Extrinsic and intrinsic drivers of natural killer cell clonality

Clonal expansion of antigen‐specific lymphocytes is the fundamental mechanism enabling potent adaptive immune responses and the generation of immune memory. Accompanied by pronounced epigenetic remodeling, the massive proliferation of individual cells generates a critical mass of effectors for the control of acute infections, as well as a pool of memory cells protecting against future pathogen encounters. Classically associated with the adaptive immune system, recent work has demonstrated that innate immune memory to human cytomegalovirus (CMV) infection is stably maintained as large clonal expansions of natural killer (NK) cells, raising questions on the mechanisms for clonal selection and expansion in the absence of re‐arranged antigen receptors. Here, we discuss clonal NK cell memory in the context of the mechanisms underlying clonal competition of adaptive lymphocytes and propose alternative selection mechanisms that might decide on the clonal success of their innate counterparts. We propose that the integration of external cues with cell‐intrinsic sources of heterogeneity, such as variegated receptor expression, transcriptional states, and somatic variants, compose a bottleneck for clonal selection, contributing to the large size of memory NK cell clones.


| Diversification of receptor repertoires as a substrate for clonal selection
[16] Consequently, the vast diversity of constantly evolving pathogens needs to be met by diverse immune receptors ready to mount an immune response to a multitude of different antigens.8][19][20] Complemented by the non-templated addition of nucleotides to create additional junctional diversity, 21,22 this process produces a large breadth of specificities.Positive and negative selection shape the T-cell receptor repertoire to high enough affinities to scan for foreign peptide antigens on the major histocompatibility complex, while at the same time maintaining selftolerance, generating specificities to an approximate average of 100,000 CD8 and 1,000,000 CD4 T-cell epitopes. 23 innate lymphocytes, NK cells lack re-arranged antigen receptors and instead integrate different signals via a range of different germline encoded receptors, which are expressed by much larger populations of cells. 24Virtually, all NK cells can sense cytokines such as Interleukin-15 (IL-15), IL-12, or type I interferons (IFN) in their crosstalk with myeloid and stromal cells releasing these inflammatory signals. 25,26Cellular targets opsonized with antibodies can also be recognized by the majority of the NK cell pool via the low-affinity IgG receptor CD16, inducing antibody-dependent cytotoxicity. 27,280][31] Conversely, diversification mechanisms generate variegated expression patterns for some NK cell receptors, resulting in heterogeneous responsiveness across the NK cell population.This stochastic receptor (co-)expression is the fundamental mechanism generating a diverse repertoire of NK cells with different combinatorial specificities.The probably most closely studied receptor family in this regard are the inhibitory KIRs and their functional orthologs from the Ly49 family.Importantly, both KIRs and Ly49s display specificity to certain motifs within groups of MHC class I alleles that enable the "missing-self" response in which NK cells react to MHC class I downregulation, for example, in the context of viral infections or malignant transformation. 32,335][36][37][38][39][40][41] We will discuss below, how random monoallelic expression contributes to these stable, variegated expression profiles.Although achieved by different mechanisms, the diversification of specificities is an important parallel between T and NK cells, providing an anticipatory pool of cells able to respond to a wide range of different pathogens, while at the same time resulting in a restricted pool of cells able to respond to each individual pathogen.Despite this conceptual similarity, individual NK cell receptors are expressed at frequencies that are several orders of magnitude higher than those of antigen-specific TCRs, and most NK cell receptors recognize self-ligands upregulated during infection or cell stress and, therefore, are not truly pathogenor antigen-specific.However, several examples of direct recognition of pathogen-derived structures have been described and previously reviewed, 24 with the most prominent example being CMV.

| Antigen-specific NK cell memory to CMV
The specific recognition of CMV by murine NK cells and their role in controlling the infection was initially based on the differential susceptibility of different laboratory mouse strains to MCMV infection.
