conditions are restored. Whereas halofuginone-treated Rosa26 T cells showed comparable cytokine production after rescue, halofuginone-treated ATF4 and CEBPG KO cells displayed sustained defects even after rescue, suggesting failure to adapt (Fig 5D). ATF4 and CEBPG KO CD8⁺ T cells also showed dampened cytokine production capacity after exposure to tumor conditioned media that was not observed in controls (Fig 1D, Fig 5E–G). Collectively, these findings demonstrate that an ISR-mediated transcriptional program is essential for CD8⁺ T cell functional adaptation to environmental stress, but largely dispensable in replete environments.

Figure 5: ATF4 and CEBPG are required CD8⁺ T cell functional resilience to environmental stress.

A) Experimental schematic: Rosa26 (control), ATF4 KO or CEBPG KO CD8⁺ T cells were cultured +/− halofuginone for 24 hours and restimulated during the final 6 hours. Alternatively, cells were exposed to halofuginone for 24 hours, then returned to normal media for 24 hours prior to restimulation.

B) Intracellular cytokine production in DMSO or halofuginone treated Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells after 24 hours.

C) CD98 expression in DMSO or halofuginone treated Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells after 24 hours.

D) Intracellular cytokine production in Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells following 24 hours recovery from DMSO or halofuginone treatment.

E) Experimental schematic: Rosa26 (control), ATF4 KO or CEBPG KO P14 CD8+ T cells were cultured in tumor supernatant or control media for 24 hours and restimulated during the final 6 hours.

F) Intracellular cytokine production in Rosa26 (control), ATF4 KO, or CEBPG KO CD8+ T cells cultured for 24 hours in tumor conditioned media (Tumor Sup) or control RPMI 1640.

G) Representative flow cytometry plots of intracellular IFNg and TNFa production in Rosa26 (control), ATF4 KO, or CEBPG KO CD8+ T cells cultured for 24 hours in tumor conditioned media (Tumor Sup) or control RPMI 1640.

H) Experimental schematic: Rosa26 (control), ATF4 KO or CEBPG KO P14 CD8⁺ T cells were transferred into gp33-MC38 tumor-bearing mice to assess anti-tumor function. n = 30 mice/group from 3 independent replicates.

I) gp33-MC38 tumor growth in recipient mice receiving an adoptive transfer of 2×10^6 Rosa26 (control), ATF4 KO, or CEBPG KO P14 CD8+ T cells at day 7.

J) Intracellular IFNg production in Rosa26 (control), ATF4 KO or CEBPG KO dLN P14 CD8⁺ T cells (total cells) and amongst IFNg+ cells, relative to Rosa26 (control) MFI for each experimental replicate.

K) Representative flow cytometry plots of intracellular IFNg production in CD45.1 Rosa (control), ATF4 KO or CEBPG KO dLN P14 CD8+ T cells.

L) Frequency of tumor infiltrating P14 CD8⁺ T cells from CD45.1 Rosa (control), ATF4 KO or CEBPG KO donors.

M) Tox1 expression in tumor infiltrating Rosa26 (control), ATF4 KO or CEBPG KO P14 CD8⁺ T cells, relative to Rosa26 (control) MFI for each experimental replicate.

N) Representative flow cytometry plots of intracellular Tox1 expression in tumor infiltrating CD45.1 Rosa (control), ATF4 KO or CEBPG KO P14 CD8+ T cells.

Data points represent technical replicates (B-D and F) or individual mice (I-N). Error bars represent SEM. Individual experiments performed at least twice. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001 by two-way ANOVA with Šídák's multiple comparisons test (B-D and F) and Brown-Forsythe and Welch ANOVA tests (J, L, and M).

Lastly, we asked whether these transcription factors were essential for anti-tumor T cell responses in vivo. In contrast to control cells, both ATF4- and CEBPG-deficient CD8⁺ T cells failed to suppress tumor growth, showing dampened cytokine production capacity (Fig 5H–K). Further, ATF4-deficient cells displayed markedly decreased intratumoral frequency, while CEPBG KO cells exhibited elevated expression of TOX (Fig 5L–N). Altogether, we find that the ISR, from initiating kinases to downstream transcription factors, is required for CD8⁺ T cell resilience to environmental stress and optimal immunity against solid tumors.

ATF4 and CEBPG coordinate adaptive metabolic rewiring to restrain runaway stress

To understand how ATF4 and CEBPG sustain CD8⁺ T cell function under environmental stress, we performed RNA-seq on DMSO or halofuginone-treated control (Rosa26), ATF4 KO, or CEBPG KO T cells (Fig 6A and E). Long-term ISR induction led to widespread transcriptional reprogramming, including many metabolic genes previously observed in acute nutrient stress such as amino acid transporters and synthesis enzymes (Fig 6A and B, Fig S6A). However, halofuginone treatment also drove the expression of a dysfunction-related gene signature, including anti-proliferative programming (Btg1, Cdkn1b), proapoptotic programming (Fas, Ddit3), and inhibitory receptors (Tigit, Lag3), along with a widespread decrease transcripts related to cell cycle progression, central carbon metabolism, and cholesterol biosynthesis (Fig 6B–E, Fig S6A and B). While ATF4 and CEBPG were essential for the expression of a discrete module of genes (Cluster 4), their loss also led to a global amplification of halofuginone-altered programming (Fig 6B and C, Fig S6A). Concordantly, we observed increased p-eIF2a in ATF4 KO cells and increased ATF4 levels in CEBPG KO halofuginone treated cells (Fig S6C). These data illustrate a distinction between adaptive and maladaptive programming within the broader transcriptional response to stress; While environmental stress primes both adaptive and dysfunction-related programming, ATF4/CEBPG regulate a selective sub-network of targets that mitigate stress amplification and promote resolution.

Figure 6: An ATF4-CEBPG axis drives metabolic rewiring to prevent runaway stress-driven dysfunction.

A) Heatmap of all differentially-expressed transcripts in DMSO or halofuginone treated Rosa26, ATF4 KO, and CEBPG KO CD8⁺ T cells after 24 hours. Left annotation: k-means cluster membership and expression value in Rosa26 control condition. Left main: Data from halofuginone-treated conditions are plotted relative to Rosa26 cells cultured in DMSO. Right main: Data from halofuginone-treated conditions are plotted relative to Rosa26 cells treated with halofuginone.

B) Heatmap of changes in transcript expression for genes in Cluster 4. Genes containing highly conserved ATF4 binding sites from Örd et al., 2023 noted in purple.

C) Heatmap of changes in transcript expression for genes in Clusters 1–3.

D) Heatmaps of changes in transcript expression for all annotated genes from indicated pathways.

E) Euler diagram of differentially expressed genes across all groups due to ATF4 (red fill) / CEBPG (blue fill) KO in the presence of DMSO (black outline) or halofuginone (green outline); or due to halofuginone treatment (grey)

F) Euler diagram of differentially expressed genes in only ATF4 KO (red fill) or CEBPG (blue fill) KO, in the presence of DMSO (black outline) or halofuginone (green outline)

G-H) Oxygen consumption rate (OCR) (I) and extracellular acidification rate (ECAR) (J) in DMSO (closed) or halofuginone (open) treated Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells after 24 hours.

I) Labeling of selected metabolites from U-13C glucose over final 18 hours in Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells treated with DMSO or halofuginone for 24 hours.

J) Labeling of selected metabolites from U-13C glutamine over final 18 hours in Rosa26 (control), ATF4 KO, or CEBPG KO CD8⁺ T cells treated with DMSO or halofuginone for 24 hours.

