Ketogenesis-generated β-hydroxybutyrate is an epigenetic regulator of CD8+ T-cell memory development

Glycogen has long been considered to have a function in energy metabolism. However, our recent study indicated that glyco- gen metabolism, directed by cytosolic phosphoenolpyruvate carboxykinase Pck1, controls the formation and maintenance of CD8+ memory T (Tmem) cells by regulating redox homeo- stasis1. This unusual metabolic program raises the question of how Pck1 is upregulated in CD8+ Tmem cells. Here, we show that mitochondrial acetyl coenzyme A is diverted to the keto- genesis pathway, which indirectly regulates Pck1 expression. Mechanistically, ketogenesis-derived β-hydroxybutyrate is present in CD8+ Tmem cells; β-hydroxybutyrate epigenetically modifies Lys 9 of histone H3 (H3K9) of Foxo1 and Ppargc1a (which encodes PGC-1α) with β-hydroxybutyrylation, upregu- lating the expression of these genes. As a result, FoxO1 and PGC-1α cooperatively upregulate Pck1 expression, therefore directing the carbon flow along the gluconeogenic pathway to glycogen and the pentose phosphate pathway. These results reveal that ketogenesis acts as an unusual metabolic pathway in CD8+ Tmem cells, linking epigenetic modification required for memory development. The development of long-lived CD8+ memory T (Tmem) cells requires the use of fatty acids for energy production through oxida- tive phosphorylation2–5. Fatty acid breakdown occurs in the mito- chondria where the fatty acids are converted to acetyl coenzyme A (acetyl-CoA), which is further oxidized within the tricarboxylic acid (TCA) cycle. In addition to producing ATP for energy supply, the conversion of acetyl-CoA by the mitochondrial β-oxidation sys- tem also yields reduced flavin adenine dinucleotide (FADH2) and reduced nicotinamide adenine dinucleotide (NADH), which can generate cellular reactive oxygen species (ROS) through reverse electron transport in the electron transport chain6.

Preventing excessive ROS production or enhancing detoxification therefore represents an alternative mechanism that regulates the metabolic fitness and survival of the CD8+ Tmem-cell population. Indeed, we recently reported that an active gluconeogenesis–glycogenolysis cycle exists in CD8+ Tmem cells1. In contrast to fatty acid oxidation (FAO), which generates energy, the gluconeogenesis–glycogenolysis cycle in CD8+ Tmem cells provides antioxidant defence by generating NADPH and reduced glutathione1. To retain high gluconeogenic activity, cytosolic phosphoenolpyruvate carboxykinase (Pck1), an important rate-limiting enzyme in gluconeogenesis, is upregulated substantially in CD8+ Tmem cells; Pck1 catalyses oxaloacetate (OAA) to facilitate the generation of 6-phosphoglucose1. Normally, mito- chondrial acetyl-CoA is condensed by OAA to citrate; however, a low level of TCA cycle activity (such as a low amount of OAA) results in the diversion of acetyl-CoA to the ketogenesis pathway, through which acetyl-CoA condenses into acetoacetyl-CoA, giv- ing rise to the formation of acetoacetate (AcAc), β-hydroxybutyrate (BHB) and acetone7,8. Although FAO can be the source that provides acetyl-CoA for ketogenesis, the role of FAO in T-cell memory devel- opment is inconsistent9,10. Here we hypothesize that overconsump- tion of OAA by Pck1-triggered gluconeogenesis may lower OAA levels and impair the recycling of acetyl-CoA, leading to activation of ketogenesis in CD8+ Tmem cells.
To determine whether ketogenesis occurs in CD8+ Tmem cells, C57BL/6 (CD45.2+) mice were adoptively transferred with 1 × 105 CD45.1+ OT-I T cells, and then infected with 5 × 105 colony-form- ing units (CFUs) of Listeria monocytogenes that express ovalbumin (Lm-OVA).