3][44][45][46] The MCMV protein m157, which has an MHC-like protein fold, serves as the specific ligand for Ly49H on the surface of infected cells, inducing NK cell cytotoxicity and cytokine release. 47,48[51] Early viral control by Ly49H + NK cells is associated with their preferential activation and proliferation, expanding their frequencies and absolute numbers during acute MCMV infection. 50,51Elegant transfer studies utilizing Tyrobp −/− hosts unable to mount endogenous Ly49H + NK cell responses demonstrated the robustness of Ly49H + NK cell expansion from a relatively small number of naïve progenitors.In this setting, a small but significant pool of memory cells persists with a terminally mature phenotype marked by expression of KLRG1 and Ly6C. 7Ly49H + memory NK cells (also referred to as "adaptive") are more readily activated and produce higher levels of IFNγ upon stimulation via activating receptors, while being less responsive to bystander activation via pro-inflammatory cytokines or heterologous infection with Listeria monocytogenes, underlining their functional remodeling. 7,52,53Upon transfer, memory NK cells provide enhanced protection against MCMV infection, likely driven by their more pronounced effector functions, as their expansion kinetics are comparable to naïve Ly49H + NK cells. 7,54 humans, observations of NK cell receptor profile imprinting by HCMV infection were suggestive of a similar memory NK cell response. 6Many of the defining phenotypic features of human memory NK cells were already included in these initial observations, including increased levels of the activating heterodimeric receptor CD94/NKG2C, lower co-expression of its inhibitory counterpart CD94/NKG2A and the natural cytotoxicity receptors NKp30 and NKp46, as well as an enrichment of KIR + and ILT2 + cells within the NKG2C + population of HCMV + individuals. 6Key studies following acute HCMV infection after hematopoietic stem cell transplantation (HSCT) demonstrated that NKG2C + NK cells co-expressing the terminal maturation marker CD57 + expand during acute HCMV infection and persist for months after viremia. 55,56Similar dynamic expansion was observed after kidney transplantation. 57Complexes of CD94 with NKG2A and NKG2C both recognize the non-classical MHC class I molecule HLA-E loaded with peptides.2][63] However, these complexes of HLA-E and gpUL40 peptides can be specifically recognized by NKG2C + NK cells, inducing their activation and expansion in a sequence-specific manner. 64Similar to the murine system, human memory NK cells are efficient producers of IFNγ and TNF when triggered via activating receptors, whereas their effector functions in response to cytokines are reduced due to their downregulation of cytokine receptor components. 65Demethylation of the regulatory region CNS1 in the proximity of the IFNG gene is one mechanism likely enabling such potent production of this effector cytokine. 668][69][70] As will be discussed below, activation by the pro-inflammatory cytokines IL-12 and IL-18 plays an important role in this remodeling, which in combination with IL-15 can recapitulate many aspects of the CMVinduced remodeling in vitro. 64,67Accordingly, a large fraction of remodeled chromatin regions is bound by the IL-12-induced transcription factor (TF) STAT4, and its genetic deletion disrupts the epigenetic remodeling during MCMV infection. 69Enrichment of AP-1 family TFs in memory NK cell-associated open chromatin further suggests a role for this transcription factor family in HCMV-induced imprinting, 9,69,71 an interesting parallel with memory phenotypes of adaptive lymphocytes, as well as immune training of different immune and non-immune cell types. 72 sum, memory NK cells induced by CMV are characterized by their mature phenotype and high functionality, achieved by pronounced transcriptional and epigenetic remodeling.Specific recognition of CMV-derived ligands via unrelated human and murine receptors is an interesting parallel between these species that explains this unique interaction.

| Clonal NK cell responses to CMV
Clonal selection as the basis for CMV-specific NK cell memory was originally postulated based on their focused receptor expression profiles, especially of the MHC class I specific receptors from the Ly49 and KIR families in mice 54 and humans, 73,74 respectively.However, their lack of uniquely re-arranged antigen receptors precluded genetic analysis as a gold-standard test for their clonal origin.A recent study elegantly addressed the ability of Ly49H + NK cells to undergo clonal expansion by retroviral barcoding with combinations of fluorescent proteins. 8Such barcoded Ly49H + NK cells expanded drastically upon transfer and MCMV infection of immunodeficient Rag2 −/− Il2rg −/− recipients, with some clones surpassing progeny of 10,000 cells at the peak of expansion.How this clonal progeny evolves in time and whether such drastic expansion of individual clones also occurs during natural infection of wildtype mice with higher naïve Ly49H + progenitor remains to be tested.
6][77] Increased clonal size in RhCMV + animals implied the involvement of RhCMV, although the NK cell compartment in RhCMV − macaques also contained sizeable clones, especially early after transplantation, pointing toward a potential engraftment bias.Nevertheless, an intriguing segregation of clones by KIR profiles and expression of KLRC2 transcripts by expanding cells in RhCMV + monkeys suggest that some of these clones might represent memory NK cells.Indeed, as the RhCMV protein Rh67 is a functional homolog to the HCMV gpUL40 protein containing an MHC-E-stabilizing peptide, 78,79 it is tempting to speculate that the recognition mechanism might be conserved.Experimental infection with RhCMV initially induced polyclonal expansion, with discernible clonal shifts occurring around 2 months after infection. 77e to the lack of specific reagents for macacine NKG2C, it has so far remained challenging to specifically define RhCMV-induced memory NK cell clones, but ongoing studies combining barcoding with single-cell RNA sequencing (scRNA-seq) 80 and novel antibodies specifically designed for macaque antigens hold promise to resolve these issues. 81 humans, we recently revisited the hypothesis of NK cell clonal expansion and persistence utilizing mitochondrial mutations as cellendogenous barcodes. 9The mitochondrial genome exists in many copies (100-1000s) per cell and has a higher mutation rate than F I G U R E 1 The memory NK cell compartment is composed of diverse clonal identities.NK cells of an exemplary HCMV + donor were analyzed by scATAC-(A) and scRNA-seq (B), combined with cell surface phenotyping using nucleotide-labeled antibodies and quantification of mitochondrial variants (C). 9 Dimensionality reductions are based on chromatin accessibility and transcriptomes and are overlaid with the surface levels of memory NK cell-defining markers NKp30 and NKG2C, as well as self-MHC-specific KIR2DL1/S1/S3/5 and KIR3DL1.These analyses reveal several subpopulations of memory NK cells distinguished by chromatin accessibility, transcriptomes, and cell surface phenotypes.Selected NKp30 − NKG2C + subpopulations expressing KIR2DL1/S1/S3/5 (red) or KIR3DL1 (blue) in a mutually exclusive fashion correspond to abundant clonotypes defined by mitochondrial mutations.These clonally enriched subpopulations can also be identified by flow cytometry (D).
