Graphical Abstract

   graphic file with name fx1.jpg
   [41]Open in a new tab

Highlights

     * •
       γ-chain cytokines revert newly differentiated CD8^+ T cells to a
       naive-like phenotype
     * •
       These “naive-revertant” are primed for secondary challenge
     * •
       Their chromatin landscape is reminiscent of memory cells
     * •
       Specific signaling pathways and transcription factors are involved
     __________________________________________________________________

   Biological Sciences; Molecular Biology; Immunology

Introduction

   The development of T cell memory is essential for long-term health but
   it remains uncertain how this population is maintained over the many
   decades of human lifespan. Current models of CD8^+ T cell
   differentiation propose a unidirectional and irreversible pathway
   whereby initial antigen stimulation triggers sequential differentiation
   of naive (T[N]) cells into central memory (T[CM]), effector memory
   (T[EM]), and effector (T[Eff]) T cells ([42]Klebanoff et al., 2006).
   T cells gradually acquire increasing effector function but this is
   associated with a progressive reduction in the capacity for
   differentiation and self-renewal, i.e., “stemness.” In settings of
   persistent antigenic stimulations T cells may progressively lose
   effector functions and proliferative capacity such that they eventually
   become exhausted.

   Two subsets of phenotypically naive CD8^+ T cells with features of
   memory cells have recently been described. “T-memory stem cells”
   (T[SCM]), which display enhanced capacity for self-renewal and
   multipotent proliferative potential ([43]Gattinoni et al., 2011), are
   believed to be minimally differentiated and located between T[N] and
   T[CM] in the differentiation pathway ([44]Gattinoni et al., 2017).
   “Memory T cells with naive phenotype” (T[MNP]) also exhibit broad
   polyfunctional capability ([45]Pulko et al., 2016) and are thought to
   be functionally imprinted at an early stage of differentiation between
   CD8^+ T[N] and T[CM] subsets. Despite sharing many characteristics,
   T[SCM] and T[MNP] differ in their extended phenotype and it is
   uncertain if they represent distinct and stable subsets or derive from
   a common precursor with phenotypic plasticity. Although CD8^+ T[SCM]
   can be produced in vitro by activating T[N] cells in the presence of
   interleukin (IL)-7, IL-21, and the glycogen synthase-3β inhibitor
   TWS119 9 ([46]Sabatino et al., 2016), the physiological mechanisms
   leading to the generation of both these cells and T[MNP] are largely
   unknown.

   Given the importance of cytokines as key regulators of T cell-mediated
   immunity, we analyzed the effect of different cytokines on T cell
   differentiation after primary stimulation, using T cells from human
   cord blood (CB), which are unlikely to have encountered antigen and
   therefore have a very low frequency of T[SCM] ([47]Gattinoni et al.,
   2011). We observed that recently differentiated CD8^+ memory T cells
   can undergo lineage reversion to a naive-like phenotype when exposed to
   γ-chain cytokines and that these naive-revertant cells share extensive
   phenotypic and functional characteristics with both T[SCM] and T[MNP].
   This work describes a new pathway of T cell differentiation and
   provides a unifying theory for the generation of T cells with a
   “naive-memory” profile.

Results

IL-7 Induces Recently Differentiated CD8^+ Memory T Cells to Revert to a
Naive-like Phenotype

   CB mononuclear cells (CBMCs) were activated with anti-CD3 plus IL-2,
   and the differentiation stage of CD8^+ T cells was evaluated by CD45RA
   and CCR7 co-expression ([48]Klebanoff et al., 2006). As expected,
   activation induced an expansion of T[CM] (CD45RA^−/CCR7^+) and T[EM]
   (CD45RA^−/CCR7^-) subsets with a concurrent reduction in T[N]
   (CD45RA^+/CCR7^+) ([49]Figures 1A and 1B). T[Eff] (CD45RA^+/CCR7^-)
   were not generated in significant number and were not considered
   further.

Figure 1.

   [50]Figure 1
   [51]Open in a new tab

   IL-7 Induces Reversion of Recently Differentiated Memory CD8^+ T Cells
   to a Naive-like Phenotype

   (A) Flow cytometric analysis of phenotypic changes in CD8^+ T cells
   after activation and successive incubation with 25 ng/mL IL-7. CBMCs
   were activated with anti-CD3 plus IL-2, and when the percentage of
   CD8^+ T[N] dropped below 20%, in this case day 5, cultures were
   maintained in IL-7. Numbers indicate the percentage of cells in each
   quadrant. Single representative experiment out of 50.

   (B) Kinetics of phenotype reversion of CD8^+ T cells from the 50
   different CB samples. Each symbol represents one sample. The shaded
   area indicates the interval of time when IL-7 was added for the first
   time.

   (C) CD8^+ T cell proliferation after activation and IL-7
   administration. CBMCs were stained with CFSE either before activation
   (left panels) or at day 9, during phenotype reversion (right panels).
   At the indicated time points, cell phenotype and CFSE content were
   assessed for T[N] (light gray dots) and T[CM] (dark gray dots). Dashed
   lines indicate basal content in CFSE. Single representative experiment
   out of three.

   (D) Flow cytometry evaluation of IL-7-dependent phenotype reversion in
   recently differentiated T[CM] and T[EM]. After activation T[CM] and
   T[EM] were negatively selected. The two cell subpopulations were then
   incubated with IL-7 and monitored for phenotype changes over time.
   Single representative experiment out of three, for each subset.

   (E) The effect of different cytokines on phenotype reversion. CBMCs
   were activated, and when the percentage of CD8^+ T[N] dropped below 20%
   the indicated cytokines were added. The percentage of the cells in the
   different subsets is shown when the percentage of CD8^+ T[N] reached
   the nadir (upper panel) and afterward, when it reached the plateau
   (lower panel). Data from six samples. Paired t test analysis between
   the T[N] levels at nadir and plateau. ∗∗ = p < 0.001.

   (F) Viability of cells incubated with different cytokines. Activated
   CBMCs were incubated from day 4 with each cytokine or medium, and CD8^+
   T cell viability was evaluated by flow cytometry using 7-AAD uptake.
   Data are represented as means ± 1SD of three samples.