OVA-specific CD8+ effector T cells (Teff) and Tmem cells were sorted after staining for CD45.1 and CD8 on day 6 and 30 after infection, respectively. We found much higher levels of AcAc and BHB, but not acetone, in CD8+ Tmem cells compared with naive T cells (Tn) or Teff cells, analysed using mass spectroscopy with liq- uid chromatography (LC–MS; Fig. 1a) and confirmed by fluoro- metric assay (Fig. 1b). Moreover, CD4+ Tmem cells that were isolated from the mice that were adoptively transferred with OT-II T-cells and immunized against OVA also showed higher levels of BHB compared with Tn and Teff cells from these mice (Extended Data Fig. 1a). We next differentiated OT-I cells into Tmem or Teff cells in vitro on the basis of a protocol that we described previously1. Consistent with the above results, IL-15-induced Tmem cells expressed significantly higher levels of BHB and AcAc than IL-2- primed Teff cells (Fig. 1c). In this study we also analysed human CD8+ Tn and Tmem cells isolated from donor peripheral blood. Some isolated Tn cells were stimulated into Teff cells using anti-CD3/ CD28 microbeads. Similar to the mouse Tmem cells, higher BHB levels in human CD8+ Tmem cells were found compared with Tn and Teff cells (Extended Data Fig. 1b).

These results suggest a possible link between CD8+ Tmem cells and ketogenesis. In fact, enzymes involved in ketogenesis, such as mitochondrial acetyl-CoA acet- yltransferase 1 (Acat1), mitochondrial 3-hydroxy-3-methylgluta- ryl-CoA synthase 2 (Hmgcs2), 3-hydroxy-3-methylglutaryl-CoA lyase (Hmgcl) and 3-hydroxybutyrate dehydrogenase 1 (Bdh1), were upregulated in the mouse and human CD8+ Tmem cells described above but not in Teff or Tn cells (Fig. 1d–f, Extended Data Fig. 1c,d), suggesting that CD8+ Tmem cells might have active ketogenesis. Furthermore, upregulation of these enzymes was also found in mouse CD4+ Tmem cells (Extended Data Fig. 1e,f). As acetyl-CoA is the source for generating AcAc and BHB, we also performed 13C-tracing assays to clarify the carbon source. Using 13C-glucose, we found that IL-15-induced CD8+ Tmem cells had a lower abundance of m+2-labelled BHB and m+2 AcAc compared with Tn or Teff cells (Extended Data Fig. 1g). However, by provid- ing 13C-pyruvate to directly generate acetyl-CoA, we detected a greater abundance of m+2 BHB and m+2 AcAc in CD8+ Tmem cells (Extended Data Fig. 1h), suggesting that acetyl-CoA might not be channelled from glycolysis in CD8+ Tmem cells. By contrast, in cells treated with 13C-palmitate, a greater abundance of m+2 BHB and m+2 AcAc was detected in CD8+ Tmem cells compared with Tn orTeff cells (Extended Data Fig. 1i), suggesting that fatty acid oxida- tion might provide the acetyl-CoA for ketogenesis.We next investigated whether and how ketogenic metabolite(s) affect CD8+ Tmem cells. Serum BHB concentrations can increase to 2 mM after 24 h fasting11. We therefore added a physiologically relevant concentration of AcAc and BHB (2–10 mM) to the cul- ture.

Treatment with BHB, but not AcAc, increased the number of IL-15-induced CD8+ T cells that exhibited a typical central memory phenotype (Fig. 2a, Extended Data Fig. 2a) and upregulated the expression of memory-related transcription factors Tcf7, Lef1 and Bcl6 (refs. 12,13; Fig. 2b, Extended Data Fig. 2b), suggesting that the ketogenic metabolite BHB is involved in the memory development of CD8+ T cells. In line with this result, only a slight increase in the level of BHB was found in the AcAc-treated Tmem cells (Extended Data Fig. 2c). To confirm this result in vivo, BHB was intraperitone- ally (i.p.) injected into the mice 7 d after adoptive transfer with OT-I cells and Lm-OVA infection. A systemic increase of CD8+ Tmem cells, determined by both phenotypic markers (Fig. 2c) and transcription factors (Fig. 2d), was observed in multiple compartments on day 30 compared with control groups that were not treated with BHB. Here we also confirmed that Tmem cells generated in BHB treated mice were functional. By rechallenging the mice with 5 × 105 CFUs Lm-OVA, we found that OT-I Tmem cells responded strongly, as evidenced bythe upregulation of CD25, CD69 and interferon-γ (Extended Data Fig. 2d,e). Notably, a further increase in the number of OT-I T cells was observed in the BHB-treated mice (Extended Data Fig. 2f). In addition to BHB treatment, we also treated the mice with a carbohy- drate-free ketogenic diet (KD), a 15% carbohydrate non-ketogenic diet with high-fat (HF) or a normal diet beginning on day 7 after adoptive transfer of OT-I cells and Lm-OVA infection. Mice in the KD group displayed the highest levels of BHB and the increased number of CD8+ Tmem cells, whereas mice in the HF group had the same blood BHB levels and CD8+ Tmem cell numbers as the mice in the normal diet group (Extended Data Fig. 2g,h).