K) Model of ATF4 / CEBPG-mediated metabolic and functional adaptation in CD8⁺ T cells under stress

Bar and line graphs represent technical replicates. Error bars represent SEM. Individual experiments performed at least twice. Statistical significance for all tracing data was assessed by two-way ANOVA with Tukey’s multiple corrections test (Table S2 and S3).

Whereas stressed ATF4 and CEBPG KO cells showed similar gene expression profiles, each transcription factor regulated distinct signatures in control and stress conditions, indicating highly context-specific activities and distinct regulatory networks (Fig 6E and F). This dichotomy was apparent in canonical ISR target genes and Cluster 4, which displayed strong ATF4/CEBPG-dependent expression during stress and little to no dependency in control conditions (Fig 6B, S6B). When evaluating each gene set for the presence of highly conserved, CHIP-seq annotated ATF4 binding sites, ATF4 bound genes were significantly enriched in halofuginone-treated cells, with none detected in DMSO CEBPG KO cells (Fig 6B, S6B and E).⁶⁴ ATF4 and CEBPG also regulated the stress-induced expression of many other transcription factors, including Atf5, Batf3, and Nr4a1 (Fig S6D). These data support a model in which ATF4 and CEBPG participate in a cooperative transcription factor network regulating unique gene sets in stress and unstressed conditions.

We next investigated the role of ATF4 and CEPBG in supporting metabolic function in CD8⁺ T cells under stress. Long-term halofuginone treatment diminished the glycolytic and respiratory rate of CD8+ T cells, with ATF4 or CEBPG KO further decreasing metabolic capacity compared to Rosa26 control cells (Fig 6G and H, Fig S7A). ATF4 or CEBPG KO also impaired glycolysis in control conditions, indicating a broader role in T cell glycolytic capacity. (Fig 6G, Fig S7A). Altogether, these data indicate that while ongoing stress responses dampen central carbon metabolism, ATF4 and CEBPG maintain metabolic capacity under stress.

Lastly, we explored the ATF4/CEBPG-dependent metabolic processes sustaining T cells function during stress, performing stable isotope labeling using U-¹³C-labeled glucose or glutamine in ATF4- and CEBPG-deficient T cells in the presence or absence of halofuginone (Table S2 and S3). Halofuginone decreased glucose-derived lactate production, which was further diminished with ATF4 or CEBPG KO and accompanied by decreased glycolytic intermediate abundance (Fig 6I, Fig S7B). Conversely, halofuginone elevated production of glucose-derived amino acids in an ATF4- and CEBPG-dependent manner, including alanine, serine, and glycine which support T cell expansion and function (Fig 6I, Fig S7C)^65–68. ATF4 and CEBPG promoted the mRNA expression of the alanine synthesis enzyme Gpt2 and the alanine transporters Slc1a4 and Slc1a5 during stress, leading to an ATF4/CEBPG-dependent increase in both labeled and unlabeled fractions (Fig S7D). Halofuginone treatment also led to reduced glutamine- and glucose-derived carbons entering the tricarboxylic acid (TCA) cycle, which was further enhanced by the loss of ATF4 and CEBPG (Fig 6J, Fig S7E and F). Halofuginone-treated ATF4 and CEBPG KOs also displayed a buildup of aspartate, suggesting dysregulated aspartate-asparagine homeostasis from decreased Asns expression (Fig 6B, Fig S7E and F). Lastly, consistent with ISR-driven CD98 expression, ATF4 and CEPBG were required to maintain leucine and isoleucine levels under halofuginone treatment (Fig 5A, Fig S9G).

These findings support a model whereby ATF4 and CEBPG orchestrate CD8⁺ T cell stress resilience through an “adaptive” metabolic program supporting amino acid accumulation and preservation of central carbon metabolism (Fig 6K). ATF4- and CEBPG-dependent metabolic gene expression mitigates the maladaptive effects of environmental stress and preserves function, despite dampened central carbon metabolism. In their absence, CD8⁺ T cells under stress fail to maintain amino acid levels and mitochondrial anaplerosis, leading to collapse of central carbon metabolism, amplification of a dysfunction-related transcriptional profile, and immunological dysfunction.

Discussion

Gene-regulatory mechanisms that control metabolism and cell behavior in response to environmental perturbation have been a longstanding focus of biology^69–71. Adaptive metabolic responses to extracellular nutrient levels enable unicellular organisms to grow or persist across diverse biochemical conditions^72–78. In multicellular organisms, cells must balance nutrient utilization for growth and self-maintenance against usage for host protective lineage-specific functions. Studies of transformed cells have revealed how both intrinsic and environmental signals regulate cell growth and metabolism, as well as how cells are able to grow or survive at the expense of the host when these regulatory systems fail^1,2,79,80. How non-transformed cells prioritize nutrient use across growth, survival, and/or lineage-specific functions during environmental stress remains comparatively unknown. Effector T cells are tasked with providing protective immunity in metabolically-disrupted sites such as infected tissue or tumors, offering a dynamic system to understand the coordination of metabolism, gene expression, and nutrient allocation under environmental stress, with potential immunotheraputic applications.

We propose that “biosynthetic plasticity” — the ability to rapidly alter global nutrient handling and reprioritize nutrient allocation to discrete metabolic and cellular programs in response to environmental change — represents a unique mechanism preserving lineage-specific function across biochemical contexts. In comparison to fuel choice plasticity, where varying carbon sources are consumed to maintain energetic capacity resulting in distinct functional outcomes, biosynthetic plasticity allows cells to sustain an existing functional program despite diminished energetic capacity^{33,66,81–90}. Our work reveals CD8⁺ T cells under acute stress rapidly modulate nutrient-sensitive signaling and translational activity, reconfiguring cellular nutrient handling while shifting the translational priority of discrete cellular pathways. This early environmental stress response licenses an adaptive transcriptional program and establishes an alternate metabolic mode that mitigates cell dysfunction and prevents metabolic collapse, promoting functional resilience required for anti-tumor immunity.

Our work adds to a growing body of literature evincing a key role for translational control and metabolism-translation cross-talk in T cell responses^{30,31,48,91–101}. We find the mTOR and ISR pathways jointly govern metabolic activity, translational capacity, and regulatory programming to shape immune function under stress, similar to other cellular systems^{10,50,56,58,59,102–106}. In T cells, mTORC1 activity supports effector differentiation, proliferation, and function while stress-responsive signaling has been shown to exert both adaptive and maladaptive outcomes^{7–9,34,37,107–115}. Indeed, we observed that acute nutrient stress leads to diminished function, decreased mTORC1 signaling and increased ISR activity. This loss of mTORC1 activity dampens T cell function while decreasing global translational activity, repressing the translation of growth-related programming and increasing intracellular amino acid content. Concurrently, the ISR induces an adaptive program that is efficiently translated under stress and restores T cell function over time. Thus, these signaling responses harmonize global translational and biosynthetic activity with environmental constraints, promoting cellular resilience to nutrient stress and robust protective immunity across varying metabolic microenvironments.

We further highlight a key role for the ISR in T cell resilience to TME stressors, including nutrient deprivation and hypoxia. We find that CD8⁺ T cells require the capacity to flexibly respond to these various stress sources through the ISR, rather than one sensor or stressor having a dominant role. Thus, while specific sources of TME stress cause T cell dysfunction through distinct mechanisms, the ISR represents a central stress-response switchboard promoting resilience. Accordingly, we find that ATF4 and CEBPG are similarly required for protective CD8⁺ T cell anti-tumor immunity and ATF4 and CEBPG KOs under stress show increased expression of dysfunction-related targets (e.g. Btg1, Nr4a1, Lag3, Tigit)^{24,25,29,86,116–122}. Similarly, both GCN2 and HRI were required to limit exhausted T cell differentiation as was CEBPG to limit TOX expression. Thus, inability to control environmental stress can be a key driver of exhaustion alongside chronic antigen exposure.