nuclear DNA; as mutations often reach high levels of heteroplasmy (proportion of mitochondrial genomes containing a specific mutation) by vegetative segregation, random genetic drift, and relaxed replication, 82 it is an ideal substrate for single-cell genotyping for the reconstruction of clonality relationships. 83These mutations can be detected in a modified single-cell assay for transposase accessibility by sequencing (scATAC-seq) that captures chromatin accessibility together with high-throughput sequencing of the mitochondrial genome. 83,84Mutations can be called with a specialized analysis pipeline ("mgatk") that accounts for systematic errors by considering the strand-concordance of heteroplasmy and variance over cells when identifying high-confidence variants.Using this approach, we found that NKG2C + memory NK cell populations in healthy HCMV + individuals contained massively expanded clones identified by individual mitochondrial mutations. 9Donors had varying numbers of memory NK cell subpopulations consisting of one or several clones.While these all shared a characteristic epigenetic signature defining memory NK cells, clones were distinguished by pronounced differences in chromatin accessibility, partially reflected in gene expression and cell surface phenotypes (Figure 1).Consistent with the observation of stable memory NK cell frequencies based on KIR profiles, 73 memory NK cells were clonally stable in their specific epigenetic states in longitudinal follow-ups of up to almost 2 years, suggesting they are self-sufficient in their maintenance. 9An observation that was surprising to us is their remarkable clonal size, which for some was in the range of several percent of the memory NK cell population.It should be noted that due to the many copies of the mitochondrial genome per cell and the lack of information on when mitochondrial mutations were acquired and drifted to high enough heteroplasmy to be reliably detected and identify clonotypes, the precision of the clonal size estimates is limited at the upper limit; with the relatively strict approach that we applied to define clonotypes, almost only those carrying homoplasmic mitochondrial mutations passed filters for downstream analysis.Therefore, our results only define the minimal size of each clone, which might actually be even bigger.These massive expansions are on par with the size of the largest CD8 T-cell clones from other studies, typically also identified in the context of HCMV infection, [85][86][87] underlining its capacity to induce large antigen-specific expansions in the adaptive but also the innate immune system.

| Clonal selection by antigen-specific receptors is highly efficient
The recruitment of naïve T cells is highly efficient, as has been shown in different infection models spanning a range of infection doses. 1 Consistently, the early stages of CD4 T-cell responses are characterized by high TCR diversity. 88Additionally, as long as the activation threshold is overcome, the efficiency of CD8 T-cell recruitment is largely independent of TCR affinity. 89Hence, the initial selection of T cells is mainly dependent on the respective precursor frequency in the T-cell repertoire, which for the average peptide-MHC complex is estimated to encompass 1 in 100,000 CD8 + and 1 in 1,000,000 CD4 + antigen-specific T cells, representing an important clonal bottleneck. 23,90,91 if antigen-specific precursor frequency is the main bottleneck for the initial recruitment of T cells into the response, how does this apply to NK cell responses to CMV? Ly49H + cells transferred into Tyrobp −/− hosts showed near complete recruitment as judged by CFSE dilution. 7Accordingly, expansion amplitudes scale with precursor frequency, indicative of highly efficient recruitment of transferred cells in this setting where they face little competition.Upon transfer into C57BL/6 mice, the expansion amplitude of the transferred cells is strongly reduced, suggesting competition with larger numbers of endogenous NK cells limits their expansion. 7However, rapid BrdU labeling of a high proportion of endogenous Ly49H + cells in MCMV-infected C57BL/6 mice suggests that the competition does not restrict recruitment, 51 similar to what has been observed for T cells. 89Given that approximately half of all NK cells in C57BL/6 mice express Ly49H, 51 precursor frequencies are hardly restricted by the main antigen-specific receptor.Although NKG2C + NK cell frequencies in HCMV − individuals are typically around 5%-10%, and therefore considerably lower than naïve Ly49H + NK cell frequecies, 92 they are still vastly more abundant than antigen-specific precursors in the T-cell compartment.However, for both, T cells and NK cells, engagement of their antigen receptors is not the only requirement for efficient clonal expansion.Instead, the integration of various signals plays a pivotal role in shaping fate decisions within antigen-specific lymphocyte responses.This underscores that heterogeneous exposure to these signals and varying responsiveness are critical determinants influencing clonal expansion.