   (G) The kinetics of phenotype reversion of CD8^+ T cells activated with
   different artificial stimuli. Data are represented as means ± 1SD of
   three samples. Paired t test analysis between the T[N] levels at nadir
   and plateau. ∗ = p < 0.05, ∗∗ = p < 0.001.

   (H) The kinetics of phenotype reversion of CD8^+ T cells undergoing
   successive cycles of activation/IL-7 incubation. Newly generated CD8^+
   T[Nrev] cells were twice re-stimulated with PHA and induced to revert
   twice with IL-7 when the percentage of T[N] dropped below 20%. Single
   representative experiment out of three.

   (I) Flow cytometry analysis of phenotype changes of CD8^+ T[Nrev] upon
   activation with the cognate antigen. CB T lymphocytes were activated,
   retrovirally transduced with the SSC-TCR, and induced to revert their
   phenotype with IL-7 (left panels). Afterward, cells were incubated with
   peptide-pulsed DCs (central panels). IL-7 was then added again, driving
   the transduced cells to revert their phenotype (right panels). Plots
   were gated on CD8^+ T cells. The upper panels show the percentage of
   transduced (black dots) and non-transduced (gray dots) CD8^+ T cells.

   In order to investigate the role of cytokines in determining the fate
   of recently differentiated memory CD8^+ T cells, IL-7 was added to the
   culture medium when the proportion of CD8^+ T[N] dropped below 20% of
   the CD8^+ population, typically around 1 week after activation (6.7 ±
   1.9 days, mean ±1 SD, n = 50). No further activation stimulation was
   given. In the first 3 days following addition of IL-7 the percentage of
   CD8^+ T[N] continued to diminish, reaching a nadir of 8.4% ± 6.40
   ([52]Figure 1B). However, with continuing IL-7 incubation the great
   majority of CD8^+ T cells started to re-express CD45RA and reverted
   back to a phenotype resembling T[N] and characterized by co-expression
   of CD45RA, CCR7, CD62L, and CD27 and loss of CD45RO expression
   ([53]Figures 1A and [54]S1A). We termed these cells that reverted to a
   naive-like phenotype as “T naive-revertant” (T[Nrev]). This
   re-acquisition of a naive-like phenotype by CD8^+ memory T cells
   reached a plateau by 13–28 days after initial activation (20 ±
   4.7 days) and typically represented over 70% of the CD8^+ T cell
   population (mean 71% ± 12, range 45%–95%). As such, this value was only
   slightly below the mean of 87% of CD8^+ T[N] at day 0 (±5.8, range
   74–98). All samples followed a similar pattern, although there was
   variation both in the time taken to reach the peak T[Nrev] level and in
   the magnitude of the T[Nrev] population at plateau ([55]Figure 1B).
   These differences in the percentage of cells with naive phenotype
   between day 0 and nadir, and between nadir and plateau, were highly
   significant (p = 1.99 × 10^−50 and p = 2.69 × 10^−38, respectively, by
   paired t test).

   To demonstrate that the phenotypic reversion of differentiated T cells
   was not due to selective death or proliferation of individual T cell
   subsets, we enumerated the cells within each cell subset and monitored
   their proliferation. The total number of all cell subsets before and
   after reversion remained largely unchanged ([56]Table S1), and no cell
   proliferation was detected after the addition of IL-7 and during the
   period of phenotypic reversion ([57]Figure 1C). To further confirm that
   the phenomenon was due to modulation of cellular phenotype, recently
   differentiated CD8^+ T[CM] and T[EM] were purified, incubated with
   IL-7, and tracked. Phenotype reversion was again demonstrated for over
   80% of the purified T[CM] and T[EM] ([58]Figure 1D).

T[Nrev] May Undergo Several Rounds of Differentiation and Reversion

   We next assessed whether this property was unique to IL-7 or shared by
   other cytokines. Recently differentiated CBMCs were incubated with
   single and multiple combinations of the γ-chain cytokines IL-2, IL-7,
   IL-15, IL-4, and IL-21. IL-6, an inflammatory cytokine, was also
   incorporated.

   Phenotypic reversion was observed with several of these cytokines, but
   IL-7 was the most potent agent ([59]Figure 1E). Interestingly, CD8^+
   T cells cultured with IL-6 or IL-21 were driven toward a more
   differentiated phenotype with a substantial increase in T[Eff] cells.
   No synergistic effect was observed when IL-7 was administered together
   with IL-2, IL-4, or IL-15. In addition, the ability of individual
   cytokines to promote dedifferentiation to T[Nrev] cells was correlated
   with their ability to maintain CD8^+ T cell survival in vitro
   ([60]Figure 1F), whereas IL-6, IL-21, or medium alone led to
   substantial cell death. This suggests that reversion may be a default
   physiological program of recently differentiated CD8^+ T cells when a
   survival stimulus is provided.

   Phenotypic reversion also takes place following differentiation induced
   by mitogenic stimuli other than soluble anti-CD3 ([61]Figure 1G). The
   percentage of differentiated CD8^+ T cells undergoing IL-7-dependent
   phenotype reversion after activation with phytohemagglutinin (PHA) and
   staphylococcus enterotoxin B (SEB) was similar to those following
   activation with anti-CD3. Activation with CD3/CD28 beads led to a
   smaller proportion of differentiated cells reverting to a naive
   phenotype, but we were unable to fully remove the beads, some of which
   were still attached to the cells when IL-7 was added, and it is likely
   that the resulting continuous antigen stimulation explains the lesser
   reversion achieved. However, the kinetics of reversion were similar
   demonstrating that co-stimulation does not prevent phenotypic
   reversion.

   We further assessed whether cells could undergo more than one cycle of
   phenotypic reversion[.] Since serial rounds of anti-CD3 stimulation led
   to a high rate of cell death, PHA was used for two further rounds of
   activation, each followed by IL-7 incubation ([62]Figure 1H).
   Phenotypic reversion was observed after each cycle of activation and
   IL-7 treatment, indicating that CD8^+ T[N] can undergo repeated cycles
   of differentiation and reversion.