While our study reveals the essentiality of the ISR in preventing CD8⁺ T cell dysfunction within the TME, recent gain-of-function studies have shown that enforcing ATF4 expression can be also be detrimental to anti-tumor immunity¹²³. These seemingly disparate findings help illustrate the hormetic nature of the ISR: appropriate signaling allows stress resolution, stress signaling deficits prevent resilience, and sustained or unresolved stress signaling leads to programmed cell dysfunction and death. Additional work is required to dissect how environmental stress and ISR-driven programming contribute to dysfunctional CD8⁺ T cell states.

Though ATF4 and CEBPG regulate a narrow subset of stress-sensitive genes, they cooperatively support metabolic rewiring and are required to prevent the amplification of dysfunction-related programming. Of note, while the ATF4 and CEBPG regulomes are similar under stress, each factor differs significantly between stress and control conditions, suggesting that the environmental context shapes their activities. This observation, alongside ATF4/CEBPG’s regulation of other transcription factors during stress, highlights the dynamic and interconnected nature of stress adaptation responses, presenting an opportunity for further dissection of environmental feedback on this larger stress-sensitive regulatory network.

Going forward, it will be important to understand the molecular mechanisms enforcing specificity in stress-induced regulatory responses, including alternate modes of translation, upstream open reading frames, unique RNA elements, post-transcriptional modifications, and control of translation through regulatory RNAs or RNA-binding proteins^{44,45,47,91,92,99,100,124,125,125–132}. Manipulating these processes would open avenues for preserving translational “priority” of desired transcripts in cell and gene therapies. Additionally, emerging technologies could allow for systematic dissection of the combinatorial transcription factor-environment interactions underlying stress-adapted and dysfunctional states in T cells and other non-transformed cell types^133–137. Similarly, it will be important to understand how environmental conditions across microenvironments impact the functional programs of other immune cell lineages^138–143. This could inform strategies utilizing metabolic conditioning or cell engineering for therapeutic applications^{20,22,115,144–146}. As sites of peripheral immune function contain multiple cell types and stressors, additional models of immune cell adaptation across tissue and disease contexts will enable a more complete understanding of environmental stress on immunity^138,147,148. Lastly, the stress-sensitive targets across our study’s datasets may prove useful for discovering factors promoting stress-resilience or susceptibility to be therapeutically targeted beyond the immune system - for example, in cancer, where stress adaptation pathways, translational and metabolic status, and environmental conditions modulate therapeutic efficacy or resistance^60,149–164.

Limitations of the Study

This study reveals the capacity of CD8⁺ T cells to adapt to environmental nutrient stress and recover effector function. To uncover the underlying molecular mechanisms, we employed reductionist nutrient deprivation and ISR agonist systems. While these studies revealed a conserved network of shared stress-adaptive responses, there were many nutrient-stress specific responses whose mechanisms remain undefined. The relative kinetics of stress adaptation and the reversibility of stress resilience across in vivo settings are also unclear, as the sources of environmental stress likely differ across models and tumor types. Furthermore, the MC38 tumor model used in this work may not capture the role of the ISR in other cancer types or autochthonous tumors, where the nature of environmental stress may evolve over time. Likewise, as metabolite tracing studies were performed in RPMI-1640, the roles of ATF4 and CEPBG in metabolic adaptation in more physiologic environments warrant further investigation.

Resource availability

Lead contact

Requests for further information, resources, and reagents should be directed to the lead contact, Will Bailis (bailisw@chop.edu).

Materials availability

This study did not generate new, unique reagents.

Data and code availability

• All sequencing data generated or analyzed during this study are included in this manuscript and are available at the Gene Expression Omnibus accession numbers GSE287894 and GSE287987.

• This study did not generate any original code.

• Any additional information required to reanalyze the data reported in this paper can be requested from the lead contact.

STAR Methods

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

Mice

C57BL/6J (B6, Jax #000664), Tg(TcraTcrb)1100Mjb (OT-I, Jax #003831), B6.Cg-Tcratm1MomTg(TcrLCMV)327Sdz/TacMmjax (P14, Jax #037394), and B6.SJL-Ptprca Pepcb/BoyJ (CD45.1, Jax #002014) mice were obtained from The Jackson Laboratory and maintained under specific pathogen-free conditions. Experiments were performed with male or female mice aged between 6 and 9 weeks of age. All experiments were performed in accordance with the Institutional Animal Care and Use Committee of the Children’s Hospital of Philadelphia (IAC 21–001325).

Cell Lines

MC38 cells were a gift from the lab of Ken Cadwell (University of Pennsylvania, Philadelphia, PA). gp33-expressing MC38 cells were a gift from the lab of E. John Wherry (University of Pennsylvania, Philadelphia, PA). MC38 and gp33-MC38 cells were maintained at <80% confluence by trypsinization and passaging every two days for one week before use in further experiments. Cells were cultured in Dulbecco's Modified Eagle Medium (DMEM, Gibco #11965118) 10% fetal bovine serum Corning, #MT35071CV), additional 2mM L-glutamine (Gibco, #25030081), and 1x penicillin/streptomycin (Gibco, #15140122).

METHODS DETAILS

CD8+ T cell Isolation and Stimulation

Bulk mouse CD8+ T cells were isolated from spleen and peripheral lymph nodes by ACK-lysis and magnetic negative selection using an EasySep^™ Mouse CD8+ T Cell Isolation Kit (STEMCELL Technologies, #19853). For experiments without electroporation, purified CD8+ T cells were initially stimulated for 24 hours on tissue culture plates coated with 2.5ug/mL anti-CD3ε (clone 145–2C11, BioLegend, #100360, RRID:AB_2800555) and 2.5u/.mL anti-CD28 antibodies (clone 37.51, BioLegend, #102122, RRID:AB_11147170), then removed from stimulation for further culture. For experiments involving electroporation, cells were stimulated for 24 hours on plates coated with 5ug/mL anti-CD3ε and 5ug/mL anti-CD28 antibodies before electroporation.

Media and Cell Culture

Initial T cell stimulation and culture was performed in supplemented lymphocyte culture medium (“T cell culture medium”, TCM): Roswell Park Memorial Institute (RPMI) 1640 Medium (Gibco, #11875093), supplemented with 10% fetal bovine serum (Corning, #MT35071CV), additional 2mM L-glutamine (Gibco, #25030081), 1mM sodium pyruvate (Gibco, #11360070), 25mM HEPES (Gibco, #15630080), 1x penicillin/streptomycin (Gibco, #15140122), 55uM 2-mercaptoethanol (Gibco, #21985023), and 5ng/mL recombinant mouse IL-2 (BioLegend, #575406).

MC38-conditioned supernatant was generated by passaging MC38 cells for one weekend allowing the cell cultures to reach 80% confluence, then replacing the culture media with T cell culture media (above) for 48 hours. Supernatant was then centrifuged for 15m at 4000xg to remove cells and utilized for further experiments.

All nutrient-stress media contained an RPMI-derived base medium: “Control” = RPMI-1640 Medium (Gibco, #11875093); No Glucose = RPMI-1640 Medium, no glucose (Gibco, #11879020); No Glutamine = RPMI 1640 Medium, no glutamine (Gibco, #21870076); No BCAA, RPMI 1640 Medium w/ L-Glutamine, w/o L-Isoleucine, L-Leucine, L-Valine (US Biological, #R8999–20); No Arginine = RPMI 1640 Medium for SILAC (Gibco, #A33823) + 40mg/L-Lysine (Thermo, #J62225.36); no Methionine = RPMI 1640 Medium w/o L-Glutamine, Methionine (US Biological, #R8999–06) + 2mM L-Glutamine (Gibco, #25030081)), 10% dialyzed fetal bovine serum (Cytiva, #SH30079.03), additional 2mM L-glutamine (except “No Glutamine” condition, Gibco #25030081), 1mM sodium pyruvate (Gibco, #11360070), 25mM HEPES (Gibco, #15630080), 1x penicillin/streptomycin (Gibco, #15140122), 55uM 2-mercaptoethanol (Gibco, #21985023), and 5ng/mL recombinant mouse IL-2 (BioLegend, #575406).

