Abstract

   The rapid recall ability is a hallmark of memory CD8^+ T cells, but the
   underlying mechanisms remain incompletely understood. Here we find that
   histone variant H2A.Z is expressed at higher levels in memory CD8^+ T
   cells than in naïve cells. Furthermore, in memory CD8^+ T cells H2A.Z
   is deposited at the promoters and enhancers, particularly super
   enhancers, of those genes involved in recall responses, while H2A.Z
   deficiency in memory CD8^+ T cells inhibits recall responses in vitro
   and in vivo. Mechanistically, multi-omics analyses show that H2A.Z
   maintains a poised epigenetic landscape on those recall response genes
   to potentiate a rapid transcription activation. Accordingly, H2A.Z
   deposition on these genes is induced by TCR/CD28 signals, and is
   cooperated by IL-7/IL-15 signals. Together, our data suggest that H2A.Z
   may orchestrate a specific epigenetic landscape during memory T cell
   differentiation to facilitate a rapid recall response.

   Subject terms: T cells, Immunological memory, Epigenetics in immune
   cells, Tumour immunology
     __________________________________________________________________

   The ability to pursue a stronger recall response on secondary antigen
   encounter is a hallmark for T cell immunity. Here the authors show that
   the histone component, H2A.Z, may contribute to CD8 T cell recall
   readiness by depositing to specific locations and introducing a
   specific epigenetic landscape in CD8 memory T cells to facilitate more
   prompted activation of related genes.

Introduction

   Memory CD8^+ T cells are crucial for defending the body against acute
   and chronic infections, as well as cancers^[44]1–[45]3. Inducing
   functional memory CD8^+ T cells is also essential for effective vaccine
   immunity^[46]4,[47]5. These poised cells persist long-term and rapidly
   respond upon antigen re-exposure^[48]6. Compared to naïve T cells,
   memory T cells exhibit faster, more robust activation and effector
   function. This unique capacity enables efficient control of recurrent
   pathogens and cancers. A key question is how the rapid recall ability
   of memory T cells is established and regulated.

   The molecular basis of the rapid recall response in memory T cells
   remains poorly understood. Early studies revealed differences at the
   protein level, particularly within T cell activation signaling
   pathways. While initial CD3 signaling events, as measured by CD3
   complex and ZAP70 phosphorylation, are indistinguishable between memory
   and naïve CD8^+ T cells^[49]7, memory CD8^+ T cells exhibit increased
   phosphorylation of LAT and ERK upon antigen stimulation. Another study
   reported quicker and enhanced nuclear translocation of NF-κB p50 in
   memory CD4^+ T cells compared to naïve cells following TCR
   stimulation^[50]8. Critically, however, these protein-level changes are
   transient modifications; they are not inherently stable and cannot be
   inherited through cell divisions during memory T cell homeostasis.

   Epigenetic modifications represent heritable changes in chromatin
   states without altering the DNA sequence. Such modifications frequently
   accompany T cell activation and differentiation^[51]9. Recent studies
   implicate epigenetic programming as critical for the rapid recall
   response of memory T cells^[52]10. A genome-wide analysis of H3K4me3
   and H3K27me3 in memory versus naïve CD8^+ T cells revealed
   significantly higher H3K4me3 deposition at poised gene loci in memory
   cells^[53]11. Bivalent chromatin domains, marked by both H3K4me3 and
   H3K27me3, were also enriched at silent gene loci in resting memory
   CD8^+ T cells; these genes were rapidly induced upon activation.
   Similarly, resting memory CD4^+ T cells exhibit an activating chromatin
   state at rapidly inducible genes (e.g., IFNG, IL4 and IL13) associated
   with recall responses^[54]12. However, the mechanisms establishing this
   poised epigenetic landscape remain unclear.

   H2A.Z, a highly conserved histone H2A variant, is incorporated into
   chromatin by ATP-dependent remodeling complexes in a
   replication-independent manner, influencing chromatin architecture and
   gene regulation^[55]13,[56]14. Recent studies demonstrate that H2A.Z
   deposition in gene promoter regions regulates nucleosome stability and
   increases chromatin accessibility. By weakening histone-histone
   interactions within the nucleosome^[57]14–[58]16, H2A.Z establishes a
   chromatin environment permissive for transcription factor and RNA
   polymerase II recruitment^[59]17,[60]18, enhancer RNA transcription,
   enhancer-promoter interactions^[61]19, and ultimately modulates
   transcriptional activity. Additionally, H2A.Z influences histone
   modifications by recruiting complexes like MLL and PRC2 to promoters
   and enhancers, thereby regulating H3K4me3 and H3K27me3
   methylation^[62]18,[63]20. While extensively studied in yeast, plants,
   ESCs, and other model systems^[64]14,[65]21,[66]22, the role of H2A.Z
   in T cells remains largely unexplored. Notably, in memory CD4^+ T
   cells, H2A.Z deposition occurs at IFNG, IL4, and IL13 gene loci,
   correlating positively with other active chromatin marks^[67]12. A key
   outstanding question, therefore, is how H2A.Z contributes to the rapid
   recall response in memory T cells.

   Here, transcriptome analysis reveals elevated expression of H2A.Z in
   memory CD8^+ T cells compared to naïve cells. Further investigation
   shows that H2A.Z is selectively enriched at the promoter and enhancer
   regions of key effector genes in resting memory T cells, establishing a
   poised chromatin state conducive to rapid transcriptional activation.
   This deposition is cooperatively regulated by TCR/CD28 and cytokine
   (IL-7/IL-15) signaling. Collectively, these findings uncover an
   H2A.Z-mediated epigenetic licensing mechanism in memory CD8^+ T cells
   that enables their rapid recall response. Targeting this epigenetic
   pathway may enhance the efficacy of vaccines and T cell-based
   immunotherapies.

Results

Identification of activation-recalled genes in memory CD8^+ T cells

   Upon re-stimulation ex vivo, memory CD8^+ T cells rapidly activate
   their cytotoxic machinery^[68]23. To validate this rapid recall
   response in our study, the classical Listeria
   monocytogenes-OVA (LM-OVA) infection-induced OT-I memory T cell model
   was adapted^[69]24. Memory OT-I cells were identified and isolated from
   the LM-OVA infected mice at least 30 days later, as described
   previously^[70]24. Naïve OT-I cells were isolated from naïve OT-I
   transgenic mice. Equal number of memory and naïve OT-I cells were mixed
   and adoptively transferred (ATF) into recipient mice followed by LM-OVA
   rechallenge (Fig. [71]1a). The expansion of memory OT-I T cells is
   significantly higher than naïve OT-I T cells at day 4 or day 7
   post-infection (Fig. [72]1b, c). To directly test the TCR signal per se
   on memory T cell recall and avoid the influence of other potential
   signals in vivo, resting memory and naïve OT-I T cells were isolated
   (Fig. [73]1d and Supplementary Fig. [74]1a, b) and stimulated with
   OVA[257-264] peptide in vitro. Significantly higher amount of IFN-γ was
   produced in memory OT-I T cells compared to naïve OT-I T cells
   (Fig. [75]1e). To more systemically evaluate the recall response of
   memory OT-I T cells, naïve and memory OT-I T cells were sorted and
   stimulated with anti-CD3 and anti-CD28 antibodies (Abs) for 6 or
   24 hours (h), or left unstimulated for RNA-seq analysis (Fig. [76]1f
   and Supplementary Fig.[77]1a, b). Principal-component analysis showed
   that stimulation caused potent transcriptomic change from the resting
   state in both memory and naïve OT-I cells (Fig. [78]1g). To
   characterize memory T cell recall response, we captured the genes that
   responded more robustly (at least 2-fold) at either 6 or 24 h in memory
   OT-I cells than in naïve OT-I cells, and defined these genes as memory
   T cell activation-recalled genes (ARGs). 80 ARG genes were up-regulated
   in memory OT-I cells compared to naïve OT-I cells (Supplementary
   Table [79]1). GO/KEGG analysis showed that these ARGs were enriched in
   functions such as “cytokine-cytokine receptor interaction”, “regulation
   of MAPK cascade”, “inflammatory response” (Ifng, Ccl3 and Xcl1) and
   “immune effector process” (Gzmb and Tnf) (Fig. [80]1h, i). It is worth
   noting that memory T (IL-7R^high KLRG1^low) cells include both central
   memory T (T[CM]) and effector memory T (T[EM]) cell populations
   (Supplementary Fig. [81]1c). To examine whether T[CM] and T[EM] may
   have different recall response, we isolate T[CM] and T[EM] separately
   and examine their recall response of the ARGs we previously defined
   using total memory T cells. The data show that both T[CM] and T[EM]
   have largely similar changes on these ARG induction compared to naïve
   cells (Supplementary Fig. [82]1d–f), suggesting a common phenotype of
   both T[CM] and T[EM] on the recall activation. Thus, these results
   validated the rapid recall response of memory CD8^+ T cells and defined
   a group of signature genes characterizing the recall response.

Fig. 1. Memory CD8^+ T cells respond rapidly and robustly than naïve CD8^+ T
cells.