In line with this ketogenesis, we found that BHB levels were increased in the liver of mice in the KD group (1.44 nmol mg−1 protein) compared with those in the HF or control diet groups (0.056 and 0.053 nmol mg−1 protein, respectively; Extended Data Fig. 2i). Notably, the BHB level in the IL-15-induced CD8+ Tmem cells was 0.1 nmol mg−1 pro- tein (Fig. 1a,b). By contrast, transducing OT-I T cells with Bdh1- targeting short hairpin RNA (shBdh1) to knock down the transition from AcAc to BHB led to a reduction of BHB and an accumula- tion of AcAc (Fig. 2e, Extended Data Fig. 2j–l). Consistent with this, Bdh1 knockdown decreased the m+2 BHB abundance of CD8+ Tmem cells when cultured in the presence of 13C-palmitate (Extended Data Fig. 2m). Bdh1 knockdown impaired the formation of IL-15- polarized Tmem cells in vitro or Lm-OVA-induced CD8+ Tmem cells in the spleen, liver and lung (Fig. 2f,g, Extended Data Fig. 2n,o). However, such impaired Tmem-cell formation could be rescued by the provision of BHB but not AcAc (Fig. 2f). This reduction in memory formation caused by Bdh1 knockdown might be due to the impaired cell survival, as Bdh1 knockdown resulted in an increase of annexin V+CD8+ Tmem cells, which had no effect on the frequency of Ki-67+ or BrdU+ proliferating cells (Extended Data Fig. 2p). In line with this, neither provision of BHB nor Bdh1 knockdown affected in vitro polarization of Teff cells by IL-2 or Lm-OVA-primed Teff-cell responses in vivo (Fig. 2h,i, Extended Data Fig. 2q–t), suggesting that the ketogenic metabolite BHB has an important role in regu- lating CD8+ Tmem-cell formation. Here we further translated this regulation mechanism of CD8+ Tmem cells into a tumour immuno- therapy model. We found that i.p. injection of BHB enhanced OT-I T-cell-mediated destruction of OVA-B16 melanoma and pro- longed mouse survival (Fig. 2j); by contrast, mice that were adop- tively transferred with shBdh1 OT-I T cells exhibited unsubdued tumour growth and shortened survival, which could be rescued by treatment with BHB (Fig. 2k).

Compared with Bdh1, Acat1 is an upstream ketogenesis-regulating enzyme. We found that Acat1 knockdown also blunted BHB levels (Extended Data Fig. 3a–c), and impaired the memory formation of OT-I T cells but had no effect on Teff-cell responses (Extended Data Fig. 3d–g).BHB is usually converted back into acetyl-CoA and then enters the TCA cycle. When we used 13C-BHB to treat CD8+ Tn, Teff or Tmem cells, we found that 13C could be incorporated into the TCA cycle intermediates such as citrate, fumarate, α-KG and malate. A slightly higher ratio of 13C-α-KG, -fumarate and -malate was found in Tmem cells compared with Teff cells (Extended Data Fig. 3h–k), indicating that BHB can be used as energy source for Tmem-cell metabolism. However, when we measured the oxygen consumption rate (OCR) using a Seahorse XF Analyzer, we found that the OCR in BHB- treated CD8+ Tn, Teff or Tmem cells was not altered (Extended Data Fig. 3l–n). Furthermore, we found that knockdown of Bdh1 did not affect the OCR of CD8+ Tmem cells (Extended Data Fig. 3o), suggest- ing that the major role of BHB in CD8+ Tmem cells does not include function as energy molecule.Notably, BHB treatment resulted in the upregulation of Pck1 (which is a key enzyme that triggers gluconeogenesis) in IL-15- induced or Lm-OVA-induced CD8+ Tmem cells (Fig. 3a,b); this upreg- ulation of Pck1 could be inhibited by transduction with shBdh1 or shAcat1 (Fig. 3c, Extended Data Fig. 3p), indicating a connectionbetween BHB production and Pck1 expression. Our previous study showed that Pck1 activates the gluconeogenesis–glycogenolysis cycle to fuel flux into the pentose phosphate pathway, leading to enhanced generation of NADPH, which protects against ROS to support the long-term survival of CD8+ Tmem cells1. Indeed, Bdh1 knockdown decreased the levels of glycogen, NADPH/NADP+ and GSH/GSSG but increased ROS in CD8+ Tmem cells; however, such a decrease could be rescued by provision with BHB (Fig. 3d–g). Moreover, the 13C-glucose tracing showed substantial content of m+2 PEP, m+2 G1P/G6P and m+2 R5P in CD8+ Tmem cells; how- ever, knockdown of Bdh1 considerably decreased the contents of these compounds (Fig. 3h).