   In order to demonstrate that phenotypic reversion is also possible
   after activation with cognate antigen, CBMCs were transduced with a
   gene encoding a T cell receptor (TCR) specific for a peptide from the
   Epstein-Barr virus (EBV) LMP2 protein ([63]Frumento et al., 2013).
   Following activation and retroviral transduction, cells acquired a
   predominantly T[CM]/T[EM] phenotype but reverted to T[Nrev] when
   incubated with IL-7. Cells were then re-challenged with peptide-pulsed
   autologous dendritic cells (DCs) and underwent differentiation again to
   CD8^+ T[EM] within 5 days ([64]Figure 1I). At this point IL-7 was
   re-added, and after a further 9 days a second reversion to T[Nrev] was
   attained, demonstrating that phenotypic reversion is also possible
   after stimulation with cognate antigen presented by professional
   antigen-presenting cells.

CD8^+ T[Nrev] Proliferate and Differentiate Rapidly into Functional Effector
Cells following Secondary Stimulation

   As T[Nrev] are antigen-experienced cells that have previously undergone
   differentiation and expansion we were interested to assess their
   proliferative potential when compared with primary T[N]. After
   re-stimulation T[Nrev] differentiated into memory subsets more rapidly
   than T[N] and exhibited a higher proliferation rate ([65]Figures 2A and
   2B). T[Nrev] also rapidly acquired effector function, and when
   EBV-specific TCR-transduced T[Nrev] were re-stimulated with
   peptide-pulsed DCs and driven to a T[EM] phenotype the cells expressed
   perforin and granzyme B ([66]Figure 2C) and exerted cytolytic activity
   against target cells ([67]Figure 2D).

Figure 2.

   [68]Figure 2
   [69]Open in a new tab

   CD8^+ T[Nrev] Cells Have Excellent Differentiation and Proliferative
   Potential and the Amplitude of Reversion Depends on the Cell Source

   (A) Flow cytometry analysis of phenotype changes in CD8^+ T[Nrev] and
   T[N] cells from the same CB sample following stimulation with PHA.
   Single representative experiment out of three.

   (B) Flow cytometry analysis of the proliferation of CD8^+ T[Nrev] and
   T[N] cells. Cells from the samples shown in the panel above were
   stained with CFSE at day 0 and activated with PHA. The CFSE content in
   the two cell subsets is shown at the indicated time points. Dashed
   lines represent basal content of CFSE. Single representative experiment
   out of three.

   (C) Flow cytometry analysis of perforin and granzyme B expression by
   the re-stimulated TCR-transduced CD8^+ T[Nrev]. The intracellular
   expression of perforin and granzyme B were assessed in T[Nrev] cells
   transduced with the SSC-TCR after re-stimulating the cells twice with
   cognate peptide-pulsed DCs.

   (D) Cytotoxic assay of re-stimulated, SSC-specific TCR-transduced
   T[Nrev]. The T cells were incubated in a standard ^51Cr cytolytic assay
   with target cells consisting of HLA A∗1101-transduced T2 cells loaded
   with either 1 μg/mL (diamonds), 10 ng/mL (squares) or 1 ng/mL
   (triangles) of SSC peptide. The peptide solvent, i.e., DMSO, was used
   as control (crosses). The percentage of target cell killing at
   different E:T ratios is indicated.

   (E) T[N] cells were isolated from CBMCs and PBMCs, activated with PHA,
   and then incubated with IL-7. The percentage of cells in the different
   subsets was measured by flow cytometry at the indicated time points.
   Data are represented as mean ± 1SD of five CB samples and eight PB
   samples. Analysis by paired t test.

   (F) Differential accessibility of peak regions identified in PB T[N]
   cells versus CB T[N] cells. The x axis indicates Log2 fold change, and
   the y axis indicates –Log10 p value of all peaks. Colored points
   indicate differentially accessible chromatin sites with inaccessible
   sites as blue and accessible sites as red. Data are from three CB and
   three PB samples.

   (G) TF-binding motif enrichments at DACs more accessible in T[N] from
   PB and from CB. Significant (FDR<0.1) pathway enrichments identified
   within DEG lists from T[Nrev].

   (H) CD8^+ T[N], and T[CM] and T[EM,] were isolated from adult PB,
   activated with PHA, and incubated with IL-7 from day 5. The percentage
   of cells in the different subsets was measured by flow cytometry at the
   indicated time points. Data are represented as mean ± 1SD of three
   samples. Paired t test.

T Cells from Cord Blood and Adult Donors Differ in Degree of Reversion and
Chromatin Accessibility

   In order to demonstrate that phenotypic reversion phenomenon is not
   just restricted to T cells from CB we next enriched CD8^+ and CD4^+
   T[N]cells from the blood of adult donors and compared their ability to
   revert to T[Nrev] following in vitro activation ([70]Figure 2E).
   Although the vast majority of CB-derived CD8^+ T cells could be induced
   to revert to T[Nrev], this was seen in less than half of naive CD8^+
   T cells from peripheral blood (PB) of adult donors. The reduced degree
   of reversion in PB cells could not be related to differences in basal
   phenotype as these were identical in T[N] cells from CB and PB
   ([71]Figure S1B). Instead, a remarkable difference was found in the
   chromatin landscape of T[N] from the two sources ([72]Figure 2F), which
   is predicted to influence a range of biological processes
   ([73]Figure S2A). Compared with T[N] from PB, T[N] from CB showed
   increased accessibility to sites binding transcription factors (TFs),
   in particular those from the basic region/leucine zipper motif (bZIP)
   family, of which BATF was the one with the highest number of open
   binding sites ([74]Figure 2G). Moreover, T[Nrev] showed more open
   chromatin regions annotated as regulatory sites for genes involved in
   TCR signaling ([75]Figure S2B).

   Only recently differentiated T[CM] and T[EM] were able to undergo
   reversion and not established T[CM] (CCR7^+/CD45RA^−) and T[EM]
   (CCR7^-/CD45RA^−) CD8^+ T cells from adult blood ([76]Figure 2H). This
   indicates that, despite similar phenotype, the capacity for
   IL-7-induced dedifferentiation is observed only within recently
   differentiated T[CM] and T[EM] and is relatively less efficient in
   adult donors.