   [83]Fig. 1
   [84]Open in a new tab

   a Experiment setup. CD45 congenically distinct naïve (CD62L^hiCD44^lo)
   OT-I cells (T[N]) and memory (CD127^hiKLRG1^lo) OT-I cells (T[M]) were
   sorted from spleens, mixed at a 1:1 ratio, and subsequently
   co-transferred into recipients, followed by infection with LM-OVA. b, c
   The frequencies of donor cells in the spleen of recipient mice were
   determined at day 4 (n = 4 mice) (b) and day 7 (n = 5 mice) (c)
   post-infection. d Experiment setup. Naïve (CD62L^hiCD44^lo) OT-I cells
   and memory (CD127^hiKLRG1^lo) OT-I cells were isolated from spleens for
   in vitro stimulation. e Resting memory and naïve OT-I cells were
   stimulated with OVA[257-264] peptide, IFN-γ production was determined
   by flow cytometry (n = 3 biologically independent samples). Values
   denote geometric mean fluorescence intensity (gMFI). f RNA-seq were
   performed as described in the schematic diagram (n = 2 biologically
   independent samples). g Principal components analysis (PCA). h
   Metascape analysis showing the Gene Ontology terms and KEGG pathways
   enriched in the list of ARG genes. i, Heatmap of selected genes from
   ARGs. Data are representative of two (b, c and e–i) independent
   experiments. Data are mean ± SD and the P value of two-tailed paired
   Student’s t test (b and c). ATF, adoptively transferred. Source data
   are provided as a [85]Source Data file.

H2A.Z regulates memory CD8^+ T cell recall response

   To explore the epigenetic mechanism of memory T cell recall response,
   we selectively compared the expression of epigenetics-related genes
   between resting naïve OT-I T cells and LM-OVA infection-induced resting
   memory OT-I T cells. We plotted the log2 ratio of gene abundance
   between these two types of cells, and ranked all genes in the data set
   by increasing fold-change. It was observed that H2afz, encoding histone
   variant H2A.Z.1, is one of the top ranked genes highly expressed in
   resting memory OT-I cells (Fig. [86]2a), while H2afv (encoding histone
   variant H2A.Z.2) is not. Similar results were found using data from a
   previous study (Fig. [87]2b)^[88]25. Since H2A.Z.1 and H2A.Z.2 are
   highly conserved with only three amino acids difference, in the latter
   of this study we detected and manipulated total H2A.Z. At the protein
   level, significantly higher level of H2A.Z expression was found in
   memory OT-I T cells compared to naïve or effector OT-I T cells
   (Fig. [89]2c, Supplementary Fig. [90]2a), while no significant
   difference was found between naïve and effector cells (Supplementary
   Fig. [91]2a). Whether the increased expression of H2A.Z in memory T
   cells plays any important roles remains unknown. Recent studies have
   shown that H2A.Z is associated with “poised” genes^[92]14. We
   hypothesized that H2A.Z may regulate the recall response of “poised”
   memory T cells. To directly test this, we constructed a conditional
   knockout (cKO) mouse model, GzmB-Cre^+H2A.Z^f/f OT-I mice
   (Supplementary Fig. [93]2b), in which H2A.Z is specifically deleted in
   CD8^+ T cells upon activation^[94]26. This model bypasses the potential
   influence of H2A.Z on T cell development and early effector response,
   therefore allowing us to study the role of H2A.Z in late effector
   response and memory formation. CD45.2^+ congenically marked naïve cKO
   OT-I or wild type (WT) OT-I cells were isolated and adoptively
   transferred separately into WT CD45.1 recipients, followed by LM-OVA
   infection. GzmB-Cre effectively deleted H2A.Z in effector OT-I T cells
   at day 7 (Supplementary Fig. [95]2c, d) and day 30 post-infection
   (Supplementary Fig. [96]2e). In addition, consistent with the purpose
   of this mouse model, GzmB-Cre-induced H2A.Z deletion does not affect
   OT-I T cell early activation and proliferation during the in vitro
   assay (Supplementary Fig. [97]3a–c). To explore whether H2A.Z
   influences naïve T cell primary response in vivo, CSFE-labeled naive WT
   and H2A.Z cKO OT-I T cells were transferred into WT mice, which were
   then infected with LM-OVA on the next day. At day 3 and 7
   post-infection, the adoptively transferred OT-I T cells were analyzed.
   The results show that H2A.Z deficiency impairs OT-I cells proliferation
   (Supplementary Fig. [98]3d, e) and effector T cell survival
   (Supplementary Fig. [99]3f, g) in vivo.

Fig. 2. Conditional deletion of H2A.Z reduces Ag-specific memory CD8^+ T cell
recall responses.

   [100]Fig. 2
   [101]Open in a new tab

   a, b The log2 ratio of gene abundance between the memory and naïve OT-I
   cells, line plot with genes ranked according to increasing log2 ratios.
   Based on data from Fig. [102]1F (a) and GEO dataset [103]GSE15907 (b).
   c H2A.Z expression was analyzed in naïve (CD62L^hiCD44^lo) and memory
   (CD127^hiKLRG1^lo) OT-I cells (n = 5 mice per group). Values denote
   geometric mean fluorescence intensity (gMFI). d Experiment setup. CD45
   congenically distinct naïve OT-I cells from wild-type (WT) and GzmB-Cre
   H2A.Z^f/f (cKO) OT-I mice were mixed at a 1:1 ratio and co-transferred
   into recipients. On the next day, these mice were infected with LM-OVA.
   On day 30-45, WT and cKO memory OT-I cells were sorted and re-mixed at
   a 1:1 ratio (1.5 × 10^4 mixed cells/mice) and co-transferred into
   congenically marked WT recipients and followed by LM-OVA infection. e
   At day 7 post-infection, the frequencies of donor cells in the spleen
   of recipient mice were determined (n = 10 mice). f Experiment setup.
   CD45 congenically distinct naïve OT-I cells from WT and Rosa26 cre-ERT2
   H2A.Z^f/f (icKO) OT-I mice were mixed at a 1:1 ratio and transferred
   into recipients. On the next day, these mice were infected with LM-OVA.
   After TAM injection, WT and icKO memory OT-I cells were sorted and
   re-transferred into congenically marked WT recipients and followed by
   LM-OVA infection. g, h On day 7 post-infection, the frequencies of
   donor cells in the blood (g) and spleen (h) of recipient mice were
   determined (n = 10 mice). i, j Splenocytes isolated from recipient mice
   were stimulated with the OVA[257-264] peptide for 6 h, then cytokine
   production from OT-I cells was measured. Values denote gMFI (n = 9
   biologically independent samples). Data are representative of two
   independent experiments in (c), three independent experiments in (e and
   g–j). Data are mean ± SD. Statistical analysis was performed using
   two-tailed unpaired Student’s t test (c) or two-tailed paired Student’s
   t test (e and g–j). Source data are provided as a [104]Source Data
   file.

   To test whether H2A.Z influence memory T cell generation, WT and cKO
   naïve OT-I T cells were mixed transferred into WT host. Donor WT, donor
   cKO and host were congenically distinct. Mice were infected with LM-OVA
   and memory OT-I T cells were detected 30 days later. H2A.Z deficiency
   resulted in significantly reduced memory OT-I T cells (Supplementary
   Fig. [105]4a), but CD127 (IL-7R) expression was not significantly
   altered (Supplementary Fig. [106]4b, c), suggesting an important role
   of H2A.Z for memory T cell generation or maintenance. This is also
   consistent with our previous finding that H2A.Z is required for naïve T
   cell homeostasis proliferation^[107]27. To further elucidate the role
   of H2A.Z in memory T cell formation at early time, we employed an OT-I
   T cell co-transfer model and found that H2A.Z is required for OT-I MPEC
   differentiation at day 7 upon LM-OVA infection (Supplementary
   Fig. [108]4d, e) and the expression of CD127 is also reduced in H2A.Z
   deficient T cells (Supplementary Fig. [109]4f). To further investigate
   whether H2A.Z also regulates the maintenance of memory T cell, we have
   also analyzed the viability of memory T cells by flow cytometry at day
   30 post-infection. The results showed that H2A.Z deletion did not
   affect the survival of memory T cells (Supplementary Fig. [110]4g).
   Thus, at least during the observed period, H2A.Z might play minor role
   for the maintenance of memory T cells. However, whether the long-term
   maintenance needs H2A.Z requires further study.

   To specifically test the role of H2A.Z in memory T cell recall
   response, LM-OVA infection-induced memory OT-I T cells of WT and cKO
   genotypes were re-isolated and adoptively transferred with equal number
   into a new set of recipient mice, followed by LM-OVA rechallenge
   (Fig. [111]2d). At day 7, significantly impaired recall response was
   found in cKO OT-I memory T cells compared to that in WT OT-I memory T
   cells (Fig. [112]2e).

   To further specify the role of H2A.Z in memory T cells, a
   tamoxifen-induced gene deletion model, Rosa26-creERT2, was
   used^[113]28. H2A.Z deletion was induced in Rosa26-creERT2^+H2A.Z^f/f
   (icKO) OT-I T cells when memory has been fully established (day 45
   after primary infection) (Supplementary Fig. [114]5a, b). H2A.Z icKO
   and WT OT-I memory cells were then isolated and mixed transferred to
   congenically distinct host. Their memory recall responses were detected
   upon LM-OVA rechallenge. Not only is recall expansion impaired in H2A.Z
   icKO OT-I cells (Fig. [115]2f–h), but the production of effector
   molecules such as IFN-γ and TNF-α (Fig. [116]2i, j and Supplementary
   Fig. [117]5c), as well as Perforin and Granzyme B (Supplementary
   Fig. [118]5d, e), is also significantly reduced. Together, these data
   suggest that H2A.Z regulates the recall response of memory CD8^+ T
   cells in a cell-intrinsic mechanism.