| Additive signal strength modulates clonal expansion
An important feature of memory T-cell populations is their overrepresentation of TCRs with high antigen affinity compared to the naïve compartment.Especially upon repeated antigen exposure, the TCR repertoire of antigen-specific cells narrows more and more toward high-affinity receptors. 93As the initial recruitment of naïve T cells is unaffected by TCR affinity beyond the activation threshold, the control of clonal output in response to signal strength is achieved by tuning the magnitude and duration of clonal expansion. 89 Wild isolates of MCMV carry different m157 variants of which only some are recognized by Ly49H from C57BL/6. 95In humans, we have shown that variations in the binding affinity between NKG2C and the HLA-E-restricted polymorphic gpUL40-derived HCMV peptides lead to significant disparities in the recognition of HCMV-infected cells. 64,96This differential recognition modulates effector functions of memory NK cells upon restimulation and the degree of proliferation of NKG2C + NK cells from HCMV − individuals. 64Mathematical modeling of in vitro culture kinetics suggests that signals received via NKG2C accelerate division times and reduce death rates, resulting in the accumulation of NKG2C + NK cells.Accordingly, we observed pronounced expansion of NKG2C + NK cells with a memory phenotype in two patients reactivating HCMV strains that carried the most efficiently activating peptide variant.Based on these findings, we propose that a part of the wide variation in NKG2C + memory NK cell frequencies between HCMV + individuals might be driven by the gpUL40 peptide sequences of the infecting HCMV variants.Whether this has an effect on clonal diversity and size is an interesting question that still needs to be addressed.Additionally, engagement of the inhibitory NKG2A is another important factor influencing signal 1, as responses of NKG2A + NKG2C + cells to HLA-E + target cells are dominantly inhibited. 97Accordingly, NKG2C + memory NK cells are usually NKG2A − , 6 although exceptions have been observed. 9,65,73Moreover, recent studies indicate that heterogeneity in expression levels of the CMV-specific receptors impacts how Signal 1 is perceived.Increased per-cell expression of Ly49H and NKG2C on murine and human memory NK cells, respectively, suggest the preferential expansion of cells expressing high levels of the receptor. 6,8,92,98This concept is further supported by correlations of population and clonal size with receptor levels. 8,98Indeed, transfer of Ly49H + NK cells sorted for high and low expression revealed that although both subsets could undergo expansion upon MCMV infection, the magnitude of the response was higher for the Ly49 hi cells, even in a non-competitive setting. 98These findings suggest that increased avidity to CMV-antigens achieved by high receptor expression gives NK cells a competitive advantage.
Interestingly, NKG2C − and Ly49H − memory NK cell expansions have been described, suggesting other activating receptors might be able to compensate for its absence. 53,73,99,100Activating KIRs have been discussed to replace NKG2C, as they were found to be expressed by NKG2C − expansions from healthy individuals and after HCMV reactivation in patients undergoing HSCT with cord blood from NKG2Cdeficient donors. 73,100However, in another study, several individuals with KIR haplotype A/A, who lack activating KIRs genetically, displayed NKG2C − memory NK cell expansions, suggesting activating KIRs are not strictly required for their emergence. 99 Since DNAM-1 expression is dynamically regulated, it remains unclear whether the predominance of DNAM-1 − cells in the memory phase is due to changes in expression or preferential survival, although increased apoptosis of DNAM-1 + cells at days 3 and 6 postinfection supports the latter hypothesis. 1014][105] In humans, CD2 is highly expressed by most memory NK cells and has been demonstrated to support NK cell activation when combined with engagement of NKG2C or CD16. 64,99Consequently, we could show that CD2 supports the expansion of NKG2C + NK cells in vitro, especially in response to low-affinity peptides. 64Mechanistically, CD2 feeds into different signaling pathways, enhancing NFAT, AP-1, and NF-kB activation. 106-109However, we observed that at least strong stimulation via NKG2C combined with IL-15 is sufficient to induce the specific proliferation of NKG2C + cells without CD2 engagement, 64 and have found rare cases of CD2 − memory NK cell populations (unpublished).Hence, while CD2 might be necessary to overcome the activation threshold and thereby effectively restrict the responsive naïve population in situations of low activation by other signals, in other cases, it might rather act to support expansion.Also for receptors mediating signal 2, quantitative differences in expression can influence clonal competition, as selective expansion of T cells with higher CD28 expression has been described. 110ile the high expression of CD2 by memory NK cells supports such a mechanism, experiments assessing expansion capacity in relation to receptor expression levels remain to be performed.