Figure 3.

   [77]Figure 3
   [78]Open in a new tab

   CD8^+ T[Nrev] Can Acquire Phenotypic and Functional Characteristics of
   Other Early Memory T Cell Subsets and Are Present In Vivo

   (A) The expression of the markers that discriminate early T[Nrev] from
   T[N] and recently differentiated T[CM] was measured by flow cytometry.
   Single representative experiment out of three.

   (B) Kinetics of CD25 and CD127 expression by CD8^+ T[Nrev] in the
   presence of either IL-2 or IL-7. After phenotype reversion had
   occurred, cells were either maintained in 25 ng/mL IL-7 or switched to
   culture in 30 U/mL IL-2. The absence of one cytokine led to increased
   expression of its cognate receptor. The mean fluorescence intensity
   (MFI) is shown. Data are represented as means ± 1SD of three samples.
   Unpaired t test analysis between the MFI levels at day 28, ∗ = p <
   0.05.

   (C) Kinetics of the expression of discriminatory markers for early
   T[Nrev] cells. The MFI was measured at different time points during
   activation and reversion. Data are represented as means ± 1SD of three
   samples.

   (D) Scheme of the differences in phenotype between the memory CD8^+
   T cell subsets showing naive-like phenotype.

   (E) T[Nrev] were re-stimulated with PMA plus ionomycin, and the
   percentages of cells expressing granzyme B, IFNγ, TNF-α, and
   combinations thereof were measured after 3 and 14 h. Data are
   represented as mean of three samples.

   (F) Cells with the T[Nrev] signature are present among the T[N] from
   tonsils, but not from PB. The expression of CD95 and integrin β7 was
   measured gating CCR7^+/CD45RA^+ CD8^+ T cells from tonsils of seven
   patients with recurrent acute tonsillitis or from PB of seven unrelated
   healthy individuals.

T[Nrev] Phenotype Overlaps with T[SCM] and T[MNP] during In Vitro Culture

   Using available data ([79]Christensen et al., 2001, [80]Hendriks
   et al., 2003, [81]Hermiston et al., 2003, [82]Ishida et al., 1992,
   [83]Kim et al., 2006, [84]O'Shea et al., 1992, [85]Zehnder et al.,
   1992) we identified 28 membrane-bound proteins that exhibit a
   differential pattern of expression following T cell activation and
   differentiation. Antibodies against these proteins were then used to
   contrast the phenotypic profile of T[N] and T[Nrev]. The pattern of
   expression of integrin β7, CD25, CD127, CD95, CXCR3, and CD49d was
   found to discriminate between these two subsets ([86]Figure 3A).
   Interestingly, CD95 and CXCR3 are also distinct markers of T[SCM]
   ([87]Gattinoni et al., 2011) and CD49d is a marker of T[MNP] ([88]Pulko
   et al., 2016). T[Nrev] and T[CM] are clearly distinguished by CCR7 and
   CD45RA expression, but additional differences were also found in the
   expression of CD25, CD45RO, CD69, CD95, CD120b, CD122, and PTK7
   ([89]Figures 3A and [90]S3). These differences demonstrate that the
   phenotypic correlates of reversion extend substantially beyond
   differential expression of CCR7 and CD45RA.

   We next went on to investigate the relationship between T[Nrev] cells
   and T[SCM] and T[MNP] subsets. In particular we were interested in the
   stability of the T[Nrev] phenotype and how this could be modulated by
   culture conditions. CD8^+ T[Nrev] cells had a very similar phenotype to
   T[SCM] and T[MNP], and all three expressed a common profile of
   CD45RA^+/CCR7^+/CD27^+/CD62L^+/CD45RO^−/CXCR3^+/CD31^+/CD122^lo.
   Indeed, the only difference between early T[Nrev] and the other two
   subsets was that both T[MNP] and T[SCM] express the IL-7 receptor
   (CD127) and T[SCM] lack expression of the IL-2 receptor (CD25), whereas
   T[Nrev] exhibit a CD127^lowCD25^high phenotype. However, expression of
   the receptors for IL-7 and IL-2 is down-regulated on T cells in the
   presence of their respective cytokines ([91]Minami et al., 1993,
   [92]Park et al., 2004, [93]Vranjkovic et al., 2007). When T[Nrev] were
   deprived of IL-7 and maintained in IL-2 for 2 weeks they adopted a
   phenotype almost indistinguishable from T[SCM] and T[MNP] with rapid
   decrease in CD25 expression and progressive increase in CD127
   expression ([94]Figure 3B). After an additional 2 weeks of culture with
   IL-2 there was further progressive loss of CD49d, CD95, CXCR3, and
   integrin β7 ([95]Figure 3C) such that late T[Nrev] acquire a phenotype
   approaching that of primary T[N], although their epigenetic signature
   clearly identifies them as a different population. We were unable to
   monitor the cells further due to increased cell death. These results
   suggest that the cytokine milieu and the time since activation account
   for the modest phenotypic diversity between T[Nrev], T[SCM], and T[MNP]
   ([96]Figure 3D). T[Nrev] also share functional characteristics with
   T[MNP] as the latter cells express granzyme B and secrete IFNγ and
   TNF-α after re-stimulation with phorbol myristate acetate (PMA) plus
   ionomycin ([97]Pulko et al., 2016). Resting T[Nrev] also became
   polyfunctional after the same treatment, r ([98]Figure 3E).

   To assess the in vivo relevance of phenotypic reversion we also looked
   for the presence of cells with the T[Nrev] signature in the blood and
   secondary lymphoid tissue (tonsils) of adult donors. CD8^+ T cells with
   a CCR7^+CD45RA^+ naive phenotype and expression of CD95 and integrin β7
   were present within tonsil but were not seen in blood (p = 0.017,
   unpaired t test, [99]Figure 3F). The population of memory T cells
   within tonsil had a CD95^low and integrin β7^− phenotype
   ([100]Figure S4) and was clearly distinguishable from the putative
   T[Nrev].