   To further examine the role of H2A.Z in human memory T cells, we
   knocked down its expression using the ProteanFect CRISPRMax-Cas9 mRNA
   Gene Editing Kit in an IL-7/IL-15-induced human CD8^+ T cell memory in
   vitro system^[119]29. We then assessed the expression of the effector
   molecule IFN-γ following stimulation with either Human CD3/CD28 T Cell
   Activator or PMA/Ionomycin. The results showed that H2A.Z deficiency
   led to a reduction in IFN-γ expression in human CD8^+ T memory cells
   upon re-stimulation. These results indicate that H2A.Z indeed plays a
   crucial role in the recall response of human memory CD8^+ T cells
   (Supplementary Fig. [120]6a–d).

Epigenetic features predetermine memory CD8^+ T cell gene induction

   To further define the molecular mechanisms underlying the rapid recall
   response in resting memory CD8^+ T cells compared to naïve cells, we
   first profiled the global epigenome (including H2A.Z deposition,
   chromatin accessibility, H3K4me3, H3K27me3, H3K27ac and Pol II S5P
   enrichment) in memory and naïve OT-I T cells via native ChIP-seq,
   CUT&Tag, and ATAC-seq. Differential peak analysis revealed that H2A.Z
   is deposited differently in memory and naïve OT-I cells genome-wide
   (Supplementary Fig. [121]7a). Great Gene Regional Enrichment of
   Annotations Tool (GREAT)^[122]30,[123]31 analysis of the differential
   peaks showed distinct biological processes enriched in the two cell
   types (Supplementary Fig. [124]7a). The H2A.Z peaks specific to memory
   T cells were enriched with genes involved in “regulation of cell
   adhesion and activation” and “regulation of lymphocyte proliferation”,
   etc., while naïve T cell-specific H2A.Z peaks are enriched with genes
   involved in “DNA damage checkpoint”, “negative regulation of cell cycle
   process”, “mitotic G1/S transition checkpoint”, etc. These data suggest
   H2A.Z may mainly regulate activation in memory T cells while
   homeostasis in naïve T cells. Further analysis showed largely similar
   patterns of ATAC signal, H3K27ac and Pol II S5P between memory and
   naïve T cells (Supplementary Fig. [125]7b–d). However, H3K4me3 showed
   different patterns (Supplementary Fig. [126]7e). Memory T cell-specific
   H3K4me3 peaks are enriched with genes involved in “lymphocyte
   migration”, “regulation of lymphocyte chemotaxis” and “protein
   acetylation”, etc., while naïve T cell-specific H3K4me3 peaks are
   enriched with genes involved in “leukocyte cell-cell adhesion” and “T
   cell activation and differentiation”, etc, probably suggesting
   H2A.Z-independent regulations. To further explore the global functions
   of H2A.Z in chromatin landscapes, we analyzed the chromatin
   accessibility and histone mark changes at the differential H2A.Z sites
   in naïve and memory T cells. The results showed that the permissive
   chromatin landscapes (such as chromatin accessibility and H3K4me3,
   H3K27ac and Pol II S5P) were highly enriched at the naïve T cell-unique
   H2A.Z sites in naïve T cells, but not in memory T cells (Supplementary
   Fig. [127]7f). In contrast, at memory T cell-unique H2A.Z sites, the
   permissive chromatin landscapes were also highly enriched in memory T
   cells, but not in naïve T cells (Supplementary Fig. [128]7f). These
   data suggested that highly enriched H2A.Z is associated with the
   deposition of permissive chromatin landscapes.

   Induction of ARGs is characteristic of memory T cell recall response.
   Comparing the epigenetic features of ARG loci, we found memory T cells
   showed significantly higher deposition of H2A.Z on the ARG promoter
   regions, as well as enhancer regions (Fig. [129]3a, b). Furthermore, we
   also found that the ARG loci in resting memory OT-I T cells had other
   unique epigenetic features compared to those in resting naïve OT-I
   cells. These include lower levels of repressive H3K27me3 enrichment and
   higher levels of permissive H3K4me3, H3K27ac and Pol II S5P enrichment
   in the promoter and enhancer regions (Fig. [130]3a, b). These
   characteristic epigenetic features were typically observed in several
   important ARG gene loci (e.g., Ifng, Gzmb, Ccl3, Ccl4 and Xcl1)
   (Fig. [131]3c). In contrast to the highly inducible ARGs, the silent
   genes, as defined in a previous study^[132]12, showed significantly
   lower level of H2A.Z deposition, as well as deposition of H3K4me3,
   H3K27ac and Pol II S5P (Supplementary Fig. [133]8). Thus, the overall
   permissive epigenetic modifications may set ARGs at a poised state
   conferring rapid recall response of memory T cells.

Fig. 3. Epigenetic features predetermine memory CD8^+ T cell gene induction.

   [134]Fig. 3
   [135]Open in a new tab

   a–c Naïve (CD62L^hiCD44^lo) OT-I CD8^+ splenocytes and day 30-45
   post-infection memory (CD127^hiKLRG1^lo) OT-I cells were sorted from
   spleens. H2A.Z and Pol II S5P CUT&Tag-seq were performed (n = 2
   biologically independent samples per group). Data are representative of
   two independent experiments. Other data (ATAC-seq, H3K4me3, H3K27me3
   and H3K27ac) in the Figure are from published data ([136]GSE89036). The
   Figure shows the density of peaks around TSS (a) and enhancer regions
   (b). Box plots showing signal strength of inducible gene locus in T[N]
   and T[M] using normalized read counts of ATAC-seq and ChIP-seq or
   CUT&Tag-seq peaks in each condition. Statistical significance was
   determined with two-sided paired Wilcoxon test. c Sequencing tracks at
   selected gene loci as displayed on the Integrative Genomics Viewer
   (IGV), with gene structure and transcription orientation marked on the
   top. CNS, conserved non-coding sequence elements.

   The structure scheme of the super enhancer (SE) (Supplementary
   Fig. [137]9a) illustrates its key role in the gene transcription
   process. It recruits transcription factors and RNA polymerase II to
   target gene promoters and promotes the formation of the mediator
   co-activator complex^[138]30. H3K27ac is a commonly used hallmark to
   identify putative genome-wide enhancers^[139]32. Since we found that
   the enrichment of H3K27ac is significantly increased at ARG gene loci
   in memory OT-I T cells than in naïve OT-I cells (Fig. [140]3b), we
   hypothesized that SEs may exhibit differences in memory and naïve OT-I
   cells. To obtain a genome-wide landscape of SEs in memory T cells, we
   analyzed H3K27ac signal in memory and naïve OT-I T cells through the
   rank ordering of super-enhancers (ROSE) algorithm^[141]30. We
   identified 559 super-enhancer regions in memory T cells (Supplementary
   Fig. [142]9b), much more than in naïve OT-I cells (56 super-enhancer
   regions) (Supplementary Fig. [143]9c), indicating their significance in
   memory T cell function. Specifically, 9 of ARGs in memory OT-I cells
   (Supplementary Fig. [144]9d), including Ifng, Gzmb, Ccl3, Ccl4 and
   Xcl1, are associated with super enhancers, while none was found
   associated with SE in naïve OT-T cells (Supplementary Fig. [145]9e).

H2A.Z establishes the poised epigenetic features of ARGs in memory T cells

   To further investigate the relationship between H2A.Z deposition and
   the other epigenetic features, we profiled the epigenome in WT and
   H2A.Z cKO memory OT-I T cells. In WT memory OT-I cells, at the
   genome-wide, H2A.Z is consistently correlated with chromatin
   accessibility, as well as with marks of gene activation such as H3K4me3
   and H3K27ac, but H2A.Z is poorly associated with the gene silencing
   mark H3K27me3 in the promoter and enhancer regions (Fig. [146]4a, b).
   To specifically determine the effect of H2A.Z on ARG loci, we compared
   WT and H2A.Z cKO memory OT-I T cells. The data demonstrate that H2A.Z
   depletion significantly alters the epigenetic landscape of ARG
   promoters and enhancers in resting memory T cells, including decreased
   chromatin accessibility, reduced levels of Pol II S5P and the active
   histone marker H3K27ac, and increased levels of the repressive histone
   marker H3K27me3 (Fig. [147]4c–e). Thus, H2A.Z may serve as “a master
   regulator” in memory CD8^+ T cells to establish a permissive poised
   epigenetic landscape on ARG loci for rapid recall response.

Fig. 4. H2A.Z deletion leads to chromatin landscape changes in memory T
cells.

   [148]Fig. 4
   [149]Open in a new tab

   a–e Naïve OT-I cells sorted from the spleens of WT (OT-I H2A.Z^f/f) or
   cKO (OT-I H2A.Z^f/f GzmB-Cre) mice were transferred into hosts,
   followed by LM-OVA infection. Day 30-45 post-infection memory
   (CD127^hiKLRG1^lo) OT-I cells were sorted from spleens. ATAC-seq and
   CUT&Tag were performed (n = 2 biologically independent samples per
   group). Data are representative of two independent experiments. a, b
   Heat map comparing enrichment signals of H2A.Z, ATAC, H3K4me3,
   H3K27me3, H3K27ac and Pol II S5P at promoter regions (TSS) and enhancer
   regions in wild-type T[M] cells. All heatmaps were generated using the
   same ordered regions ranked by decreasing H2A.Z levels in T[M] cells. c
   The Figure shows the density of peaks around TSS. d Genome browser
   snapshots for representative genomic loci showing H2A.Z enrichment and
   various epigenetic marks. e The Figure shows the density of peaks
   around enhancer regions. Box plots showing signal strength of inducible
   gene locus in WT and cKO T[M] cells using normalized read counts of
   ATAC-seq and ChIP-seq or CUT&Tag-seq peaks in each condition.
   Statistical significance was determined with two-sided paired Wilcoxon
   test.