All cells were cultured in a CO2 incubator at 37C at 5% CO2.

Pharmacological Agents

All experiments utilizing pharmacological agents below were performed in “Control” media as described in the nutrient-stress media section above. All compounds were resuspended in DMSO (VWR #25–950-CQC) and used at the following final concentrations unless otherwise specified: Torin 1 (500nM, Cayman Chemical, #10997), Halofuginone (hydrochloride) (50nM, Cayman Chemical, #13370), 4EGi-1 (50uM, Cayman Chemical, #15362), GCN2iB (2uM, Cayman Chemical, #35897).

CRISPR-Cas9 KO by Electroporation

KO cells were generated via electroporation of Cas9-RNPs using a Thermo Neon instrument. Cas9-RNPs were generated by combining by 1.25uL 200uM target-specific crRNA (IDT), 1.25uL 200uM ATTO-550-labeled tracrRNA (IDT, #1075928), and 2.5uL nuclease-free duplex buffer (IDT) and incubating for 5m at 95C, then mixed with 5uL Cas9-NLS (QB3-Berkeley MacroLab) and incubated for 15m before use in electroporation. 2–10 × 10⁶ activated T cells were washed once with PBS, resuspended in 100uL Buffer R (Thermo #MPK10096), mixed with 10uL Cas9-RNP, then electroporated with the following parameters (1600V, 10ms, 3 pulses) and cultured in “T Cell Media” without Pen/Strep for 24 hours before further use. Viability and RNP uptake were quantified at 24 hours post-nucleofection by flow cytometry.

crRNA target sequences:

Mouse Rosa26 locus: ACTCCAGTCTTTCTAGAAGA

Mouse Atf4: AAGTTTAATAAAAGTCGACC

Mouse Cebpg: CAGAGAGCGGAACAATATGG

Mouse Eif2ak1 (HRI): CGTGGACGGCGAGCGCGACA

Mouse Eif2ak4 (GCN2): AAAGCCCGGACATACTCCTC

Restimulation and Flow Cytometry

All flow cytometry for in vitro profiling was performed on a CytoFLEX LX or CytoFLEX S cytometer (Beckman Coulter). Intracellular cytokine production was measured by adding brefeldin A (Invivogen) 30 min after transferring cells to a plate coated with 2.5ug/mL anti-CD3ε (clone 145–2C11, BioLegend, #100360, RRID:AB_2800555) and 2.5ug.mL anti-CD28 antibodies (clone 37.51, BioLegend, #102122, RRID:AB_11147170) and culturing for 6 hours before staining. Cells were first stained with surface antibodies and viability dye (30m at RT) diluted in PBS, fixed with 1x Fixation/Permeabilization Buffer for 15m at room temperature (eBioscience^™ Foxp3 / Transcription Factor Staining Buffer Set, Invitrogen, #00–5523-00), and lastly stained with intracellular antibodies diluted in 1x Permeabilization Buffer (eBioscience^™ Foxp3 / Transcription Factor Staining Buffer Set, Invitrogen, #00–5523-00) for 30m at room temperature. Reagents utilized: eBioscience Fixable Viability Dye eFluor 780 (1:1000, Thermo Fisher Scientific, #65–0865-18); CD8 alpha Monoclonal Antibody (53–6.7), FITC (1:300, Invitrogen #MA1–10303, RRID: AB_11153636); CD98 Monoclonal Antibody (RL388), PE, eBioscience^™ (1:300, Invitrogen #12–0981-81, RRID:AB_465792); Granzyme B Monoclonal Antibody (NGZB), PE-eFluor^™ 610, eBioscience^™, (1:300, Invitrogen, #61–8898-82, RRID: AB_2574670); IFN gamma Monoclonal Antibody (XMG1.2), APC, eBioscience^™ (1:300, #17–7311-82, RRID: AB_469504); TNF alpha Monoclonal Antibody (MP6-XT22), eFluor^™ 450, eBioscience^™ (1:300, #48–7321-82, RRID:AB_1548825).

RT-qPCR

RNA was isolated with the Zymo Quick-RNA MicroPrep Kit (Zymo Research, #R1050) with DNase I treatment according to the manufacturer’s instructions. cDNA was synthesized with 1ug of purified RNA using Thermo Maxima H Minus Reverse Transcriptase (Thermo Scientific, #EP0751) according to manufacturer’s instructions using 25 pmol oligo(dT)₁₈ (Thermo Scientific, #SO132) and 25 pmol random hexamer (Thermo Scientific, #SO142) primers and the following thermocycler protocol: 10 min at 25°C, 15 min at 50°C, 5 min at 85°C. cDNA was diluted 1:10 in molecular biology water and quantified via real-time quantitative polymerase chain reaction using PowerTrack^™ SYBR Green Master Mix (Thermo Scientific, #A46110) and 1.66uM forward and reverse primers on a Bio-Rad CFX96 or CFX384 instrument (Bio-Rad). 3–4 technical replicates were averaged, each quantified transcript was normalized to Rps18, and data from all samples were plotted as relative expression versus a chosen sample.

Mouse Atf3 F - AGAGTGCCTGCAGAAAGAGTCA

Mouse Atf3 R - CGGTGCAGGTTGAGCATGTAT

Mouse Ppp1r15a F - TCCTCTAAAAGCTCGGAAGGTACAC

Mouse Ppp1r15a R - TCTCGTGCAAACTGCTCCCA

Mouse Sesn2 F - ACCTTCCGTGCCCAGGATTAT

Mouse Sesn2 R - CCTGGAACTTCTCATCCAGCAG

Mouse Nupr1 F - CTGGCGGGCATGAGAGGAAG

Mouse Nupr1 F - TCTGTGGTCTGGCCTTATCTCCA

Mouse Slc3a2 F - CACTGGGGAGCGTACTGAATCC

Mouse Slc3a2 F - AGTCCTGCTTGCGACACACTC

Mouse Slc7a5 F - AGCTGTGGCTGTGGACTTCG

Mouse Slc7a5 R - CCCATTGACAGAGCCGAAGCA

Mouse Slc7a11 F - CCAAGGGCATACTCCAGAACA

Mouse Slc7a11 R - TAGGACAGGGCTCCAAAAAGT

Western Blot

Cells were pelleted, washed in PBS and lysed in RIPA lysis buffer (50 mM Tris (pH 8.0), 1 mM EDTA, 150 mM NaCl, 1% NP-40, 0.5% Na-deoxycholate, 0.1% SDS) containing 10 mM NaF, 1 mM Na3VO4, and cOmplete^™, Mini, EDTA-free Protease Inhibitor Cocktail (Roche, #11836170001). Protein lysate concentration was quantified via Bradford assay (Bio-Rad, #5000006) using bovine serum albumin as standard. 15–20ug protein lysate was mixed with reducing sample buffer (Boston BioProducts, #BP-111R), and samples were separated by SDS-PAGE on Any kD^™ Mini-PROTEAN^® TGX^™ Precast Protein Gels (BioRad, #456–8126) followed by transfer to a nitrocellulose membrane using the Trans-Blot Turbo Transfer System (BioRad) on the Turbo setting. Membranes were blocked with 5% w/v dry nonfat milk in Tris-buffered saline containing 0.1% Tween 20 (TBS-T) for 60 min. Primary antibodies were diluted in 1x Tris Buffered Saline (TBS) with 1% Casein (BioRad, #1610782) overnight at 4°C with shaking. Five TBS-T washes of 2 minutes each were performed before incubation with secondary antibody (diluted in 5% w/v dry nonfat milk in TBS-T) for 2 hours at room temperature with rocking. Three additional washes were performed before visualizing the chemiluminescent signal with SuperSignal West Pico PLUS (Thermo Scientific, #34580) or Femto Chemiluminescent Substrate (Thermo Scientific, #34095) on a ChemiDoc MP Imaging System (BioRad). For phospho-epitopes, blots were stripped using Restore^™ Western Blot Stripping Buffer (Thermo Scientific, # 21059) after visualization of phospho-epitopes before incubating with total antibody. Actin was quantified by incubation with anti-actin fAb at room temperature for 1h. For western blot time courses involving multiple gels, an identical (pooled) protein sample from cells at t=0 of the experiment was loaded on every gel as an internal control to account for changes in transfer efficiency. All membranes within the time course were imaged for the same amount of time for a given target.