Substantial Chromatin Reorganization Is Observed during Differentiation from
T[N] to T[EM] and This Is Partially Retained following Reversion

   Epigenetic modifications have a profound regulatory influence on CD8^+
   T cell differentiation and function ([101]Henning et al., 2018,
   [102]Moskowitz et al., 2017). We next used ATAC-seq to investigate the
   profile of chromatin landscape remodeling during T cell differentiation
   and reversion. ATAC-seq analysis was performed on purified T[N], T[CM],
   T[EM], and T[Nrev] CD8^+ populations following in vitro culture.
   ATAC-seq read density profiles at phenotype-defining genes such as CCR7
   and GZMB were compatible with lineage-specific expression
   ([103]Figure S5).

   Differentially accessible chromatin sites (DACs) were identified within
   memory subsets and contrasted with the profile in naive cells
   ([104]Figures 4A and [105]S6A). The majority of epigenetic
   modifications were acquired relatively late in differentiation during
   transition from T[CM] to T[EM]. In particular, only 26 DACs developed
   during transition from naive to central memory cells but this increased
   markedly to 5,829 with further differentiation to T[EM]. Interestingly,
   51% of these DACs were lost during reversion to T[Nrev] but 2,830 DACs
   still remained within T[Nrev] cells ([106]Figures 4A and 4B). Read
   density distributions at T[EM] and T[Nrev] DACs further indicate that
   the chromatin landscape of T[Nrev] cells retains similarity to T[EM]
   cells ([107]Figures 4C and 4D) and is supported by principal component
   analysis of sample-wise chromatin accessibility, which identified
   unique groups for the T[N], T[CM], and T[EM] populations and alignment
   of T[Nrev] with the memory subtypes ([108]Figure S6B).

Figure 4.

   [109]Figure 4
   [110]Open in a new tab

   Chromatin Landscape Changes during Differentiation from T[N] to T[CM]
   and T[EM] and Reversion to T[Nrev]

   (A) Pairwise comparisons of chromatin accessibility at peak regions.
   The x axes indicate Log[2] fold change, and the y axes indicate
   unadjusted –Log[10] p value of all peaks[.] Colored points indicate
   differentially accessible chromatin sites with inaccessible sites as
   blue and accessible sites as red.

   (B) DACs count for accessible and inaccessible regions in comparison
   with T[N] are shown for T[CM], T[EM], and T[Nrev].

   (C) Tracking from T[N] to T[Nrev] the accessibility of DACs becoming
   inaccessible (T[N] to T[EM] closing)/accessible (T[N] to T[EM] opening)
   upon differentiation to T[EM].

   (D)Tracking from T[N] to T[Nrev] the accessibility of DACs becoming
   inaccessible (T[N] to T[Nrev] closing)/accessible (T[N] to T[Nrev]
   opening) upon reversion to T[Nrev.] The y axis is chromatin
   accessibility in units of rlog normalized mapped reads at the peak
   site.

   (E) Overlaps of differentially accessible sites. Overlaps of DACs more
   accessible in T[CM]/T[EM]/T[Nrev] compared with T[N] (left) and DACs
   less accessible in T[CM]/T[EM]/T[Nrev] compared with T[N] (right).

   We next investigated the balance of open and closed chromatin during
   differentiation and reversion. A relative increase in the number of
   inaccessible chromatin regions was observed during differentiation with
   4,316 closed and 1,299 open DACs in T[EM] compared with T[N]
   ([111]Figure 4E). This profile was retained within the T[Nrev]
   population with values of 2,194 and 617 DACs, respectively. The great
   majority of DACs within T[Nrev] were shared with the T[EM] population,
   although 216 and 546 regions were uniquely open and closed,
   respectively, within this subtype.

Biochemical Pathways Associated with Reversion Can Be Identified by
Epigenomic and Transcriptional Analysis of T Cell Subsets

   In order to examine the relationship between chromatin landscape and
   gene expression, DAC regions were next annotated with the gene whose
   transcriptional start site was nearest to the peak summit.
   As anticipated, differential chromatin accessibility was observed at
   genes encoding phenotypic markers that discriminate T cell subsets
   including CD25, CD127, integrin β7, and CXCR3 ([112]Figure 5A)[.]

Figure 5.

   [113]Figure 5
   [114]Open in a new tab

   Chromatin Accessibility Identifies Biochemical Pathways Specific for
   T[N] and T[Nrev]

   (A) ATAC-seq signal tracks at selected markers that discriminate early
   T[Nrev] from T[N] and recently differentiated T[CM]. Gene diagrams
   (bottom) show alternative transcripts with black boxes indicating
   exons. Each subset signal is aggregated across the constituent samples,
   n = 4(T[N]), 3(T[CM]), 3(T[EM]), 4(T[Nrev]). The y axes are in units of
   reads per million mapped reads.

   (B) DAC sites in T[Nrev] versus T[N] cells. Normalized mapped read
   density (RPKM) of aggregated T[N] (n = 4) and T[Nrev] (n = 4) mapped
   ATAC-seq reads at differentially accessible chromatin sites (centered
   on peak summit, extended +/− 5 kbp).

   (C) Pathways significantly enriched (adjusted p value <0.1) in genes
   closest to T[N] accessible or T[Nrev] accessible DACs.

   (D) Efficiency of reversion in the presence of ATRA. The percentage of
   CD8^+ T[Nrev] was recorded at day 24 from activation in three CBMC
   samples. Unpaired t test.

   (E) After activation CBMCs were incubated with the indicated
   combination of IL-7 25 (ng/mL), ATRA (10^−8M), and BMS (493 3 ×
   10^−6M). Data are from three CBMC samples. One-way ANOVA.

   Genomic regions enrichment of annotations (GREAT) analysis ([115]Table
   S2) revealed that the 1,648 DAC sites that became closed in T[EM,] and
   were not reopened in T[Nrev], were enriched for pathways related to TCR
   signaling in naive T cells and CXCR4-mediated signaling. Sites that
   were opened during reversion of T[EM] to T[Nrev] showed enrichment for
   IL-7 signal transduction and Lck/Fyn-mediated initiation of TCR
   activation, whereas the 546 DAC sites uniquely closed in T[Nrev] were
   enriched for pathways involved in generation of T cell cytotoxicity.