   Since we found that the enrichment of H3K27ac is significantly
   decreased at ARG gene loci in H2A.Z deficient memory OT-I T cells
   (Fig. [150]4e), we hypothesized that H2A.Z may also regulate SEs at ARG
   gene loci in memory T cells. Indeed, deficiency of H2A.Z in memory OT-I
   T cells results in a significant decrease in H3K27ac levels at SE
   regions at genome-wide (Supplementary Fig. [151]9f). Moreover, at the
   SE regions of ARGs, H3K27ac levels are also significantly decreased in
   H2A.Z deficient memory OT-I cells (Supplementary Fig. [152]9g), as
   specifically demonstrated in Ifng, Gzmb, Ccl3, and Ccl4 loci
   (Supplementary Fig. [153]9h). Thus, H2A.Z might not only regulate ARG
   poised induction at regular promoter and typical enhancer regions in
   memory T cells, but also via SE.

H2A.Z regulates ARG induction during memory T cell recall response

   Our data showed that H2A.Z plays an important role in the establishment
   of poised epigenetic landscape in ARG loci. Whether it is required for
   ARG induction during memory T cell recall response is still unknown. To
   do this, LM-OVA infection-induced WT and cKO memory OT-I cells were
   isolated and stimulated in vitro with OVA[257-264] peptide
   (Fig. [154]5a). RNA-seq analysis was performed for transcriptome
   comparison. The results show that H2A.Z deficiency in memory OT-I cells
   significantly reduced the induction of genes related to
   “cytokine-cytokine receptor interaction”, “chemotaxis” (Ifng, Gzmb,
   Ccl3 and Tnfsf11) and “regulation of interleukin-17 production” (Ccn1,
   Ccl1 and Tnfsf14) (Fig. [155]5b). Specifically, impaired ARG induction
   was also found by gene set enrichment analysis (GSEA) (Fig. [156]5c),
   including Ifng, Gzmb, Tnf, Il2, Xcl1, Ccl3, Ccl4, etc. (Fig. [157]5d).
   Some of them were also confirmed by qPCR (Fig. [158]5e). Flow
   cytometric analysis showed that cKO memory OT-I cell also produced
   reduced levels of IFN-γ and Granzyme B, upon OVA[257-264] peptide
   (Fig. [159]5f) or IL-12/IL-18 restimulation (Fig. [160]5g). Moreover,
   TAM-induced icKO memory OT-I cell showed consistent results, including
   reduced levels of IFN-γ, TNF-α and Granzyme B, upon OVA[257-264]
   peptide (Supplementary Fig. [161]10a–c) or IL-12/IL-18 restimulation
   (Supplementary Fig. [162]10d). These data provided direct evidence that
   H2A.Z is a master regulator to facilitate ARG induction in memory CD8^+
   T cells for their recall response.

Fig. 5. H2A.Z regulates the expression of activation-recalled genes in memory
T cells.

   [163]Fig. 5
   [164]Open in a new tab

   a–g Naïve OT-I cells sorted from the spleens of WT (OT-I H2A.Z^f/f) or
   cKO (OT-I H2A.Z^f/f GzmB-Cre) mice were transferred into hosts followed
   by LM-OVA infection. Day30-45 post-infection T[M] (CD127^hiKLRG1^lo)
   cells were sorted from spleens. a Experiment setup. WT and cKO T[M]
   cells were stimulated in vitro with 200 nM OVA[254-264] for 24 h, RNA
   was extracted for transcriptome analysis (n = 2 biologically
   independent samples per group). b Functional enrichment analysis of
   T[M]activation-recalled genes. c Enrichment plots of defined ARG gene
   set in comparison of cKO versus WT CD8^+ T[M] cell transcriptomes with
   GSEA. d Heatmap of selected genes from ARGs. e qPCR analysis of Ifng,
   Gzmb, Xcl1, Prf1, Cd44, Il2, and H2afz mRNA levels in WT and cKO T[M]
   cells stimulated with 2 µg/ml plate-bound anti-CD3/CD28 Abs for 24 h
   (n = 3 biologically independent samples per group). f, g The splenic
   cells from recipients were stimulated with OVA[257-264] peptide or a
   combination of IL-12 and IL-18. f IFN-γ and Granzyme B production from
   OT-I cells was measured by flow cytometry. Representative flow plots
   show the frequency of CD44^+IFN-γ^+ (top) and CD44^+ Granzyme B^+
   (bottom) cells upon 24 h peptide stimulation (n = 4 biologically
   independent samples per group). g Representative flow plots show the
   frequency of CD44^+IFN-γ^+ cells upon 6 h IL-12/IL-18 stimulation
   (n = 5 biologically independent samples per group). Data are
   representative of two independent experiments in (b–g). Data are
   mean ± SD. Statistical analysis was performed using two-tailed unpaired
   Student’s t test (e–g).

   ARGs is a set of genes that are quickly/robustly induced upon
   re-stimulation in memory T cells. In addition to ARG, many other genes
   are also induced upon re-stimulation in memory T cells, although in a
   lower degree compared to ARGs. We determined them as activation induced
   genes (AIGs) hereafter for comparison. In the deficiency of H2A.Z,
   significant number of AIGs also showed impaired induction in memory T
   cells upon stimulation at genome-wide (Supplementary Fig. [165]11a),
   suggesting a general effect of H2A.Z for gene induction. Comparing the
   level of H2A.Z deposition, we found these AIGs are even deposited
   higher levels of H2A.Z at their promoters compared to those on ARGs
   (Supplementary Fig. [166]11b). These data revealed that H2A.Z
   deposition level is not positively correlated with the induction speed
   or extent of corresponding genes. Moreover, during secondary expansion,
   many other genes are induced at later time points, our data showed that
   many genes are also controlled by H2A.Z at Day7 post-reinfection
   (Supplementary Fig. [167]11c–e).

TCR/CD28 and IL-7/IL-15 signals coordinately induce H2A.Z deposition at ARG
loci

   The preceding data indicates that H2A.Z regulates the expression of ARG
   via epigenetic priming. However, it is still unclear which signals
   drive the enrichment of H2A.Z at ARG gene loci. T cell activation
   signals (TCR/CD28 signaling) and cytokine signals (IL-7 or IL-15, etc.)
   are crucial signals controlling memory CD8^+ T cell
   formation^[168]33,[169]34. We wondered that the deposition of H2A.Z in
   memory T cells might be regulated by these signals. To do this, naïve
   OT-I T cells were stimulated for a short-term with different
   extracellular signals (TCR/CD28, IL-7, IL-15) to evaluate their early
   and direct effect on H2A.Z deposition (Fig. [170]6a). PCA analysis of
   genome-wide differential peaks highlighted significant differences
   between the groups, which were largely driven by TCR signaling, as
   shown in PC1 (Fig. [171]6b). Differential peak analysis showed that
   TCR/CD28 and cytokine signals caused extensive and differential H2A.Z
   enrichment genome-wide in CD8^+ T cells (Fig. [172]6c). Gene Regional
   Enrichment of Annotations Tool (GREAT) analysis^[173]35 on differential
   peaks showed distinct biological processes regulated by TCR/CD28 and
   IL-7 combination signals compared to TCR/CD28 signals alone
   (Fig. [174]6d), while TCR/CD28 and IL-15 combination signals are
   largely similar to TCR/CD28 signals alone, suggesting the differential
   role IL-7 and IL-15 on H2A.Z deposition selection. TCR/CD28 signals
   induced H2A.Z deposition on genes related to “regulation of adaptive
   immune response”, “regulation of TNF, IFN-gamma, and IL-2 production”,
   etc. TCR/CD28 and IL-7 combination signals induced H2A.Z deposition on
   genes related to “regulation of cell activation”, “cytokine
   production”, “cell adhesion and proliferation”, and “chromatin
   organization and remodeling”.

Fig. 6. TCR/CD28 and cytokine signals induce H2A.Z deposition at specific
gene loci.

   [175]Fig. 6
   [176]Open in a new tab

   a Experiment setup. Naïve OT-I cells were stimulated by 1 µg/ml
   plate-bound anti-CD3/CD28 Abs for 24 h (Activated T cells, TA), and
   then stimulated with 50 ng/ml IL-7 or IL-15 for another 24 h,
   respectively. The cells were collected at different steps for H2A.Z
   CUT&Tag-seq (n = 2 biologically independent samples per group). Data
   are representative of two independent experiments. b Principal
   components analysis (PCA). c Quantification of differential H2A.Z peaks
   (p-value < 0.01; |M value | > 1) between subsets on a logarithmic scale
   (across genome-wide regions). d Network analysis (GREAT) of
   biological-process GO term enrichment of the genes with unique
   H2A.Z-deposited peaks induced by changes in TCR/CD28, TCR/CD28  +  IL-7
   or TCR/CD28  +  IL-15 signals. Top GO biological processes terms and
   −log10 BinomFdrQ (binomial FDR q-value) are shown. e The Figure shows
   the density of peaks around TSS (upper panel, promoter regions) and
   enhancer regions (lower panel). Box plots showing signal strength of
   ARGs loci in Naive, TA, TA IL-7, and TA IL-15 using normalized read
   counts of H2A.Z CUT&Tag-seq peaks in each condition. Statistical
   significance was determined with two-sided paired Wilcoxon test. f
   Sequencing tracks at selected gene loci as displayed on the IGV, with
   gene structure and transcription orientation marked on the top.