In addition, pro-inflammatory cytokines acting as Signal 3 are essentially required for the expansion and differentiation of memory NK cells.Especially IL-12 and IL-18 are important in the early phases of the response. 25While NK cells lacking the IL-12 receptor develop normally and efficiently kill MHC class I-deficient or m157expressing target cells, they fail to expand and mature in response to MCMV infection. 111Consistent with IL-12 activation of STAT4, the effect of IL-12 receptor deficiency is largely phenocopied by NK cells from Stat4 −/− mice.Similarly, IL-18 receptor-deficient NK cells have a competitive disadvantage in mixed bone marrow chimeras infected with MCMV. 112However, neither their survival nor their activity during a recall response are affected by IL-18 receptor deficiency, suggesting a stage-specific requirement for IL-18 during the primary response, in contrast to the complete inability of IL-12 receptordeficient NK cells to form memory. 111,112 Accordingly, genetic deletion of the main signaling adaptor Myd88 recapitulates the effects of IL-18 receptor deficiency, 112 although this might be partially due to interrupted IL-33 signaling. 113A secondary effect has been attributed to type I IFN, as NK cells from Ifnar1 −/− mice do not have a proliferation deficit.Instead, increased apoptosis severely stunts their ability to expand in numbers, with evidence for increased upregulation of NKG2D-ligands on Ifnar1 −/− NK cells leading to killing by other NK cells. 114Deficiency of STAT1 phenocopies this effect, consistent with the role of STAT1 homo-and STAT1/STAT2 heterodimers in type I IFN signaling. 115As the other ISGF3 complex components STAT2 and IRF9 are independently required, canonical type I IFN signaling seems to be the basis of an auto-regulatory feedforward loop to support NK cell survival. 116Together, the strict dependency of NK cell memory formation on pro-inflammatory cytokines suggests differential exposure, as well as differences in the intrinsic ability to respond to these cytokines might affect clonal selection and expansion.1][122][123][124][125] Hence, future studies might show whether these subsets might be preferentially recruited into effector and memory differentiation.Since stimulation with cytokines alone has been demonstrated to recapitulate some of the aspects of NK cell memory, such as enhanced IFNγ production even weeks after the initial stimulus, and is propagated across several rounds of proliferation, 117,126,127  has been suggested to enable their accumulation in transgenic mice, 128 which would be consistent with pro-survival signals mediated by phosphatidylinositol-3-kinase (PI3K) and AKT downstream of KIRs. 129This survival effect was dependent on the presence of the cognate HLA class I ligand. 128Consistently, HCMV + patients with transporter associated with antigen presentation (TAP) deficiency can present with large NKG2C + NK cell expansions that share many memory phenotype characteristics, except skewing of inhibitory KIR profiles, 130 due to drastically reduced expression of HLA class I. 131,132 While this finding was interpreted as polyclonality, it will be crucial to reassess these findings with novel methods that enable clonal resolution without relying on KIR expression profiles, potentially providing important insights into the role of KIRs in regulating clonal selection and expansion.Key studies from Lewis Lanier's laboratory addressed a similar question in mice by comparing Ly49H + cells co-expressing MHC-specific receptors in different genetic contexts with or without the respective cognate ligands. 54,133Curiously, unlicensed NK cells lacking inhibitory self-MHC-specific Ly49C/I in C57BL/6 mice proliferate more vigorously during the acute MCMV response than their Ly49C/I + counterparts, presumably due to persistent inhibition by the cognate ligands. 133In addition, the expression of inhibitory Ly49A results in enhanced apoptosis selectively in B10.D2 mice where its cognate ligand is present. 54 These observations are at odds with the predominant expression of inhibitory self-MHC-specific KIRs on human memory NK cells. 73,74 our knowledge, the evolution of Ly49 or KIR expression on CMVspecific NK cells has not been monitored closely over longer observation periods encompassing different phases of the response, which leaves room for speculation on whether the expression of (Figure 3).Based on the recognition mechanisms and skewed receptor profiles of memory NK cells described to date, especially expression of NKG2C and NKG2A, as well as CD2, and self-MHC-specific KIRs might restrict the initial recruitment.Based on these receptors, the potentially responsive NKG2C + NKG2A − self-KIR + CD2 + subset encompasses about 1% of the NK cell population in HCMV − individuals, which show average frequencies of ~5% NKG2C + NKG2A − NK cells. 92About 30% of these cells express at least one self-MHCspecific KIR 73 and 80% CD2. 99Although other receptors like ILT2 or SIGLEC7, whose expression is recurrently skewed on memory NK cells, 65,100 might add to this, the frequency of naïve cells equipped with a suitable receptor co-expression profile for the recognition of UL40-peptide-HLA-E complexes in infected cells is likely several orders of magnitudes higher than the frequency of epitope-specific T cells.Under the assumption that recruitment of NK cells is similarly efficient as has been reported for antigen-specific T cells in systemic infections, 1 even a conservative estimate of 0.01%-0.1% of potentially responsive NK cells would equate to 90,000-900,000 cells in the blood of an average healthy adult (assuming an average NK cell count of 180 cells/μL 134 and 5 L of blood volume), with many additional cells present in the bone marrow and other tissues.Therefore, clones with a frequency >1% within the memory NK cell compartment are more than 1000 times bigger than what would be expected if there was equal contribution from all potentially responsive cells.