Phospho-S6 Ribosomal Protein (Ser235/236) (D57.2.2E) XP^® Rabbit mAb, 1:5000 (Cell Signaling Technology, #4858)

S6 Ribosomal Protein (5G10) Rabbit mAb, 1:5000 (Cell Signaling Technology, #2217)

ATF-4 (D4B8) Rabbit mAb, 1:1000 (Cell Signaling Technology, #11815)

Goat anti-Rabbit IgG (H+L) Cross-Adsorbed Secondary Antibody, HRP (Invitrogen #G-21234, RRID: AB_2536530)

Goat anti-Mouse IgG (H+L) Secondary Antibody, HRP (Invitrogen, #A16066, RRID:AB_2534739)

hFAB^™ Rhodamine Anti-Actin Primary Antibody, 1:2000 (Bio-Rad, #12004164)

Polysome Fractionation

Sucrose solutions were prepared in polysome extraction buffer (10 mM HEPES, 100 mM KCl, 5 mM MgCl2, pH 7.4, 100 ug/ml Cycloheximide, 2 mM DTT). Sucrose gradients were prepared in SW41 ultracentrifuge tubes by mixing 15% and 45% sucrose solution using a BioComp Gradient Master 108. 20M cells per sample per condition were resuspended after 3 hours of culture in nutrient stress media, washed once with PBS containing 50ug/mL cycloheximide, and lysed in 500uL polysome extraction buffer containing 1% Triton-X and SUPERAse In (Thermo Scientific, #AM2696), followed by 30min incubation on ice with vortexing at 5min intervals. Lysates were centrifuged for at 12000g for 15 min at 4C, then 50uL lysate was snap frozen as “Input” RNA and 400uL lysate was loaded onto 15–45% sucrose gradients followed by centrifugation at 37K rpm for 2hr at 4°C in a SW41 rotor of an Optima XPN 80 Beckman ultracentrifuge. Gradients were fractionated with a speed of 800uL/min using a Biocomp piston gradient fractionator, which recorded the OD254nm. Fractions corresponding to 60 sec intervals were collected, labeled 1–15, and stored at −80c before RNA extraction.

250uL of each sucrose fraction was aliquoted for RNA isolation. 20uL of input RNA was diluted to 250uL with molecular biology water before processing. 750uL of Trizol LS containing 20pg of luciferase control RNA (Promega, #L4561) was added to each sample and thoroughly mixed. 1mL of 100% ethanol was added to each sample and thoroughly mixed. Each sample was loaded into a separate column of a Zymo Direct-Zol-96 plate and purified according to manufacturer’s instructions. Eluted RNA was quantified on a Nanodrop, and a pooled sample containing an equal volume of the RNA from polysome-bound fractions (fractions 9–13) was generated for downstream sequencing.

Total RNA-sequencing and mRNA-sequencing

For polysome profiling via total RNA-seq of nutrient stress conditions, RNA samples were sequenced at the Center for Applied Genomics at the Children’s Hospital of Philadelphia. RNA samples were checked for quality using an Agilent tapestation and libraries were prepared using the Illumina TruSeq Stranded Total RNA Gold according to manufacturer’s instructions. Sample input was 300ng, and samples were sequenced on a NovaSeq 6000 utilizing the S2 300 cycle kit v1.5 and demultiplexed using DRAGEN.

For polysome profiling via mRNA-seq of drug conditions, RNA samples were sequenced at Novogene. Samples were checked for quality using a tapestation and libraries were prepped for stranded mRNA sequencing at Novogene. Messenger RNA was purified from total RNA using poly-oligo-attached magnetic beads. After fragmentation, the first strand cDNA was synthesized using random hexamer primers. Then the second strand cDNA was synthesized using dUTP, instead of dTTP. The directional library was ready after end repair, A-tailing, adapter ligation, size selection, amplification, and purification. The library was checked with Qubit and real-time PCR for quantification and bioanalyzer for size distribution detection. After library quality control, different libraries were pooled based on effective concentration and targeted data amount, then subjected to Illumina sequencing.

Seahorse Extracellular Flux Analysis and Resipher

For Seahorse assay, 3 × 10⁵ cells were plated per well in a 96-well Seahorse assay plate precoated with Cell-Tak (Corning, #354240). For the mitochondrial stress test assay, cells were plated in Seahorse XF RPMI medium (Agilent, #103576–100) containing 10mM glucose, 2mM L-glutamine, and 1mM sodium pyruvate and subsequently treated with oligomycin (1.5 μM, Sigma-Aldrich, #O4876), BAM15 (2.5 μM, Cayman Chemical, #17811), and rotenone (0.5 μM, Cayman Chemical, #13995–1) / antimycin A (0.5 μM, Sigma-Aldrich, #A8674) at indicated time points. For the glycolysis stress test assay, cells were plated in glucose-free Seahorse XF RPMI (Agilent, #103576–100) with 2mM L-glutamine and 1mM sodium pyruvate prior to assay and subsequently treated with 10mM glucose, oligomycin (1 μM, Sigma-Aldrich, #O4876), and 2-deoxyglucose (50mM, Sigma-Aldrich, #D0051) at indicated time points. Data transformation was performed using Wave software.

For Resipher assay, 3 × 10⁵ cells per well were plated in a Nunc MicroWell 96 well plate (Thermo Scientific, #167008) and a Resipher sensing lid (Lucid Scientific) was placed on top. 2–4 wells were left empty for baseline correction. After connection with the Resipher hub, sensors were equilibrated for 6 hours before collecting continuous oxygen consumption measurements.

Adoptive Transfer Tumor Model

For in vivo anti-tumor efficacy and phenotyping of ATF4 and CEBPG KO T cells, gp33-expressing MC38 cells were passaged for a week in culture before engrafting 250K cells on the right flank of C57BL/6 recipient mice. On days 5, 6, and 7 post-tumor-engraftment, purified CD45.1.2 P14 CD8+ T cells were respectively stimulated, nucleofected with Cas9-RNP complexes against the Rosa26 locus (control), Atf4, or Cebpg, and 2 × 10⁶ cells were adoptively transferred into anesthetized, tumor-bearing hosts via retroorbital injection. For anti-tumor efficacy experiments, tumor volume was monitored every 2 days using calipers and calculated using the formula (L x W x W) / 2. Experimental groups were randomly assigned and evenly distributed to individual recipients across multiple cages, and the experimenter measuring tumor size was blinded to treatment groups. For TIL phenotyping, tumors were harvested at day 15 post-implantation, minced with scissors and digested in a solution of PBS containing 1% FBS, 20mM HEPES, 1mg/mL Collagenase D (Sigma, 11088858001), and DNAse I (Sigma, 10104159001) for 30 minutes at 37C with shaking. After digestion, tissue was homogenized and immune cells were enriched via centrifugation over a 40/80% Percoll density gradient for 30m at 250xg with brake = 1. Cells were then washed, treated with ACK lysis buffer, and resuspended in PBS for subsequent staining for flow cytometry. For restimulation experiments, TILs were cultured in T cell medium containing 500uM gp33 peptide and brefeldin A for 6 hours before proceeding with staining for flow cytometry.