   Next, we further analyzed whether TCR/CD28 and cytokine signals may
   promote the enrichment of H2A.Z at ARG loci. Interestingly, TCR/CD28
   signals alone induced dramatically increased H2A.Z deposition at ARG
   gene loci, and additional IL-7 or IL-15 signal further but only
   slightly increased H2A.Z deposition (Fig. [177]6e). However, at
   specific gene level, the differential roles of these signals are more
   obvious. For instance, TCR/CD28 signals alone appear to be sufficient
   to induce H2A.Z deposition at effector gene loci (Ifng and Gzmb)
   (Fig. [178]6f), while cytokine signals contribute significantly to
   H2A.Z deposition on chemokine gene loci (Ccl3, Ccl4 and Xcl1)
   (Fig. [179]6f). These data suggest that H2A.Z deposition is primarily
   regulated by TCR/CD28 signals, with specific gene loci also being
   influenced by cytokine signals cooperatively.

Transcription factors mediate site-specific H2A.Z deposition and ARG
induction

   We then further wondered how TCR/CD28 signals could induce H2A.Z
   deposition on ARG loci. We specifically focused on Ifng, a key ARG for
   memory T cell recall response. MOTIF analysis of H2A.Z binding sites on
   Ifng promoter and enhancer regions showed many consensus binding motifs
   of several transcription factors (TF) of AP-1 family and RUNX family.
   AP-1 family is one of the major TFs downstream of TCR
   signaling^[180]36. We hypothesized that AP-1 might be involved in H2A.Z
   deposition on these sites. To avoid the redundancy of AP-1 family or
   even other TFs, we chose to delete the specific TF-binding motif by
   CRISPR-Cas9 technique in EL4 cells to test this hypothesis
   (Fig. [181]7a, Supplementary Fig. [182]12a). The targeting of the
   TF-binding motif successfully led to motif mutation manifested as
   single or multiple nucleotide deletions or additions within the motif
   as determined by DNA sequencing of targeted cells (Supplementary
   Fig. [183]12b–d). Disruption of AP-1 binding motif of 15 bp nucleotides
   at the promoter region (TSS) of Ifng caused a decrease in H2A.Z
   deposition in both resting and activated EL-4 cells (Fig. [184]7b).
   Additionally, disturbance of the overlapping JUN and RUNX family
   binding motif of 15 bp nucleotides at the enhancer region (conserved
   non-coding sequence elements 22, CNS-22) upstream of Ifng also resulted
   in reduced H2A.Z deposition in resting and activated EL-4 cells
   (Fig. [185]7c). Similarly, altering the RUNX family binding motif of
   12 bp nucleotides at the enhancer region (CNS+45) downstream of Ifng
   also led to reduced H2A.Z deposition in resting and activated EL-4
   cells (Fig. [186]7d). More importantly, the induction of Ifng gene
   expression during EL-4 activation was significantly diminished when
   these TF binding motifs were mutated (Fig. [187]7e–g).

Fig. 7. Multiple TFs regulate H2A.Z enrichment at effector gene loci and
contribute to their expression.

   [188]Fig. 7
   [189]Open in a new tab

   a Experiment setup. EL-4 cells were electroporated with
   EGFP-lentiCRISPRv2_sgRNA plasmid. 24 h post-electroporation, GFP^+
   cells were sorted and collected for DNA sequencing. DNA modifications
   were quantified using TIDE. b–g The GFP^+ cells were cultured in
   1 μg/ml plate-bound anti-CD3/CD28 Abs for 24 h and then collected for
   CUT&Tag (b–d) and qPCR (e–g). Data from three biologically independent
   samples (e–g). h Experiment setup. h–k Activated OT-I cells were
   transduced with Retrovirus system. 96 h post-transduction, NT and
   target GFP^+ cells were sorted and mixed at a 1:1 ratio and
   co-transferred into recipients. On the next day, these mice were
   infected with 5 × 10^3 CFU LM-OVA. On day 30-45, the splenic cells from
   recipients were stimulated with PMA/Ionomycin for 4 h. IFN-γ production
   from OT-I cells was measured by flow cytometry. Values denote geometric
   mean fluorescence intensity (gMFI). Data from four (i) or three (j, k)
   biologically independent samples. Data are representative of two
   independent experiments in (b–g and i–k). Data are mean ± SD.
   Statistical analysis was performed using two-tailed unpaired Student’s
   t test (b–g) and two-tailed paired Student’s t test (i–k).

   To evaluate the role of specific TF binding motifs in primary CD8^+ T
   cells, we utilized a retrovirus-based CRISPR-Cas9 technique^[190]37 to
   mutate these motifs in primary murine OT-I T cells (Fig. [191]7h).
   Nontargeted and targeted OT-I cells were then co-transferred at 1:1
   ratio into congenically marked WT recipient mice, which were then
   infected by LM-OVA and memory OT-I cells were isolated at day 30
   post-infection for in vitro re-stimulation to determine their recall
   response. IFN-γ was measured by flow cytometry. The results showed that
   mutations in the specific TF-binding motifs located at the TSS, CNS-22,
   or CNS+45 regions led to a significantly decreased expression of IFN-γ
   in primary CD8^+ T cells (Fig. [192]7i–k).

   Together, these data suggest that the enrichment of H2A.Z at the gene
   loci of Ifng is regulated by specific TFs, and disruption of the TF
   binding motifs decreases the deposition levels of H2A.Z, ultimately
   leading to reduced induction of the target genes.

Discussion

   It is well-established that memory T cells mount faster and more robust
   responses than naïve T cells upon antigen stimulation. However, the
   molecular mechanisms underlying this rapid recall ability remain poorly
   defined. In this study, we demonstrate that genes enabling rapid recall
   (activation-recalled genes, ARGs) exhibit a distinct epigenetic
   landscape at their promoter and enhancer regions in memory T cells
   compared to naïve T cells. Notably, H2A.Z emerges as a key regulator of
   this poised epigenetic state. Genetic deletion of H2A.Z in primary T
   cells reversed activating epigenetic features to a repressive chromatin
   state (Fig. [193]8). Consequently, H2A.Z-deficient memory T cells
   exhibited impaired recall responses both in vitro and in vivo. These
   findings provide direct evidence that H2A.Z-dependent epigenetic
   mechanisms critically regulate the rapid recall response of memory T
   cells.

Fig. 8. Schema of H2A.Z primes an epigenetic landscape for memory CD8^+ T
cell recall response.

   [194]Fig. 8
   [195]Open in a new tab

   Resting memory T cells exhibit a poised chromatin landscape at
   activation-recalled genes (ARGs). H2A.Z is required for memory CD8^+ T
   cell recall response. H2A.Z sets up the poised chromatin landscape on
   the promoters, typical and super enhancers of ARGs for their recall
   expression. TCR/CD28 and IL-7/IL-15 signals cooperate to promote H2A.Z
   deposition at specific ARG loci.

   In this study, we generated foreign antigen (OVA)-specific memory OT-I
   T cells using the LM-OVA infection mouse model. This approach closely
   mimics memory T cells generated by natural infection or vaccination.
   This is different from the previous study in which spontaneously
   generated human virtual memory T cells were used^[196]11. In addition,
   total memory CD8^+ T cells were analyzed in our current study rather
   than separated subsets of central memory and effector memory CD8^+ T
   cells. We identified typical effector genes of the memory recall
   response as ARGs, defined by their capacity for faster or more robust
   functional responses in resting memory versus naïve T cells upon
   stimulation. These ARGs included cytotoxic molecules (Gzmb), cytokines
   (Tnf, Il2, Ifng, Lta), chemokines (Ccl3, Ccl4, Xcl1), costimulatory
   molecules (Havcr2/Tim3, Lag3, Cd70), and transcription factors (Batf3,
   Zeb2, Nr4a3), suggesting their critical roles in regulating recall
   responses. Notably, the previous human study identified distinct poised
   genes (ID2, SLA, NFIL3) and bivalent genes (MLK4, MKX, GJB7) in central
   memory and/or effector memory human CD8^+ T cells^[197]11. It should be
   noted that CD70, FAM129A, IFNG, NR4A3, TNF, TNFRSF4, TNFSF14 and TNIP3
   classified as active genes in resting human memory CD8^+ T cells
   compared to resting naïve cells in the previous study were identified
   in our study as ARGs, based on their enhanced recall kinetics
   (faster/stronger induction) upon stimulation. Thus, our study
   specifically focuses on elucidating the epigenetic basis of the ARGs
   inherent to memory T cells.

   A limitation of this study is the absence of subset-specific analysis
   within the CD8^+ memory T cell compartment, which encompasses T[CM],
   T[EM], and T[RM] populations. While recall response genes may exhibit
   subtle variations between subsets, existing evidence indicates that the
   core epigenetic features at these loci are largely conserved^[198]11,
   suggesting a common regulatory mechanism. Indeed, our data demonstrate
   that both T[CM] and T[EM] exhibit highly similar induction patterns of
   ARGs compared to naïve cells (Supplementary Fig. [199]1d–f). Given the
   profound epigenetic and functional differences between memory and naïve
   T cells significantly outweigh the distinctions among memory subsets,
   this study focused on defining the general epigenetic landscape
   characterizing memory CD8^+ T cells as a population. Whether T[CM],
   T[EM], or T[RM] utilize subset-specific epigenetic mechanisms for
   recall responses remains an important question for future
   investigation.