Hence, the limited frequency of cells equipped with a receptor repertoire optimal for responding to the signals received during CMV infection likely represents a decisive, but not the only bottleneck restricting clonal selection and expansion of NK cells.

| Signal integration connects expansion and differentiation
Clonal expansion is inseparably intertwined with the differentiation of antigen-specific cells into effectors and different memory subsets.Different signals are integrated to connect clonal expansion and differentiation (Figure 2).On the one hand, the signals received during an immune response sum up linearly in the cellular calculus to set the number of divisions the cell and its progeny perform before returning to quiescence. 110,135In T cells, this "division destiny" is associated with the amounts of the cell cycle regulator Myc, which is induced proportionally to the sum of signals. 136As the degradation rate of Myc is largely stable over time and independent of signal strength, the modulation of its production rate by signal strength establishes a dynamic equilibrium inherited across cell divisions and exerts temporal control over proliferation.
Consequently, the proliferative behavior of cells belonging to one clonal family is largely concordant and synchronous early after activation, driven by the different signals integrated independently by the founder cell. 110On the other hand, an important hub for signal integration leading to global transcriptional and epigenetic remodeling is composed of interferon regulatory factors (IRFs), activator protein 1 (AP-1), and other members of the basic region, leucine zipper (bZIP) TF family.While the originally described AP-1 FOS:JUN heterodimer serves important functions during the immediate transcriptional regulation of different T-cell effector genes, as closely studied for Il2 137(p199) , other bZIP TFs, namely BATF and BATF3, are crucial drivers of effector and memory differentiation, as well as T helper (T H ) cell polarization.Especially T H 17 differentiation has been closely studied in this context, as IL-17 production strictly depends on BATF. 138,139The leucine zipper domain of BATF family members endows them with the unique ability to form complexes with IRF4 and IRF8. 140,1413][144][145] IRF4-BATFcomplexes assemble on AP-1-IRF composite elements (AICEs) to induce the expression of polarization-associated genes. 140,141,146portantly, these complexes assemble not only under T H 17 but also under T H 2-polarizing and even under non-polarizing T H 0 conditions, 147 and are similarly involved in driving CD8 T-cell effector responses, 145,148 suggesting a near-universal role in effector cell differentiation.Indeed, a similar function of IRF4 for integrating signals has recently been demonstrated in NK cells, promoting expansion and differentiation in response to CMV infection. 71F4 is induced after MCMV infection and binds to IRF, AP-1, and AICE motifs, under the combined influence of pro-inflammatory cytokines and activating receptor stimulation.In turn, expansion and transcriptional remodeling are blunted in IRF4-deficient mice.
We observed similar cooperativity between activation via NKG2C and pro-inflammatory cytokines in humans, 9,71 consistent with the dependency on both signals for the specific expansion of NKG2C + NK cells. 64Interestingly, in this setting, activation via NKG2C in the presence of IL-15 but without IL-12 and IL-18 potently induced IRF and AP-1 TF activity. 9However, global remodeling and AP-1 TF activity were further enhanced by combined stimulation, again underlining the requirement of both signals for efficient epigenetic remodeling.While the role of BATF/BATF3 in effector and memory differentiation of NK cells has not yet been addressed, it seems likely that these are among the AP-1 TFs partnering with IRF4.
The impact of these signals on T-cell differentiation has been studied in much detail, as reviewed elsewhere. 149Briefly, increased combined initial signal strength and duration induce larger populations of short-lived effector cells (SLECs). 150,1513][154][155] IL-12 and type I IFN drive differentiation into SLECs, which proliferate more vigorously during the acute phase than memory precursor cells (MPECs). 152,156,157Consistently, despite the reduced expansion of effector T cells in mice lacking IL-12, its absence is, in fact, associated with more pronounced memory responses, 157 highlighting that the generation of memory depends not only on the amplitude of acute clonal expansion but also on the relative differentiation of expanding cells.A recent study suggests a similar fate commitment occurs for subsets of Ly49H + NK cells at the acute phase of MCMV infection. 158Lack of Ly6C at the peak of infection distinguishes a population that is more likely to survive through contraction.The authors propose the transcription factor Fli1 as a possible regulator of these alternative fates, although it seems to have a generally negative effect on the survival of both subsets, with the survival of the Ly6C − subset being slightly more enhanced by its genetic ablation.clonal size correlates strikingly with the differentiation state, underlining that clonal expansion and differentiation are interdependent, coupled processes. 159,160Stochastic dynamic models integrating quantitative data on clonal size and differentiation states suggest that naïve T cells differentiate linearly into central memory (T CM ) precursors, which can further give rise to effector memory (T EM ) precursors and, finally, effectors.As high proliferative activity is confined to the later differentiation states, the transition timing into higher proliferative states can be a major driver of differences in clonal progeny size. 159Other factors, such as the time until first division, seem to play a subordinate role based on mathematical modeling of proliferation kinetics, further supported by the observation that also "late-comer" T cells, transferred 48 h after infection, show similarly disparate behavior. 