   To further interrogate the differences between true T[N] and T[Nrev]
   cells we combined DAC analysis with transcriptional analysis of the two
   populations. In total 2,830 DAC regions had been identified between the
   two subtypes, 622 of which were more accessible in T[Nrev] and 2,208
   less accessible ([116]Figure 5B). These regions were annotated with the
   gene whose transcriptional start site was nearest to the peak summit.
   Pathway enrichment analysis with g:Profiler highlighted that genes
   becoming less accessible in T[Nrev] included those that regulate the
   major differentiation pathways for Th1, Th2, and Th17 cells
   ([117]Figure 5C). Genes associated with retinol metabolism became
   markedly more accessible in T[Nrev] and is of note given the
   pleiotropic effects on retinoic acid on T cell differentiation
   ([118]Beijer et al., 2013). Indeed, escalating doses of all-trans
   retinoic acid (ATRA) progressively inhibited reversion of IL-7-treated
   CD8^+ T cells ([119]Figure 5D) and this was blocked by a pan-retinoic
   acid receptor antagonist ([120]Figure 5E).

   Transcriptional analysis confirmed that RNA expression levels were
   strongly associated with ATAC-seq read density at transcription start
   sites and enhancer regions ([121]Figure 6A), and enrichment analysis of
   gene sets annotated at T[Nrev] DACs confirmed strong relationship
   between chromatin accessibility and expression level ([122]Figure 6B).
   Transcriptome analysis revealed 2,841 differentially expressed genes
   including major immune regulators such as FOS, JUN, KIT, and IL2RA
   ([123]Figures 6C and 6F and [124]Table S3). Moreover, 32 of the 104
   genes within the Notch signaling pathway were differentially regulated
   in T[Nrev] ([125]Figures 6D and 6G). Gene set enrichment analysis
   (GSEA) of regions closest to Notch signaling genes confirmed an
   enrichment in accessible chromatin regions in T[Nrev] cells
   ([126]Figure 6E) particularly in the HDAC2 and HDAC9-associated
   regions.

Figure 6.

   [127]Figure 6
   [128]Open in a new tab

   Chromatin Accessibility and Transcriptome Profiles of T[N] and T[Nrev]

   (A) Average mapped read density profiles (centered on TSS (top) and
   enhancer regions (bottom) generated from aggregated T[N] (left) and
   T[Nrev] (right) ATAC-seq reads. Profiles are for genes identified from
   transcriptome analysis to be expressed at low, medium, and high levels
   in T[N] and T[Nrev] cells, respectively.

   (B) Enrichment score (ES) for gene expression changing coordinately
   with chromatin accessibility. ES of genes closest to T[Nrev] accessible
   (red) and T[N] accessible (green) DACs, vertical bars indicate position
   of such genes on the axis of fold change in expression (ranking from
   most down-regulated in T[Nrev] to most up regulated in T[Nrev]).

   (C) Microarray analysis was performed on three CD8^+ T[N] samples and
   on the respective CD8^+ T[Nrev] cells. Pairwise comparison of gene
   expression in T[N] versus T[Nrev] cells showing –Log[2] p value versus
   Log[2] fold change of all genes[.] Colored points indicate
   differentially expressed genes that are down-regulated in T[Nrev]
   (blue) and up-regulated in T[Nrev] (orange).

   (D) The genes in the Notch pathway (yellow dots) that are significantly
   down-regulated and up-regulated in T[Nrev] are labeled.

   (E) Barcode plot from GSEA analysis showing ES of Notch signaling genes
   within sites ranked by chromatin accessibility. Vertical bars indicate
   position of such genes on the axis of fold change in chromatin
   accessibility, ranking from most inaccessible in T[Nrev] to most up
   accessible in T[Nrev].

   (F) As in (A) but for aggregated T[Nrev] (left) and T[Mem] (right)
   ATAC-seq reads.

   (G) RNA-seq analysis was performed on four T[Mem] samples (two T[CM]
   and two T[EM]) and on the respective CD8^+ T[Nrev] cells. Pairwise
   comparison of gene expression in T[N] versus T[Nrev] cells showing
   –Log[2] p value versus Log[2] fold change of all genes[.] Colored
   points indicate differentially expressed genes that are down-regulated
   in T[Nrev] (blue) and up-regulated in T[Nrev] (orange).

   Comparative transcriptional analysis of T[Nrev] with memory T cells
   (T[Mem]) ([129]Figures S7A–S7C) also confirmed RNA expression to be
   associated with ATAC-seq read density ([130]Figure 6F), and pairwise
   comparison revealed 447 genes down-regulated in T[Nrev] including
   several involved in T cell cytolytic activity such as GZMB, GZMH, and
   GNLY. In contrast, 267 genes were up-regulated and included proteins
   that support cell survival and act to block cell differentiation, such
   as CISH, NPDC1, HSF4, and OSM ([131]Figure 6G). Interrogation of the
   transcriptomes of T[Nrev], T[Mem], and T[N] subsets allowed
   identification of 97 signature genes specific for T[Nrev] ([132]Figures
   7A and 7B) and pathways enriched in genes that were differentially
   regulated in this subset ([133]Figure 7C). These include a number of
   intracellular signaling pathways, such as JAK-STAT and the
   RUNX-dependent regulation of WNT signaling. Comparison of TF
   transcripts in T[Nrev] and T[Mem] ([134]Figure S8) showed differences
   in the expression of TFs related to stemness and generation of T[SCM]
   ([135]Gattinoni et al., 2012, [136]Kondo et al., 2018).

Figure 7.

   [137]Figure 7
   [138]Open in a new tab

   Transcriptional Signatures of T[Nrev] Cells

   (A) Intersections of DEGs identified from differential expression
   analysis of T[Nrev] versus T[Mem] and T[Nrev] versus T[N].

   (B) Heatmap of the top 40 T[Nrev] signature genes. Z-scores were
   calculated for each gene from RNA-seq (left) and Microarray (right)
   datasets independently.

   (C) Significant (FDR<0.1) pathway enrichments identified within DEG
   lists from T[Nrev] versus T[Mem] and T[Nrev] versus T[N].