   While H2A.Z, a key epigenetic regulator, has been extensively
   characterized in diverse model systems^[200]14,[201]21,[202]22, its
   specific functions in T cell development and immune responses remain
   poorly defined. Previous correlative studies linked H2A.Z deposition to
   gene expression in T cells^[203]12,[204]38,[205]39, and we demonstrated
   its requirement for T cell proliferation via DNA replication
   licensing^[206]27. Here, we establish a novel role for H2A.Z in
   licensing the rapid recall response of memory T cells. To specifically
   interrogate this function and circumvent potential confounding effects
   on early T cell development or primary activation, we employed
   conditional knockout strategies: GzmB-Cre^+H2A.Z^f/f mice, in which
   H2A.Z depletion occurs after naïve T cell primary activation, and
   Rosa26-creERT2^+H2A.Z^f/f mice, in which H2A.Z depletion occurs after
   memory T cell establishment (tamoxifen-induced). Memory T cells
   generated from these models exhibited significantly impaired recall
   responses upon both in vitro antigen re-stimulation and in vivo
   infection re-challenge, as measured by defective effector cytokine
   production and reduced expansion. These results provide compelling
   genetic evidence that H2A.Z is essential for the rapid recall function
   of memory CD8^+ T cells.

   H2A.Z may regulate cell expansion through both origin-recognition
   complex (ORC)-dependent and -independent mechanisms. Our data reveal
   that while the number of H2A.Z deposition peaks is comparable between
   memory and naïve CD8^+ T cells, the level of H2A.Z deposition is
   significantly higher in memory cells (Fig. [207]3a, b). This elevated
   deposition may prime memory T cells for enhanced recruitment of
   SUV420H1, increased H3K20me2 modification, and consequently greater ORC
   recruitment – potentially facilitating accelerated DNA replication and
   proliferation upon re-stimulation. However, this specific mechanism
   requires further experimental validation. Beyond direct DNA replication
   regulation, H2A.Z likely controls recall expansion via transcriptional
   regulation of key genes. Notably, several identified ARGs regulate
   lymphocyte proliferation. For instance, Batf3 has been reported to
   drive the expression of c-Myc^[208]40, a critical transcription factor
   controlling T cell expansion. In addition, Zbtb32 has been reported to
   promote the proliferative burst of virus-specific natural killer cells
   responding to infection^[209]41, though its effect in T cells remains
   to be clarified. Thus, H2A.Z likely primes memory T cells for rapid
   expansion by coordinately regulating both DNA replication licensing and
   the transcription of pro-proliferative genes.

   H2A.Z establishes a permissive chromatin state for ARG transcription
   through multiple mechanisms: enhancing chromatin accessibility,
   facilitating histone modifications, and promoting RNA polymerase II
   recruitment. This functional versatility stems from at least two
   fundamental properties. First, intrinsic nucleosome destabilization:
   Compared to canonical H2A, H2A.Z possesses a shorter C-terminal tail
   deficient in positively charged residues^[210]42, reducing its
   DNA-binding affinity. This lowers the energy barrier for DNA unwrapping
   in H2A.Z-containing nucleosomes, a mechanism corroborated by
   single-particle cryo-EM studies revealing enhanced C-terminal tail
   flexibility^[211]16. These biochemical properties intrinsically promote
   higher DNA accessibility. Second, recruitment of chromatin regulators:
   H2A.Z interacts with diverse protein complexes. It associates with ISWI
   chromatin remodelers (SMARCA5 and SMARCA1)^[212]43, which enhance
   accessibility and facilitate transcription factor
   binding^[213]44,[214]45. H2A.Z also recruits MLL complexes^[215]46,
   that catalyze H3K4me3 methylation, a conserved activation mark^[216]47.
   Consistently, H2A.Z knockdown reduces binding of the core MLL subunit
   RbBP5^[217]20. While these mechanisms are central, we cannot exclude
   additional pathways. Notably, 3D chromatin reorganization has been
   linked to memory recall responses^[218]48,[219]49. Through enhanced
   factor recruitment and accessibility, H2A.Z may facilitate structural
   changes in chromatin architecture^[220]22, licensing rapid
   transcriptional induction. Beyond ARGs, H2A.Z broadly occupies other
   inducible genes (Supplementary Fig. [221]11e) and modulates their
   expression^[222]12,[223]39, consistent with its genome-wide
   distribution. However, genes exhibit differential regulatory dependence
   on H2A.Z. Given H2A.Z’s complexity, including isoforms,
   post-translational modifications, and nucleosomal
   contexts^[224]14,[225]21,[226]22,[227]50, we postulate that specialized
   H2A.Z states (e.g., specific isoforms or modifications) may be
   essential for the rapid, robust induction of ARGs, contrasting with the
   slower, moderate induction of other H2A.Z-associated genes. This
   compelling hypothesis warrants detailed future investigation.

   Epigenetic changes are not autonomous processes. Consequently, a key
   question is how the H2A.Z-dependent chromatin landscape characteristic
   of memory T cells is established. Given the profound epigenetic
   differences between naïve and memory T cells, we hypothesized that
   recall-associated changes are likely induced by extrinsic signals
   encountered during differentiation from naïve to memory T cells. To
   identify these signals and assess their direct, early effects, we
   employed a simplified in vitro short-term stimulation model. This
   approach revealed that both TCR/CD28 signaling and IL-7/IL-15 cytokine
   signaling regulate H2A.Z deposition: TCR/CD28 stimulation alone induced
   substantial H2A.Z deposition, while the addition of IL-7/IL-15 signals
   further modified H2A.Z levels, indicating synergistic regulation by
   these pathways. These findings suggest that T cells epigenetically
   encode their signal history to license enhanced recall responses.
   However, the long-term stability of this signal-induced H2A.Z
   deposition and the mechanisms governing its maintenance during memory T
   cell quiescence remain unknown.

   How is H2A.Z specifically deposited at ARG loci? While general H2A.Z
   deposition mechanisms are largely known^[228]21, the factors
   determining locus-specific targeting remain unclear. Motif analysis of
   H2A.Z binding sites within ARG loci identified consensus TF binding
   sequences, suggesting TF-dependent H2A.Z recruitment. To test this
   hypothesis and minimize functional redundancy among TFs, we mutated
   some consensus TF binding motifs within the Ifng locus using
   CRISPR/Cas9 in EL4 cells and primary murine T cells. The mutations
   significantly reduced H2A.Z deposition at the target sites and impaired
   subsequent Ifng induction upon re-stimulation.

   In conclusion, we have identified H2A.Z as an essential epigenetic
   regulator licensing the rapid recall response of memory CD8^+ T cells.
   TCR/CD28 and IL-7/IL-15 cytokine signals, acting during memory
   differentiation, cooperatively establish this H2A.Z-dependent poised
   chromatin landscape. Pharmacologically targeting this pathway
   represents a promising strategy for enhancing vaccine efficacy and T
   cell-based immunotherapies.

Methods

Study design

   The objective of this study was to define the role of H2A.Z in memory
   CD8^+ T cell recall response. In vivo studies were conducted in mice.
   In vitro studies involved primary mouse cells. Multiple experimental
   techniques were used, including flow cytometry, RNA-seq, ATAC-seq,
   ChIP-seq, and CUT&Tag analysis. Males and females were used in these
   experiments.

Animals

   OT-I (Jackson Laboratory, Stain #003831) is a C57BL/6 TCR transgenic
   strain^[229]51, expressing a receptor specific for peptide OVA[257-264]
   in the context of H2-K^b. H2A.Z^f/f mice^[230]52 (RIKEN Center,
   RBRC05765) were kindly provided by Guohong Li (Institute of Biophysics,
   Chinese Academy of Sciences, Beijing, China). GzmB-Cre mice^[231]26
   originated from JAX Lab (Stain #003734) were kindly provided by Qibin
   Leng (Guangzhou Medical University, Guangzhou, China). Rosa26 ERT2-cre
   mice^[232]28 originated from JAX lab (Strain #008463) were kindly
   provided by Bo Zhong (Wuhan University, Wuhan, China). Ly5.1 (Stain
   #002014) transgenic mice were obtained from Jackson Laboratory. All
   mice mentioned above were housed and bred in specific-pathogen-free
   (SPF) facilities under a 12-hour light/12-hour dark cycle, at
   temperatures ranging from 20 to 26°C and relative humidity between 40%
   and 70%, with ad libitum access to water and food. All animal
   experimental procedures were performed with approval (SYXK2020035 and
   ABSL-2-2021006) from the institutional committee of the Institute of
   Biophysics, Chinese Academy of Sciences. Both experimental and control
   mice were co-housed and bred together before separating. Mice are
   euthanized by carbon dioxide (CO[2]) inhalation using the euthanasia
   chamber, and double verified the death by ascertaining cardiac and
   respiratory arrest.

Cell culture

   T cells were maintained in RPMI 1640 (Gibco) supplemented with 10%
   fetal bovine serum (BI), 10 mM Hepes (Gibco), 1% NEAA (Gibco), 1%
   penicillin/streptomycin (Gibco), 50 μM β-Mercaptoethanol (Gibco) in
   humidified atmosphere with 5% CO2 at 37 °C. EL-4 mouse lymphoma cell
   line (ATCC) and Phoenix-ECO cells (ATCC) were maintained in DMEM
   containing 10% fetal bovine serum, 10 mM Hepes and 1%
   penicillin/streptomycin.