160Although recent studies have suggested more nuanced differentiation models, for example, with plasticity between subsets 163 and differences in proliferation rates between subsets over time, 164  T cells divide fast with relatively little variation. 165,166Later, some cells acquire a T CM precursor phenotype and switch to slower proliferation or quiescence.[167] An elegant study following the fate of clones marked by a lentiviral library encoding fluorescent proteins recently demonstrated that a similar coupling between clonal expansion and differentiation applies to NK cells during MCMV infection. 8More specifically, similarly heterogenous clonal burst sizes were correlated inversely with the expression of CD27, which marks immature NK cells. 168,169While these findings mirror the differential expansion kinetics of CD8 Tcell subsets, a recent pre-print points out that the more efficient expansion of the immature CD27 + NK cell subset is an important difference to the T-cell compartment, where the correlation between clonal size and differentiation can be explained by the more extensive proliferation of the mature subsets. 170By fitting stochastic models for population dynamics to clonal NK cell data generated in a similar setting, 171 the authors develop a 3-stage model in which CD27 + cells give rise to an intermediate CD27 − NK cell subset with a high proliferation rate, which differentiates into a terminal population with a high death rate. 170They propose that the mature CD27 − subsets might be distinguished by expression of Ly6C, as recently reported in the contraction phase, 158 and indeed recapitulate the inverse correlation between CD27 and Ly6C observed in experiments. 8Future studies that test these predictions experimentally, ideally combining clonal readouts with phenotypic analyses in a kinetic fashion, might provide important insights into the differentiation trajectories, expansion, and memory potential of NK cell subsets generated during acute responses.

| Stochastic dynamics of clonal expansion
Importantly, the clonal evolution of antigen-specific T-cell populations does not stop once the acute infection is resolved.
3][174] Consequently, clonal size during primary expansion is hardly predictive of the magnitude of re-expansion upon secondary challenge, whereas burst sizes of secondary and tertiary responses are highly concordant, reflecting an intrinsic difference in the ability of clonal families to re-expand. 1596][177] Novel fate-map reporter mice marking proliferative history suggest further heterogeneity within the T CM compartment, identifying a subset that exits proliferation early during the primary response and strongly contributes to reexpansion upon secondary stimulation. 164These differences in reexpansion capacity are of particular importance in the context of CMV infection, which, as a persistent virus, provides a persistent source of antigen, leading to the long-term clonal evolution of the antigen-specific repertoire. 175,178As a result, the CMV-specific CD8 memory T-cell compartment is characterized by the drastic expansion of highly functional memory T cells with a terminally mature phenotype. 4,179The progressive accumulation of MCMVspecific T cells has been studied closely in mice, where this process has been termed "memory inflation". 180,1815][186] Based on observations of proliferating T CM in lymph nodes, it has been suggested that recurrently stimulated T CM feed this pool of inflationary T cells. 186Along these lines, a recent study demonstrated that the frequency of T CM precursors within a clonal family during acute MCMV infection is a strong predictor for the magnitude of its memory inflation. 175Whether similar mechanisms shape the memory NK cell compartment in time remains largely unknown.While we have observed stable maintenance of NK cell clones in humans for up to almost 2 years, 9 our cohort contained only healthy adult individuals who were likely infected by HCMV years before the study, so we probably missed the most dynamic initial phase early after primary infection. 178187 Overall, these studies provide an exciting window into how the highly heterogeneous behavior of individual cells can produce robust responses on the population level.The quantitative nature of the studies on T-cell responses was decisive for drawing conclusions on how differentiation and proliferation are connected and which parameters have the strongest effects on clonal outcomes.Now that we have the tools to follow the clonal composition of NK cells, future studies might more closely assess their dynamics to develop a quantitative understanding of expansion kinetics and division of labor between short-term effector functions and memory formation.
Finally, it still remains largely open how the heterogeneous, seemingly stochastic behavior of cells sharing the same specificity comes about.Differences in signal strength due to stochastic encounters between lymphocytes and antigen-presenting cells or exposure to soluble factors certainly play a role in the differential programming of clonal family fate in vivo.However, even under optimal activation conditions in vitro, heterogeneity of expanding T clones persists. 166terestingly, in these settings, most variability in proliferation speed is attributed to differences within clonal families rather than differences between separate clones.Nevertheless, differences between clonal families still contribute to the heterogeneity of the expanding T-cell population as a whole, suggesting cell-intrinsic features that are heterogeneously distributed might have deterministic effects on clonal expansion.We will discuss possible sources of such heterogeneity in the next section.

| Sources and biases of somatic mutations in the hematopoietic system
Somatic variants are mutations acquired by individual cells that can be transmitted to their daughter cells, resulting in genetic mosaicism.