Consistent with this, these decreases were rescued by the addition of BHB (Fig. 3h). Furthermore, we adoptively transferred wild-type or Pck1+/− OT-I T cells into CD45.1 mice and then vaccinated the mice with Lm-OVA, followed by treat- ment with BHB 7 d later. We found that BHB treatment was unable to rescue the impairment of OT-I memory formation by Pck1 hap- lodeficiency (Fig. 3i–k). Similarly, OT-I T cells treated with 3-mer- captopicolinic acid (3-MPA) or haplodeficient for Pck1 failed to differentiate into Tmem cells in the presence of IL-15 in vitro, and such impairment could not be corrected by the addition of BHB (Fig. 3l,m). Together, these results suggest that BHB facilitates CD8+ Tmem-cell formation by upregulating Pck1 expression.We next investigated the molecular mechanism by which BHB regulated Pck1 expression. BHB is not only a ketogenic metabolite for energy supply but also acts as an epigenetic modifier, including H3K9 acetylation14–17. However, BHB treatment did not increase H3K9 acetylation in CD8+ Tmem cells (Extended Data Fig. 4a), excluding the possibility that BHB functions as an endogenous histone deacetylase inhibitor to modulate H3K9 acetylation16,17. Consistent with this, the use of either HDAC inhibitor or acetate had no effect on Pck1 expression (Extended Data Fig. 4b,c), further suggesting that BHB does not induce Pck1 expression by acetylat- ing H3K9. Both western blot and chromatin immunoprecipitation coupled with quantitative PCR (ChIP–qPCR) analyses showed that treatment with BHB had no effect on H3K4me3 or H3K27me3 of CD8+ Tmem cells (Extended Data Fig. 4d–f).

Furthermore, we found that the Pck1 promoter was highly methylated, but not as a result of modification by BHB (Extended Data Fig. 4g). Besides acetylation and methylation, BHB can also modify H3K9 by β-hydroxybutyrylation (Kbhb)18,19 and, in this case, BHB acts as the β-hydroxybutyryl donor, and an intermediate β-hydroxybutyryl- CoA is synthesized first, which can subsequently be used by p300 as a cofactor to generate lysine β-hydroxybutyrylation20. Notably, higher levels of β-hydroxybutyryl-CoA were found in CD8+ Tmem cells compared with in Teff or Tn cells (Fig. 4a), implying possible β-hydroxybutyrylation of H3K9. Moreover, the western blot analy- sis showed an increased modification of lysine residues of proteins (pan-Kbhb) as well as histone lysine residues (H3K9bhb) in CD8+ Tmem cells (Fig. 4b). Similar results were also obtained in human CD8+ Tmem or mouse CD4+ Tmem cells (Extended Data Fig. 4h). Moreover, treatment with BHB further enhanced pan-Kbhb and H3K9bhb modifications, which was confirmed by the addition of shBdh1 or shAcat1 (Fig. 4c,d, Extended Data Fig. 4i), suggest- ing that BHB is the source of β-hydroxybutyrylation in CD8+ Tmem cells. However, we found that AcAc treatment did not enhance H3K9bhb or Pck1 expression in CD8+ Tmem cells (Extended Data Fig. 4j).

Moreover, the treatment of CD8+ Tmem cells with 10 mM BHB or isotopically labelled BHB [U-13C4] further confirmed the direct contribution of BHB to H3K9bhb modification. Ten Kbhb modification sites in histone H3 with a mass shift of 86.0386 Da were identified using LC–MS/MS analysis (Supplementary Table 1–3), and 13C-labelled H3K9bhb was also identified (Fig. 4e). We also found that incorporation of 13C-β-hydroxybutyryl groups into H3K9 was increased in CD8+ Tmem cells compared with CD8+ Tn and Teff cells (Extended Data Fig. 4k) SR-18292 to glycogen and pentose phosphate pathway. These findings pro- vide insights into how CD8+ Tmem cells balance energy production and redox homeostasis, potentiating approaches to regulate T-cell memory with clinical applications.