Adoptive transfer model

   To detect primary T cell proliferation in vivo, wild-type (WT) and
   H2A.Z^f/f GzmB-Cre (cKO) naïve OT-I T cells were purified by FACS,
   labeled with 5 μM CFSE (ThermoFisher, Cat. No. [233]C34554), and
   co-transferred (mixed at a 1:1 ratio, 5 × 10^6 per mouse) to the
   congenically marked recipients. On the next day, the recipients were
   intravenously (i.v.) infected with 5 × 10^3 CFU of LM-OVA. At 3–7 days
   post-infection, the proliferation of donor cells was detected. To
   generate WT and cKO memory OT-I cells, wild-type (WT) and H2A.Z^f/f
   GzmB-Cre (cKO) naïve OT-I T cells were purified by FACS, separately
   transferred or co-transferred (mixed at a 1:1 ratio, 1.5 to 2 × 10^5
   per mouse) to the congenically marked recipients. On the next day, the
   recipients were intravenously (i.v.) infected with 5 × 10^3 CFU of
   LM-OVA. At 30–45 days post-infection, WT and cKO memory OT-I cells were
   sorted for in vitro activation or transferred into the congenically
   marked recipients for recall response detection. To generate WT and
   icKO memory OT-I T cells, OT-I wild-type (WT) and H2A.Z^f/f Rosa26
   ERT2-cre (icKO) naïve T cells were purified by FACS, separately
   transferred or co-transferred (mixed at a 1:1 ratio, 1.5 to 2 × 10^5
   per mouse) to the congenically marked recipients. On the next day, the
   recipients were i.v. infected with 5 × 10^3 CFU of LM-OVA. On day 30,
   tamoxifen (TAM) was injected intraperitoneally at 100 mg/kg for 5
   consecutive days. On day 10 following TAM injection, WT and icKO memory
   OT-I cells were sorted for in vitro activation or transferred into the
   congenically marked recipients for recall response detection. For
   comparison of naïve and memory T cell response, OT-I naïve (day 0) and
   memory T cells (at 30–45 days post-infection) were purified by FACS,
   mixed at a 1:1 ratio, and co-transferred into the congenically marked
   recipients (2 to 4 × 10^4 cells per recipient). One day after the
   transfer, mice were i.v. infected with 1 × 10^5 CFU of LM-OVA and 4
   and/or 7 days later, OT-I T cell response were analyzed. For secondary
   infection experiments, WT and cKO or WT and icKO memory OT-I T cells
   were purified by FACS from primary recipients at 30–45 days
   post-infection, mixed in a 1:1 ratio, and co-transferred into the
   congenically marked recipients (2 to 4 × 10^4 cells per recipient),
   followed by i.v. infection with 1 × 10^5 CFU of LM-OVA. On day 4 and/or
   7 post-infection, the recall responses were measured. Antibodies used
   for FACS are listed in Supplementary Table [234]2.

Listeria monocytogenes infection

   Listeria monocytogenes expressing the chicken ovalbumin (LM-OVA) was a
   kind gift from Yang-Xin Fu (University of Chicago). Bacteria were grown
   to a logarithmic phase in Brain Heart Infusion (BHI) medium
   (ThermoFisher, Cat. No. CM1135B). For the study of primary immune
   response, mice were intravenously infected with 5 × 10^3 colony-forming
   units (CFU) of ovalbumin-expressing Listeria monocytogenes (LM-OVA).
   Mice that had received OT-I cells were intravenously infected with
   5 × 10^3 CFU of LM-OVA 1 day after the adoptive T cell transfer. To
   analyze the secondary immune response, mice were rechallenged with
   1 × 10^5 CFU of LM-OVA at day 30–45 after primary infection. The recall
   response was determined 4 and/or 7 days after the secondary infection.

RNA extraction

   RNA was extracted by Easy Pure RNA Kit (Transgen, Cat. No. ER101). Add
   BB4 (containing β-mercaptoethanol) from the kit to fully lyse the
   cells, followed by ethanol precipitation, purification column
   adsorption, washing, and removal of genomic DNA with DNase I. Finally,
   30 µl of RNase-free ddH[2]O was added for RNA elution. RNA
   concentration was measured with Nanodrop, and an appropriate amount of
   RNA was used for reverse transcription or RNA library construction. RNA
   was stored at −80 °C.

qPCR

   For reverse transcription of RNA samples, Thermo Fisher Reverse
   Transcription Kit (Cat. No. EN032) was used. The RT-PCR reagents used
   were SYBR Green Mix from Takara Company, and a 96-well plate reaction
   was conducted using the ABI QuantStudio 6. Relative mRNA expression was
   calculated with the 2^-ΔΔCT method. The internal reference gene
   selected is the housekeeping gene Actb (beta-actin).

Flow cytometric analysis

   Cells were incubated with an anti-CD16/CD32 (2.4G2) mAb to block Fcγ
   receptors and then stained on 4 °C for 30 min with several combinations
   of mAbs. DAPI (1000 × dilution) was added before detection to
   distinguish between dead and alive cells. Flow cytometry was performed
   on a LSRFortessa cytometers (BD), and the data were analyzed with the
   Flowjo software (Treestar, San Carlos, CA). Cell sorting was performed
   on FACSAria instruments (BD). Intracellular staining was performed with
   eBioscience Foxp3/Transcription Factor Staining Buffer Kit (Cat. No.
   00-5523-00). Cell apoptosis was quantified by using the APC Annexin V
   Apoptosis Detection Kit (Biolegend, Cat. No. 640932) and 7-AAD
   (eBioscience, Cat. No. 00-6993-50) according to the manufacturer’s
   instructions. Briefly, the testing cells were harvested and washed with
   cold Cell Staining Buffer (Biolegend, Cat. No. 420201). Cell density
   was adjusted to 1 × 10^6 cells/100 μl in 1× Annexin V binding buffer
   and stained with Annexin V-APC and 7-AAD for 15 min at room
   temperature.

Immunoblotting

   T cells were sorted and lysed in lysis buffer (10 mM Tris-HCI, 1 mM
   EDTA, 1% SDS), containing Protease Inhibitor Cocktail (Beyotime, Cat.
   No. P1006). The cell lysates were extracted and electrophoresed on 12.5
   or 15 % polyacrylamide gels. After electrophoresis at 120 V for 30 min,
   proteins were transferred to PVDF membranes. Membranes were blocked in
   5% BSA in TBS-T (10 mM Tris–HCl, 150 mM NaCl, and 0.1% Tween 20) at
   room temperature for 1 h, and then incubated with the following
   antibodies: anti-H2A.Z (Abcam, Cat. No. ab4174) and anti-β-Actin
   (Zsbio, TA-09) in 3 % BSA in TBS-T overnight at 4 °C. After washing,
   the membranes were covered with the HRP-labeled secondary Ab at room
   temperature for 1 h, and visualized with Ultra High Sensitivity ECL Kit
   (MCE, Cat. No. HY-K1005). Images were captured by GenoSens 2000 System
   (Clinx, Shanghai, China). Densitometric quantitation of bands was
   performed using ImageJ 1.54 f software.

RNA-seq and computational analysis

   For naïve and memory OT-I cell activation, 2 × 10^5 T cell were
   inoculated with 2 µg/ml plate-bound anti-CD3 Ab (Biolegend, Cat. No.
   100340) and 1 µg/ml soluble anti-CD28 Ab (Biolegend, Cat. No. 102116)
   for 0, 6 or 24 h in 96-well plate. After treatment, RNA was isolated
   from the T cells using EasyPure RNA Kit (TransGen Biotech, Cat. No.
   ER101-01). For WT and cKO memory OT-I cell activation, 1 × 10^5 cells
   per well were stimulated with 200 nM OVA[257-264] peptide in a 96-well
   plate. After 24 h stimulation, RNA was isolated from the T cells using
   HiPure Total RNA Nano Kit (Magen, Cat. No. R4125-02). RNA libraries for
   RNA-seq were prepared using TruePrep RNA Library Prep Kit for Illumina
   (Vazyme, Cat. No. TR502-02) following manufacturer’s protocols.
   Sequencing was performed on an Illumina NovaSeq 6000 or NovaSeq X Plus
   instrument (150 bp, paired-end) at Berry Genomics Co., Ltd. (Beijing,
   China). Trim Galore (v0.6.7) and Trimmomatic (v0.39) was used to filter
   out low-quality reads. STAR (v2.7.9a) was used to align reads to the
   mouse genome. RSEM (version 1.3.1) with the command
   rsem-calculate-expression was used to calculate the FPKM values for
   gene expression in each sample. Memory T cell activation-recalled genes
   (ARGs) were defined by [1] increase of mRNA expression in activated
   over resting memory cells by at least 4-fold, [2] increase of mRNA
   expression in activated over resting naïve cells by at least 4-fold,
   [3] increase of mRNA expression in activated memory T over
   activated naïve T cells by at least 2-fold, [4] P-value < 0.01. Silent
   genes had FPKM < 2 in resting and activated memory T cells. Memory T
   cell activation-induced genes (AIGs) were defined by [1] increase of
   mRNA expression in activated over resting memory cells by at least
   4-fold, [2] FPKM > 5 in resting and activated memory T cells, [3]
   P-value < 0.01.

ATAC-seq and computational analysis

   4 × 10^4 memory OT-I T cells derived from WT or cKO mice were used to
   assess the chromatin accessibility by ATAC-seq (Assay for
   Transposase-Accessible Chromatin with high-throughput sequencing)
   technology. The TruePrep DNA Library Prep Kit V2 for Illumina (Vazyme,
   Catalog No. TD501-01) was utilized according to the manufacturer’s
   instructions to prepare the sequencing libraries. The Trim Galore
   software was used for quality control of ATAC-seq raw data, filtering
   out sequencing fragments less than 20 bp in length. The genome
   comparison software used is bowtie2 v2.5.1^[235]53. After comparison,
   MarkDuplicates in Picard was used to remove redundant comparison
   results. For quality control of BAM files after comparison, please
   refer to the official documentation of [236]ENCODE. Before peak
   analysis, the BAM file needs to be shifted according to the occupancy
   of the Tn5 enzyme. This operation is completed by alignmentSieve in
   deepTools v3.4.3^[237]54. Read visualization is completed using IGV
   software^[238]55. The bamCoverage function needs to be used to generate
   the bigwig format Input file, and RPKM is selected as the
   standardization method. The Peak calculation software used is MACS2
   v2.2.7.1^[239]56, with the threshold set at q-value < 0.05. Peak count
   calculation is implemented using the featureCounts function of Subread
   v1.6.2^[240]57 software. Peak annotation was completed using the
   ChIPseeker package.