For co-transfer experiments involving nucleofection (GCN2 and HRI KO), 250K gp33-expressing MC38 cells were engrafted on the right flank of C57BL/6 recipient mice. On days 5, 6, and 7 post-tumor-engraftment, purified CD45.1+ or CD45.1.2+ P14 CD8+ T cells were respectively stimulated, nucleofected with Cas9-RNP complexes against the Rosa26 locus (control), GCN2, or HRI, and 1 × 10⁶ Rosa26 (control) and GCN2 or HRI KO cells were each mixed and adoptively transferred into anesthetized, tumor-bearing hosts via retroorbital injection. Tumors were isolated at day 17 post implantation for immune cell enrichment and phenotyping by flow cytometry as above. Draining lymph nodes were isolated for restimulation and intracellular cytokine staining. After filtering through a 70uM cell strainer, cell suspensions were cultured in T cell medium containing 500uM gp33 peptide in the presence of brefeldin A for 6 hours before proceeding with intracellular staining as denoted above.

LC-MS Metabolomics

Following cell culture, 1.5 million cells per sample were washed with PBS, pelleted, and flash-frozen. Cell pellets were extracted in 1mL −80C 80% Methanol / 20% Optima H₂O containing a mix of 13C/15N-labeled amino acid internal standards (Cambridge Isotopes Laboratory, #MSK-A2–1.2). Samples were pulse-sonicated on ice with a sonic dismembranator (Fisher Scientific, Waltham, MA) 30 times over 15 seconds then incubated on ice for 10 min. Debris was pelleted at 12000 × g for 10 min at 4C and the solvent was dried under nitrogen gas using a blowdown evaporator (Organomation, West Berlin, MA). Dry metabolites were re-suspended in 300 μL of 5% MeOH, vortexed for 1 minute, dissolved using an ultrasonic bath for 15 min, spun down at 12000 × g for 10 min at 4C and distributed to HPLC vials for LC-HRMS analysis. A pooled QC sample was generated for monitoring intra-run variance by mixing 20 μL of each re-suspended experimental sample within the run.

Metabolites were separated using a Thermo Hypersil Gold, 150 × 2.1 mm, 1.9 μm using the UltiMate 3000 quaternary UHPLC (Thermo Scientific, Waltham, MA) equipped with a refrigerated autosampler (6C) and column heater (55C). Solvent A was water with 0.1% formic acid and solvent B was methanol with 0.1% formic acid. The gradient was as follows: 0.5% B at 0 min, 0.5% B at 2 min, 50% B at 6 min, 100 % B at 12 min, 100 % B at 16 min, and back to 0.5 % B at 17 min and kept for 3 more min for column re-equilibration. The flow rate was 0.45 ml/min. A Q Exactive HF mass analyzer (QE-HF) (Thermo Scientific, Waltham, MA) equipped with a heated electro-spray ionization (HESI) source was operated in positive mode in full scan at 120,000 resolution, AGC target = 1e6; Maximum IT = 100 ms; scan range, 70 to 800 m/z. The pooled QC samples were used for metabolite identification by generating MS/MS spectra (dd-MS2) of the top 10 features at 15,000 resolution, AGC target = 1e5, Maximum IT = 25 ms, and (N)CE/stepped NCE = 30, 50, 60v.

Stable Isotope Labeling and GC-MS Metabolomics

Following cell culture, 1.5 million cells per sample were washed with PBS, pelleted, and flash-frozen. Cell pellets were extracted in 640uL −80C 80% Methanol / 20% Optima H₂O. Extracts were dried using a speedvac and then treated with MOX for 1 hour at 60°C followed by a 45-minute incubation with tBDMS at 60°C. Samples were injected on GC-MS (Agilent Technologies) by the Analytical Core for Metabolomics and Nutrition at the British Columbia Children’s Hospital Research Institute. Metabolites were identified with m/z and retention times as shown in Table S1. Mass isotopologue distributions of selected metabolite ion fragments were quantified and corrected for natural isotope abundance using algorithms adapted from Fernandez et al¹⁸⁵. The code is available on https://github.com/Sethjparker/IntegrateNetCDF_WithCorrect (Accessed on November 25th. 2024) under MIT license.

Proteomics

Protein Extraction

Pellets were solubilized in 200 μL of extraction buffer containing 5% sodium dodecyl sulfate (SDS, Affymetrix), 50 mM TEAB (pH 8.5, Sigma), and protease inhibitor cocktail (Roche cOmplete, EDTA-free). Each sample (100 μL) was sonicated for 10 minutes at 20°C in a Covaris R230 focused-ultrasonicator (settings: Dithering Y=3.0, Speed=20.0, PIP=360.0, DF=30, CPB=200) to shear DNA and ensure complete solubilization. Samples were then centrifuged at 3000g for 10 minutes to clarify the lysate. Protein concentration was measured by intrinsic tryptophan fluorescence (excitation at 280 nm, emission at 350 nm) against an in-house E. coli lysate standard curve on a Synergy H1 microplate reader (BioTek).

In-Solution Digestion

Each sample (130 μg) was digested following the S-Trap (Protifi) manufacturer’s protocol¹⁸⁷. Briefly, proteins were reduced with 5 mM TCEP (Thermo), alkylated with 20 mM iodoacetamide (Sigma), and acidified with phosphoric acid (Aldrich) to reach a final concentration of 1.2%. Samples were then diluted with 90% methanol (Fisher) in 100 mM TEAB, loaded onto an S-trap column, and washed three times with 90% methanol in 100 mM TEAB. A 1:10 enzyme-to-protein ratio of Trypsin (Promega) and LysC (Wako) in 20 μL of 50 mM TEAB was added, and samples were digested at 37 °C in a humidity chamber for 18 hours. Peptides were eluted sequentially with 50 mM TEAB (40 μL), 0.1% TFA in water (40 μL), and 50/50 acetonitrile:water with 0.1% TFA (40 μL). The combined eluates were dried by vacuum centrifugation, desalted using the Phoenix peptide cleanup kit (PreOmics) per the manufacturer’s protocol, and eluted into autosampler vials. After drying, samples were reconstituted in 0.1% TFA containing iRT peptides (Biognosys, Schlieren, Switzerland). Peptide concentrations were measured at OD280 using a Synergy H1 microplate reader (BioTek) and adjusted to 400 ng/μL for injection.

Mass Spectrometry Data Acquisition

Samples were randomized and analyzed on an Exploris 480 mass spectrometer (ThermoFisher Scientific) coupled with an Ultimate 3000 nano UPLC system and EasySpray source. 2ug of each sample was loaded onto an Acclaim PepMap 100 75 μm × 2 cm trap column at 5 μL/min, then separated by reverse phase HPLC on a 75 μm id × 50 cm PepMap RSLC C18 column. Mobile phase A was 0.1% formic acid, and mobile phase B was 0.1% formic acid/acetonitrile. Peptides were eluted into the mass spectrometer at 210 nL/min using a 150-minute gradient from 3% B to 45% B.