   (D) TF-binding motif enrichments at DACs more accessible compared with
   T[N]. Motif enrichments were calculated on the following DACs: 4 shared
   in T[CM], T[EM] and T[Nrev]; 890 in T[EM]; 401 shared in T[EM] and
   T[Nrev]; 216 in T[Nrev].

   (E) TF-binding motif enrichments at DACs less accessible compared with
   T[N]. Motif enrichments were calculated on the following DACs: 14
   shared in T[CM], T[EM], and T[Nrev]; 2,868 in T[EM]; 1,648 shared in
   T[EM] and T[Nrev]; 546 in T[Nrev].

   (F) Expression of TFs with binding motifs enriched in loci that are
   uniquely open in T[Nrev] cells. The x axis indicates fold change in TF
   expression, and the y axis indicates -log[10] p value of TF-binding
   motif enrichment in loci uniquely more accessible in T[Nrev] cells.
   Increasing dot size indicates increasingly significant difference in TF
   expression and color indicates TF family.

   (G) As in (D) but for TFs with binding motifs enriched in loci uniquely
   closed after reversion to T[Nrev].

Reversion Is Associated with Opening of Binding Sites for RUNX and STAT
Transcription Factors and Closure of Sites that Bind BATF

   TFs can instigate chromatin remodeling and act as important regulators
   of differentiation. As such, we next assessed the relative enrichment
   of TF-binding motifs within genomic regions that became differentially
   open or closed in the different cell subsets. DACs that became more
   accessible in T[EM] were enriched in binding motifs for five TF
   complexes, including PU.1/IRF, COUP-TFII, and zinc finger members, and
   these open sites were largely retained during reversion to T[Nrev]
   cells ([139]Figures 7D and 7E). In contrast, many TF-binding motifs
   became less accessible upon differentiation to T[EM] although this
   profile was not strongly retained in the T[Nrev] subset.

   Finally, in order to gain insight into the potential regulators of
   reversion we focused on the relative access of TFs for regions of
   chromatin that were uniquely open or closed in T[Nrev] cells.
   Strikingly, chromatin regions containing binding motifs for ET1, RUNX1,
   and RUNX2, as well as STAT1, STAT4, and STAT5, all became more
   accessible during reversion, although transcriptional activity from
   RUNX1 and RUNX2 genes was suppressed in the revertant population
   ([140]Figure 7F and [141]Table S4). In contrast, there was substantial
   closure of chromatin regions containing binding motifs for BATF and
   ZNF189 ([142]Figure 7G and [143]Table S4), which is of particular note
   given the importance of BATF in driving PD-1-mediated T cell exhaustion
   ([144]Wherry, 2011).

Discussion

   There is increasing appreciation of the importance of cellular
   plasticity ([145]Blanpain and Fuch, 2014), and reversion of mature
   cells into stem cells has been confirmed in some organ systems
   ([146]Stange et al., 2013, [147]Tata et al., 2013). Here we show that
   extensive phenotypic reversion of effector CD8^+ T cells to naive
   phenotype can occur within the human immune system. Reversion was
   mediated primarily by IL-7, which is expressed by a range of stromal
   cells within secondary lymphatic organs and is therefore available to
   support reversion following local antigen-driven activation ([148]Huang
   and Luther, 2012). Indeed, cells with a T[Nrev] phenotype were located
   within human tonsil. An interesting observation was that the addition
   of IL-7 markedly decreased apoptosis, suggesting that reversion may be
   a “default” pathway of recently activated T[N] cells in the presence of
   appropriate survival signals ([149]Hong et al., 2012, [150]Mackall
   et al., 2011, [151]Surh and Sprent, 2008).

   The finding sheds light on the potential relationship between the
   T[SCM], T[MNP], and T[Nrev] populations. These three groups share a
   common core phenotype exemplified by expression of CD45RA, CCR7, and
   CXCR3. CD45RA regulates the signaling threshold in T lymphocytes
   ([152]Hermiston et al., 2003, [153]Irles et al., 2003), and CCR7
   permits entry into secondary lymphoid tissue, whereas CXCR3 plays a
   fundamental role in extravasation of CD8^+ T cells into inflammatory
   sites ([154]Cole et al., 1998, [155]Mikucki et al., 2015, [156]Peperzak
   et al., 2013). The naive-memory phenotype therefore generates an
   antigen-specific memory pool, which retains the ability to enter
   lymphoid tissue but has developed heightened responsiveness to
   re-challenge and may gain rapid access to tissue sites.

   Interestingly, the major differences in the phenotype of T[Nrev],
   T[SCM], and T[MNP] relate to expression of the IL-7 and IL-2 receptors
   and these were modulated on T[Nrev] according to the local
   concentration of IL-2 and IL-7 ([157]Minami et al., 1993, [158]Park
   et al., 2004, [159]Vranjkovic et al., 2007). T[SCM] also express
   CD95^+, whereas T[MNP] express CD49d, and we observed that both
   receptors are down-regulated during prolonged incubation, with CD95
   expression falling more rapidly than CD49d expression. As such, we
   would suggest that the three populations represent a spectrum of
   naive-memory cells that are generated by reversion and display modest
   phenotypic diversity that varies according to cytokine concentration
   and time since activation. Indeed, phenotypic transition of T[EM] to
   T[CM] populations has been observed in murine systems ([160]Wherry
   et al., 2003), and reversion from T[EM] to T[CM] and from these subsets
   to T[SCM] was described in patients following hematopoietic stem cell
   transplant ([161]Cieri et al., 2015).

   This work builds on the work of Cieri et al. who previously
   demonstrated the generation of T[SCM] from naive cells after T cell
   engagement in the presence of IL-7. However, in that study the cytokine
   was included in culture from the start of activation. We certainly do
   not rule out the possibility that T[MNP] and T[SCM] may also be
   generated by additional mechanisms such as direct differentiation from
   T[N] ([162]Cieri et al., 2013, [163]Lugli et al., 2013, [164]Sabatino
   et al., 2016, [165]Zanon et al., 2017) and CD8^+ T[SCM] have also been
   induced in vitro by activation of T cells in the presence of a glycogen
   synthase-3β inhibitor ([166]Sabatino et al., 2016). Engagement of the
   Notch pathway in memory T cells also generates T[SCM]-like CD8^+ and
   CD4^+ T cells ([167]Kondo et al., 2017), and we found a number of genes
   in the Notch pathway to be differently expressed between T[N] and
   T[Nrev] ([168]Figure 6D). Furthermore, IL-7 can further enhance with
   Notch signaling to induce a T[N]-like phenotype in recently activated
   CD8^+ T cells ([169]Kondo et al., 2018). These findings suggest that
   notch engagement and IL-7 may represent two alternative mechanisms to
   generate T[SCM]-like cells.