Native ChIP-seq and computational analysis

   Chromatin immunoprecipitation refers to the ULI-NChIP method published
   by Brind’Amour Julie et al^[241]58. Native ChIP-seq was used to conduct
   a detailed analysis of histone H3K27 acetylation (H3K27ac) in memory
   CD8^+ T cells originating from WT and cKO mice to compare their
   distribution characteristics on H2A.Z-regulated gene sites. Antibodies
   used for native ChIP are listed in Supplementary Table [242]3. For
   ChIP-seq, Trim Galore software was used to perform quality control on
   the raw data, sequencing fragments less than 20 bp in length was
   filtered out. The genome alignment software used is bowtie2. After
   alignment, MarkDuplicates in Picard is utilized to remove redundant
   alignment results, followed by the use of samtools v1.10^[243]59 to
   eliminate reads with MAPQ < 10. Read visualization is carried out using
   IGV software. It is necessary to utilize the bamCompare function to
   generate bigwig format input files, with RPKM selected as the
   standardization method. The Peak calculation software employed is
   MACS2, with the threshold set at q-value < 0.05. The count number
   calculation of the Peak is implemented using the featureCounts function
   of the Subread ([244]http://subread.sourceforge.net) software. Peak
   difference analysis is conducted using the DESeq2 software. During the
   calculation process, the Drosophila dm6 version is selected as the
   Spike-in genome to calculate the scale factor between different mouse
   samples. The difference threshold is set to |FoldChange | > 2 and padj
   <0.05. Peak annotation is completed using the ChIPseeker
   package^[245]60. Utilize the bigwig file as input, employ the
   computeMatrix function of deepTools software^[246]54 to calculate the
   density value of the transcript in each region, and use the ggplot2
   package for drawing display.

CUT&Tag and computational analysis

   CUT&Tag assay was performed using the Hyperactive Universal CUT&Tag
   Assay Kit for Illumina Pro (Vazyme, Cat. No. TD904-02) following the
   manufacturer’s instructions. Briefly, a total of 4 × 10^4 freshly
   sorted naïve or memory OT-I T cells were tagged with concanavalin
   A-coated magnetic beads and then incubated with 1 μg of primary
   antibody [rabbit polyclonal anti-Histone H2A.Z, anti-H3K4me3,
   anti-H3K27me3, anti-H3K27ac, anti-RNA polymerase II CTD repeat YSPTSPS
   (phospho S5)] overnight at 4 °C with slow rotation. Antibodies used for
   CUT&Tag are listed in Supplementary Table [247]3. After incubated with
   1 μg secondary antibody (goat anti-rabbit IgG H&L) at room temperature
   for 1 hour, cells were incubated with 0.04 μM Hyperactive pG-Tn5
   Transposon and then resuspended in tagmentation buffer for
   tagmentation. DNA was purified and amplified by 12 to 14 PCR cycles to
   generate the libraries. Libraries were sequenced on Illumina
   Nova-PE150, and 150-bp paired-end reads were generated. For CUT&Tag
   qPCR detection, 5 μl of Stop Buffer was added to extract DNA after
   tagmentation. Using the negative control (IgG) as a control, calculate
   the 2^(-ΔΔCT) values for each group to determine the relative
   enrichment fold: Relative Enrichment fold (treatment)= 2-(ΔCT treatment
   (CT treatment – CT treatment DNA Spike-in)– ΔCT IgG (CT IgG – CT IgG
   DNA Spike-in)); Relative Enrichment fold (control) = 2-(ΔCT control (CT
   control – CT control DNA Spike-in) – ΔCT IgG (CT IgG – CT IgG DNA
   Spike-in)).

   For data processing, raw sequencing data were trimmed and filtered by
   using Trim Galore. Bowtie2 (v2.5.1) was used to align reads to the
   mouse reference genome mm10 with default parameters and then sorted
   using the Samtools (v1.15). MACS2 (v2.2.7.1) was used to call peaks.
   bamCoverage (v3.5.1) was used to generate bigwig files from bam files.
   The distribution of CUT&Tag peaks was annotated with the R package
   ChIPseeker.

Enhancers were stitched and super-enhancers were identified using ROSE

   The ROSE_main.py function from ROSE^[248]30,[249]31 was used to
   identify typical- or super-enhancers based on H3K27ac ChIP-seq signals,
   with the parameters ‘-s 12,500 -t 2500’. Enhancers were annotated to
   their nearest genes using the ROSE_geneMapper.py function.
   ROSE_geneMapper.py function was used to annotate enhancers by their
   nearest genes. H2A.Z-enriched ARG gene regions were then intersected
   with the annotated enhancer regions. Among the 80 ARGs, 42 gene-related
   enhancers were analyzed.

Pathway enrichment analysis

   For functional annotation of GO and KEGG pathway enrichment was
   performed by Metascape ([250]https://metascape.org)^[251]61.
   Differential peaks were analyzed using the Gene Regional Enrichment of
   Annotations Tool (GREAT)^[252]35 for GO biological processes
   enrichment. The results from GREAT were visualized as enrichment plot
   through ImageGP^[253]62.

CRISPR Cas9 targeting

   sgRNAs were designed by using the online tool
   ([254]https://crispor.gi.ucsc.edu/) and subsequently cloned into
   LentiCRISPRv2GFP vector (Addgene, Cat. No. 82416). The lentiviral sgRNA
   vector were electroporated into EL-4 cells. After 24 h electroporation,
   GFP^+ cells were sorted and collected for DNA sequencing. DNA
   modifications were quantified using [255]TIDE. The GFP^+ cells were
   cultured in 1 μg/ml plate-bound anti-CD3/CD28 Abs for 24 h and then
   collected for CUT&Tag and qPCR. Oligonucleotide sequences used for
   EGFP-lentiCRISPRv2_sgRNA plasmid and CUT&Tag-qPCR are listed in
   Supplementary Table [256]4.

Retrovirus based CRISPR Cas9 targeting

   The retroviral sgRNA vectors were previously described^[257]37. sgRNAs
   were cloned into retroviral vector, and then the retroviral sgRNA
   vector was co-transfected into Phoenix-ECO cells with the helper
   plasmid pCL-Eco (Addgene, Cat. No. 12371) to produce retrovirus. Naïve
   OT-I T cells were isolated from OT-I Cas9^+ mice using naïve CD8^+ T
   cell isolation kit (Stem Cell technologies, Cat. No. 19858) according
   to the manufacturer’s instructions. Purified naïve CD8^+ T cells
   stimulated with plate-bound anti-CD3/CD28 Abs (10/5 µg/ml) for 24 h.
   Viral transduction was performed by spin-infection at 800 × g at 32 °C
   for 2 h with 8 μg/ml polybrene (Sigma, Cat. No. H9268-5G), followed by
   continued culture under the condition of 10 ng/ml IL-2. 96 h
   post-transduction, NT and target mAmetrine1.1^+ OT-I cells were sorted
   and mixed at a 1:1 ratio and co-transferred into recipients. On the
   next day, these mice were infected with 5 × 10^3 CFU LM-OVA. On day
   30-45, the splenic cells from recipients were stimulated with
   PMA/Ionomycin for 4 h. IFN-γ production from OT-I cells was measured by
   flow cytometry.

RNP ProteanFect System based gene editing

   Human naïve CD8^+ T cells were isolated from frozen aliquots of PBMCs
   (Milestone Biotechnologies, Cat. No. PB050C) and activated in X-VIVO 15
   Medium (Lonza, Cat. No. 02-060Q) containing 25 μl/ml of Human CD3/CD28
   T Cell Activator (Stem Cell technologies, Cat. No. 10971) for four
   days. Subsequently, mRNA cocktails (which included H2A.Z.1 sgRNA,
   H2A.Z.2 sgRNA, EGFP mRNA, and Cas9 mRNA) were transfected into the
   activated human CD8^+ T cells using the ProteanFect CRISPRMax-Cas9 mRNA
   Gene Editing Kit (ProteanFect, Cat. No. PT05). After 24 h of
   transfection, EGFP^+ cells were sorted and subsequently cultured in the
   presence of IL-7 and IL-15 for 8 days. The transfected human CD8^+ T
   cells were then restimulated with Human CD3/CD28 T Cell Activator or
   PMA/Ionomycin for 4 h before being collected for flow cytometry
   analysis. The oligonucleotide sequences used for the sgRNAs are listed
   in Supplementary Table [258]4. The full-length sgRNAs ( ~ 100
   nucleotides) were synthesized by Tsingke Biological Technology
   (Beijing, China).

Statistical analysis

   All statistical analyses were performed using Prism8 (GraphPad). The
   data were presented as mean ± SD, and a two-tailed Student’s t test or
   two-sided paired Wilcoxon test were used as indicated in the Figure
   legends. The exact value of n (number of biological or experimental
   replicates) can be found in the Figure legends and the exact P values
   were indicated in the graphed data. A P value < 0.05 was considered
   statistically significant. Most experiments were carried out at least
   two or three times, and the findings of all key experiments were
   reliably reproduced.

Reporting summary

   Further information on research design is available in the [259]Nature
   Portfolio Reporting Summary linked to this article.

Supplementary information

   [260]Supplementary Information^ (16.5MB, pdf)
   [261]Reporting Summary^ (265.5KB, pdf)
   [262]Transparent Peer Review file^ (2.6MB, pdf)

Source data

   [263]Source Data^ (1.1MB, xlsx)

Acknowledgements