Data independent acquisition (DIA) mass spectrometer settings were as follows: one full MS scan at 120,000 resolution, with a scan range of 350–1200 m/z and automatic gain control (AGC) target of 300%, and automatic maximum inject time. This was followed by variable DIA isolation windows, MS2 scans at 30,000 resolution with an AGC target of 1000%, and automatic injection time. The default charge state was 2, the first mass was fixed at 200 m/z, and the normalized collision energy for each window was set at 27.

Mass Spectrometry QA/QC and System Suitability

The suitability of the instrumentation was monitored using QuiC software (Biognosys; Schlieren, Switzerland) for the analysis of the spiked-in iRT peptides. As a measure for quality control, standard E. coli protein digest was injected in between samples and data was collected in data dependent acquisition (DDA) mode. The collected data were analyzed in MaxQuant¹⁸⁹ and the output was subsequently visualized using the PTXQC package¹⁹⁰ to track the quality of the instrumentation.

Database Searching

The DIA raw files were processed using Spectronaut 18.0 in direct DIA mode¹⁹¹. We utilized a mouse (mus musculus, 25,435 protein entries) database comprising canonical and reviewed isoforms from Uniprot, supplemented with a list of 245 common protein contaminants and iRT peptides. Enzyme specificity was set to trypsin with allowance for two potential missed cleavages. Fixed modification was specified as carbamidomethyl of cysteine, while protein N-terminal acetylation and oxidation of methionine were considered variable modifications. To ensure high confidence, a false discovery rate limit of 1% was applied for precursors, peptides, and proteins identification, while the remaining search parameters were maintained at their default settings.

QUANTIFICATION AND STATISTICAL ANALYSIS

RNA-sequencing analysis

RNA-Sequencing reads were initially assessed for quality with fastqc (0.12.1)¹⁶⁵ and multiqc (1.14)¹⁶⁶. Reads were trimmed of any remaining adapters and filtered with Trimmomatic (0.39)¹⁶⁷ using the following parameters: seedMismatches: 2, palindromeClipThreshold: 30, simpleClipThreshold: 10, Leading: 3, Trailing: 3, Avgqual: 15, SlidingWindow: 4:15, Minlength: 75. Reads passing the quality control checks were aligned to the GRCm39 mouse genome assembly¹⁶⁸ with GENCODE annotation set vM31¹⁶⁹ using STAR (2.7.10b)¹⁷⁰. Samtools (1.16.1)¹⁷¹ was used to sort alignment files by coordinate and index. Gene pseudocounts were obtained with htseq-count from HTSeq (0.11.1)¹⁷².

EdgeR (4.3.0)¹⁷³ was used to filter low expressing genes (filterByExpr), normalize by library size (calcNormExpr), and control for variance between genes (estimateDisp). Principal component analysis (PCA) was performed on scaled logCPM of genes for each sample (obtained using edgeR’s cpm function) with R’s prcomp function. Gene biotypes were pulled using biomaRt (2.61)¹⁷⁴ to interact with the ensembl database. Polysome-associated RNA-seq results were further compared to total RNA-seq results with the anota2seq R package (1.27.0)¹⁷⁵ using the following parameters: minSlopeTranslation = −1,maxSlopeTranslation = 2, minSlopeBuffering = −2, maxSlopeBuffering = 1, maxPAdj = 0.05). Sample by sample Euclidean distances were calculated using R’s dist() function and compared between assays with a Wilcoxon test. Plots were generated with either ggplot2 (3.5.1)¹⁷⁶, UpsetR (1.4.0)¹⁷⁷, gprofiler2 (0.2.3)¹⁷⁸, or ComplexHeatmap (2.18.0)^179,180.

We examined gene expression in certain gene ontology groups (genes associated with cytokine/cytokine receptor pathway, cytosolic ribosome, kegg glycolysis, kegg metabolism, and translation regulation) among the different deprivation conditions. Normalized expression (logCPM) values per gene were pulled for all samples. We examined the densities the different sample’s logCPMs across genes and calculated their area under the curve (AUC) for each gene ontology group. We then analyzed the calculated AUCs, representing gene expression in the different gene ontology groups, with a t-test comparing the different deprivations back to the control. P-values were adjusted using the FDR correction.

Genes associated with ATF4 binding sites from Örd et al., 2023 were used to identify motif enrichment.⁶⁴ Genes associated with peaks identified in at least 4 of the 7 cell lines were examined. The genes were translated into their homologs present in the Mus musculus genome using the biomaRt package in R^174,182. The upstream regulatory regions were defined from the −1000 to +100 bp of each transcription start site. These regulatory regions were then examined with SEA a tool from the MEME suite which does a simple enrichment analysis and will look for motifs in a set of sequences.^184,186 Mus musculus motifs defined in the HOCOMOCO v11 database were used as the reference motifs for SEA.¹⁸⁸ SEA returned a list of motifs from the HOCOMOCO that were enriched in our regulatory regions.

LC-MS metabolomics analysis

Metabolites were identified based on the exact mass +/− 5 ppm and retention time from authentic standards (Sigma Metabolomics Library). Peak integration was conducted using Xcalibur 4.2 (Thermo Fisher Scientific). Normalized metabolite abundances were generated by plotting the ratio of the detected metabolite peak area relative to its corresponding 13C-labeled internal standard or (if not present) the internal standard with the nearest retention time. Data was imported and plotted in R¹⁸¹ using the tidyverse family of packages¹⁸³ including ggplot2¹⁷⁶.

Proteomics analysis

Proteomics data processing and statistical analysis were conducted in R. The MS2 intensity values generated by Spectronaut were utilized for analyzing the entire proteome dataset. The data underwent log2 transformation and normalization by subtracting the median value for each sample. To ensure data integrity, we filtered it to retain only proteins with complete values in at least one cohort. To compare proteomics data across groups, we employed a Limma t-test to identify proteins with differential abundance, and we visualized the impact of these differences through volcano plots. Lists of differentially abundant proteins were generated based on criteria of adjusted P.Value <0.05, resulting in a prioritized list for subsequent bioinformatics analysis.

Statistical analysis

Statistical significance tests are noted in all figure legends for the relevant studies. We considered results with a p value less than 0.05 (p<0.05) to be statistically significant. *p < 0.05, **p < 0.01, ***p < 0.001, ****p < 0.0001. All error bars represent the standard error of the mean (SEM).

Supplementary Material

Supplemental information

Document S1. Figures S1–S7, Table S1

Table S2. Statistical analysis of U-13C glucose labeling data, related to Figure 6.