   Importantly, we observed that T[Nrev] cells were able to undergo
   several rounds of reversion in vitro, and as such this process may be
   important in protection from both acute and chronic infectious agents.
   Indeed, it is now clear that a substantial proportion of
   antigen-experienced T cells is contained within the apparent “naive”
   CD8^+ T cell pool ([170]Ellefsen et al., 2002, [171]Remmerswaal et al.,
   2012). Longer-term culture of T[Nrev] led to progressive
   down-regulation of integrin β7 and CXCR3, but it is uncertain if they
   can ultimately acquire a phenotype indistinguishable from that of
   primary T[N].

   Most of our work was performed on T cells from CB, which contains very
   few memory or effector cells. We did observe phenotypic reversion using
   T[N] cells from adult donors, although the relative proportion of cells
   that underwent reversion was sharply reduced and may reflect the
   decreased number of open chromatin sites and binding sites for TF in
   adult PB. Interestingly, the epigenetic landscape of CD8^+ T cells in
   older people is globally biased toward a differentiated phenotype
   ([172]Moskowitz et al., 2017) with a reduction in chromatin
   accessibility that is most apparent at the IL-7R locus ([173]Ucar
   et al., 2017).

   Phenotypic reversion may help to explain the paradox of preservation of
   the human T[N] pool during aging despite in vivo analyses and
   mathematical modeling suggesting that thymic output is insufficient for
   its maintenance ([174]den Braber et al., 2012, [175]Hakim et al., 2005,
   [176]Murray et al., 2003). Interestingly, it was found that the
   T[SCM]/T[N] ratio in PB increases with age ([177]Li et al., 2019),
   suggesting that during aging naive-memory cells replace T[N] and
   sustain immunological memory.

   Epigenetic regulation maintains the equilibrium between self-renewal
   and differentiation of stem cells and regulates tissue homeostasis
   throughout life. As such we were keen to understand how the chromatin
   landscape was modified following T cell activation and if these changes
   were reversible during phenotypic reversion to T[Nrev]. Interestingly,
   this was largely not the case and T[Nrev] cells retained an epigenetic
   profile that was similar to effector cells. As such, T[Nrev] become
   “epigenetically primed” for secondary activation at the same time as
   they undergo phenotypic reversion to a naive-memory phenotype. The
   chromatin modifications within T[Nrev] are likely to explain their
   ability to rapidly differentiate and acquire effector function in
   response to secondary stimulation and reveal a discrepancy between the
   degree of phenotypic and epigenetic reversion. This may partly explain
   the decrease in chromatin accessibility in phenotypically naive CD8^+
   T cells in older people ([178]Ucar et al., 2017). Furthermore, the
   observation that the chromatin landscape of T[Nrev] is much more
   extensively modified than that of T[CM] provides further confirmation
   that they have undergone sequential differentiation and reversion
   rather than minimal differentiation from T[N].

   Analysis of the distribution of TF-binding sites within differentially
   accessible chromatin regions can help to identify potential
   transcriptional regulators of differentiation. A striking observation
   was that many chromatin regions containing binding sites for BATF
   became closed during reversion. BATF is an essential regulator of CD8^+
   differentiation ([179]Kurachi et al., 2014), and PD-1 engagement on
   T cells can drive BATF-dependent terminal differentiation ([180]Quigley
   et al., 2010).

   Reversion may play a potential role in limiting T cell exhaustion in
   both physiological and pathological settings, and it is noteworthy that
   the chromatin region containing the IL-7R gene becomes poorly
   accessible in exhausted CD8^+ cells ([181]Scott-Browne et al., 2016).
   In contrast, chromatin regions enriched for binding sites of several
   TFs, including RUNX, STAT, and ETS family members, became more
   accessible within revertant subsets. Stat1 and Stat4 signaling regulate
   T cell responses to interferon and cytokine signaling ([182]Gil et al.,
   2006, [183]Nguyen et al., 2002, [184]Thierfelder et al., 1996), STAT5
   is critical in maintaining effector CD8 T cell responses ([185]Tripathi
   et al., 2010), and over-representation of ETS motifs in chromatin
   accessible regions has been observed previously in naive T cells
   ([186]Moskowitz et al., 2017). As such the “epigenetic priming” of
   naive-memory subsets appears to reflect an increased sensitivity to
   interferon and cytokines within the local microenvironment but
   protection from terminal differentiation.

   Cytokine-driven reversion of recently activated CD8^+ T cells thus
   uncovers a novel pathway for T cell differentiation and provides a
   unifying hypothesis for the existence of a naive-memory pool that
   contains T[SCM], T[MNP], and T[Nrev] populations. We also show that the
   chromatin structure of naive-revertant cells is substantially
   reorganized in comparison with the naive pool and as such they are
   epigenetically “primed for secondary activation. These observations
   will help to guide studies of fundamental mechanisms that regulate
   T cell differentiation and should also be of considerable value for
   optimal generation of naive-memory cells for adoptive T cell
   immunotherapy as less differentiated cells have been associated with
   superior engraftment, persistence, and antitumor activity
   ([187]Hinrichs et al., 2011, [188]Klebanoff et al., 2011).

Limitations of the Study

   A few limitations should be considered when interpreting our data.
   Although we identified cells with the phenotype of T[Nrev] cells in
   human tonsil, it will be of interest to assess further if these are
   generated directly in vivo. In addition, it will be important to pursue
   parallel studies within animal models to interrogate potential
   mechanisms of phenotypic reversion.

Methods

   All methods can be found in the accompanying [189]Transparent Methods
   supplemental file.

Acknowledgments