Table S3. Statistical analysis of U-13C glutamine labeling data, related to Figure 6.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Purified anti-mouse CD3ε (145-2C11)	Biolegend	CAT#100360; RRID:AB_2800555
Purified anti-mouse CD28 (37.51)	Biolegend	CAT#102122; RRID:AB_11147170
FITC CD8 alpha Monoclonal Antibody (53-6.7)	Invitrogen	CAT#MA1-10303; RRID: AB_11153636
PE CD98 Monoclonal Antibody (RL388)	Invitrogen	CAT#12-0981-81; RRID:AB_465792
PE-eFluor™ 610 Granzyme B Monoclonal Antibody (NGZB),	Invitrogen	CAT#61-8898-82; RRID: AB_2574670
APC IFN gamma Monoclonal Antibody (XMG1.2)	Invitrogen	CAT#17-7311-82; RRID: AB_469504
eFIuor™ 450 TNF alpha Monoclonal Antibody (MP6-XT22)	Invitrogen	CAT#48-7321-82; RRID:AB_1548825
Phospho-S6 Ribosomal Protein (Ser235/236) (D57.2.2E) XP® Rabbit monocloncal antibody	Cell Signaling Technology	CAT#4858
S6 Ribosomal Protein (5G10) Rabbit monocloncal antibody	Cell Signaling Technology	CAT#2217
ATF-4 (D4B8) Rabbit monocloncal antibody	Cell Signaling Technology	CAT#11815
Goat anti-Rabbit IgG (H+L) Cross-Adsorbed Secondary Antibody, HRP	Invitrogen	CAT#G-21234; RRID: AB_2536530
Goat anti-Mouse IgG (H+L) Secondary Antibody, HRP	Invitrogen	CAT#A16066; RRID:AB_2534739
hFAB™ Rhodamine Anti-Actin Primary Antibody hFAB™ Rhodamine Anti-Actin Primary Antibody	Bio-Rad	CAT#12004164
BUV395 CD45.2 Monoclonal Antibody (104)	Invitrogen	CAT#363-0454-82; RRID:AB_2925268
BUV737 CD69 Monoclonal Antibody (H1.2F3)	Invitrogen	CAT#367-0691-82; RRID:AB_2937116
BUV805 CD45.1 Monoclonal Antibody (A20)	Invitrogen	CAT#368-0453-82; RRID:AB_2896116
Pacific Blue Ly108 Monoclonal Antibody (330-AJ)	Invitrogen	CAT#134608; RRID:AB_2188093
BV650 CD8a Monoclonal Antibody (53-6.7)	Invitrogen	CAT#64-0081-82; RRID:AB_2662353
BV786 CD3e Monoclonal Antibody (145-2C11)	Invitrogen	CAT#78-0031-82; RRID:AB_2784894
PE TOX Monoclonal Antibody (TXRX10)	Invitrogen	CAT#12-6502-82; RRID:AB_10855034
Chemicals, peptides, and recombinant proteins
Recombinant mouse IL-2	BioLegend	#575406
4EGi-1	Cayman Chemical	CAT#15362
Torin 1	Cayman Chemical	CAT#10997
Halofuginone (hydrochloride)	Cayman Chemical	CAT#13370
GCN2iB	Cayman Chemical	CAT#35897
Recombinant Cas9-NLS	QB3-Berkeley MacroLab
Brefeldin A	Invivogen	CAT#inh-bfa
Fixable Viability Dye eFluor 780	Thermo Fisher Scientific	CAT#65-0865-18
ATTO-550-labeled tracrRNA	IDT	#1075928
Oligomycin	Sigma-Aldrich	CAT#O4876
BAM15	Cayman Chemical	CAT#A8674
Rotenone	Cayman Chemical	CAT#13995-1
2-deoxyglucose	Sigma-Aldrich	CAT#D0051
Collagenase D	Sigma-Aldrich	CAT#11088858001
DNAse I	Sigma-Aldrich	CAT#10104159001
SUPERAse In RNase Inhibitor	Invitrogen	CAT#AM2696
TRIzol™ LS Reagent	Invitrogen	CAT#10296028
Cell-Tak	Corning	CAT#354240
2-mercaptoethanol	Gibco	CAT#21985023
L-Glutamine	Gibco	CAT#25030081
Sodium pyruvate	Gibco	CAT#11360070
Critical commercial assays
EasySep™ Mouse CD8+ T Cell Isolation Kit	STEMCELL Technologies	CAT#19853
Buffer R	Thermo	CAT#MPK10096
Foxp3 / Transcription Factor Staining Buffer Set	Invitrogen	CAT#00-5523-00
Direct-zol-96 RNA Kits	Zymo Research	CAT#R2056
Seahorse XFe96 FluxPak	Agilent	Cat#103792-100
Zymo Quick-RNA MicroPrep Kit	Zymo Research	CAT#R1050
Thermo Maxima H Minus Reverse Transcriptase	Thermo Scientific	CAT#EP0751
PowerTrack™ SYBR Green Master Mix	Thermo Scientific	CAT#A46110
Deposited data
Nutrient stress RNA-seq and polysome profiling sequencing	This paper	GEO: GSE287987
ATF4- and CEBPg-deficient CD8 T cell RNA-seq	This paper	GEO: GSE287894
Experimental models: Cell lines
MC38	K. Cadwell - University of Pennsylvania, Philadelphia	N/A
gp33-expressing MC38	E.J. Wherry - University of Pennsylvania, Philadelphia	N/A
Experimental models: Organisms/strains
Mouse: C57BL/6J	The Jackson Laboratory	RRID:IMSR_JAX:000664
Mouse: Tg(TcraTcrb)1100Mjb, OT-I	The Jackson Laboratory	RRID:IMSR_JAX:003831
Mouse: B6.Cg-Tcratm1MomTg(TcrLCMV)327Sdz/TacMmjax, P14	The Jackson Laboratory	RRID:MMRRC_037394-JAX
B6.SJL-Ptprca Pepcb/BoyJ, CD45.1	The Jackson Laboratory	RRID:IMSR_JAX:002014
Oligonucleotides
oligo(dT)18	Thermo Scientific	CAT#SO132
Random hexamer	Thermo Scientific	CAT#SO142
crRNA: Mouse Rosa26 locus ACTCCAGTCTTTCTAGAAGA	IDT	N/A
crRNA: Mouse Atf4, AAGTTTAATAAAAGTCGACC	IDT	N/A
cRNA: Mouse Cebpg, CAGAGAGCGGAACAATATGG	IDT	N/A
crRNA: Mouse Eif2ak1 (HRI), CGTGGACGGCGAGCGCGACA	IDT	N/A
crRNA: Mouse Eif2ak4 (GCN2), AAAGCCCGGACATACTCCTC	IDT	N/A
Mouse Atf4 Forward 5’-AGAGTGCCTGCAGAAAGAGTCA-3’	IDT	N/A
Mouse Atf4 Reverse 5’-CGGTGCAGGTTGAGCATGTAT-3’	IDT	N/A
Mouse Ppp1r15a Forward 5’-TCCTCTAAAAGCTCGGAAGGTACAC-3’	IDT	N/A
Mouse Ppp1r15a Reverse 5’-TCTCGTGCAAACTGCTCCCA-3’	IDT	N/A
Mouse Sesn2 Forward 5’-ACCTTCCGTGCCCAGGATTAT-3’	IDT	N/A
Mouse Sesn2 Reverse 5’-CCTGGAACTTCTCATCCAGCAG-3’	IDT	N/A
Mouse Nupr1 Forward 5’-CTGGCGGGCATGAGAGGAAG-3’	IDT	N/A
Mouse Nupr1 Reverse 5’-TCTGTGGTCTGGCCTTATCTCCA-3’	IDT	N/A
Mouse Slc3a2 Forward 5’-CACTGGGGAGCGTACTGAATCC-3’	IDT	N/A
Mouse Slc3a2 Reverse 5’-AGTCCTGCTTGCGACACACTC-3’	IDT	N/A
Mouse Slc7a5 Forward 5’-AGCTGTGGCTGTGGACTTCG-3’	IDT	N/A
Mouse Slc7a5 Reverse 5’-CCCATTGACAGAGCCGAAGCA-3’	IDT	N/A
Mouse Slc7a11 Forward 5’-CCAAGGGCATACTCCAGAACA-3’	IDT	N/A
Mouse Slc7a11 Reverse 5’-TAGGACAGGGCTCCAAAAAGT-3’	IDT	N/A
Software and algorithms
GraphPad Prism	GraphPad Software	https://www.graphpad.com
FlowJo v10	Tree Star	https://www.flowjo.com/
Seahorse Wave Desktop Software	Agilent technologies	https://www.agilent.com
Software and algorithms

Highlights.

• T cells functionally adapt to nutrient stress through global nutrient reallocation

• Acute stress translationally prioritizes adaptive programming over growth programming

• ATF4/CEBPG rewires amino acid handling and maintain anaplerosis under stress

• The integrated stress response prevents T cell dysfunction during anti-tumor immunity