The mutational spectrum of different tissues, cancer entities, and cell types is characterized by different features such as the class of mutation, sequence context, or transcriptional strand bias.9][190] In the hematopoietic system, different mutational processes acting during development, differentiation, and acute immune responses shape the mutational load of the mature cell types.[193][194][195][196][197] Accordingly, the mutational load of HSPCs correlates strongly with age, with ~16 mutations being accumulated per cell per year.These processes continue as HSPCs differentiate into more committed progenitors and ultimately mature immune cells.Interestingly, within the lymphocyte family, clock-like mutations accumulate at a similar rate in B cells as in HSPCs, whereas this process is slightly accelerated in T cells, pointing toward cell-type specific mechanisms modulating genetic integrity. 195Notably, the mutational load is much more variable between cells from one individual than between individuals, suggesting cellular lifestyle has a more important impact than the individual's or simple clock-like processes.Further, mutational processes specifically related to the biology of T and B cells introduce additional mutations that are absent in non-lymphoid lineages.T and B cells undergo programmed somatic mutational processes mediated by recombinase-activating genes (RAG) to recombine antigen receptors, while B cells gain further somatic genetic alterations induced by cytidine deaminase (AID), resulting in somatic hypermutation (SHM) and class-switch recombination of immunoglobulin loci as part of in the germinal center reaction.Due to its ability to induce double-strand breaks, off-target activity of the RAG complex can result in structural variants typically in the context of sequences with sufficient similarity to the recombination signal sequence (RSS), and frequently co-occurring with non-templated sequences at breakpoints, indicative of TdT activity. 195Although innate lymphocytes lack re-arranged antigen receptors by definition, fate-mapping studies in mice have revealed that RAG is already expressed by a large fraction of common lymphoid progenitors (CLP), which not only differentiate into T and B cells but also innate lymphocytes and to a certain extent dendritic cells (DCs). 198,199Hence, RAG activity might also shape the genome of innate immune cells, as evidenced by the frequent presence of non-productive rearrangements within the Ig and TCR loci of NK cells and ILCs, 198,[200][201][202][203][204][205][206] which has been suggested to improve the ability of NK cells to withstand genetic stress induced by extensive proliferation. 200ile mutational signatures are proxies of the underlying mutational processes, their genomic localization correlates with features such as genic organization, replication time and direction, transcriptional activity, chromatin compaction, or its association with nucleosomes and transcription factors. 207Globally, mutations are more likely to occur in less accessible chromatin carrying histone marks of heterochromatin, 208,209 presumably due to the lower activity of different repair mechanisms at non-accessible sites. 210Similarly, the local distribution of mutations is affected by nucleosome and transcription factor occupancy, again reflecting more efficient repair of accessible regions, but also the propensity of different mutational processes to attack linker and nucleosome regions.For example, cytosine to thymidine transversions, which predominate the mutational spectrum in hematopoietic cells, are significantly depleted from nucleosome-bound regions, likely because of conformational constraints that prevent the spontaneous deamination of cytosines in transiently opened single-stranded DNA. 211As a result of these regional biases, the genomic distribution of mutations found in tumor cells years after neoplastic transformation maps back accurately to the chromatin accessibility and decoration of their most likely cell of origin. 208Applying this concept to non-malignant lymphocytes highlights important differences in their lifestyle. 195Clock-like mutations found in naïve B cells correlate best with the epigenomes of HSCs, consistent with the ongoing production of naïve B cells throughout life.In contrast, the distribution of such mutations in naïve T cells better matches their respective epigenomes, reflecting their generation in the thymus mainly early in life and their continuous acquisition of mutations as mature cells.Further, the location of mutations in both memory B and T cells is best predicted by corresponding effector/memory epigenomes.Similarly, germinal center epigenomes were most predictive of the location of B cell mutations with a characteristic somatic hypermutation-associated signature, again matching our understanding of the underlying biological process.Hence, applying such an analysis to innate lymphocytes could be a promising approach to infer their lifestyle, turnover, and dwelling time in different differentiation stages.

| Selection of somatic variants manifests in hematological conditions
[196] While the role of somatic mutations in driving the clonal proliferation of cancer cells is well established, 212,213 recently, it has become increasingly clear that also non-malignant tissues can be shaped by the genetic evolution of their cellular constituents. 2140][221][222] Clonal hematopoiesis is commonly associated with myeloid lineage bias and predictive for myeloid malignancies, consistent with the frequent detection of recurring somatic mutations associated with hematological malignancies.[221][222][223][224] Recently, genes recurrently mutated in lymphoid malignancies were found to be carried together with mosaic chromosomal alterations by a fraction of healthy individuals and to be predictive for the development of cancers of lymphoid origin, suggesting lymphoid clonal hematopoiesis exists as a separate entity. 224lection pressures on the genetic makeup of lymphocytesbesides their re-arranged antigen receptors-are only beginning to be considered as an influence on their population dynamics.That such pressures act on non-malignant lymphocytes was probably first appreciated when different autoimmune and lymphoproliferative disorders were genetically characterized. 225One prominent example is autoimmune lymphoproliferative syndrome (ALPS), of which a subtype was originally connected to germline mutations of the FAS cellular death pathway that allow for the massive expansion of CD4 − CD8 − T cells. 226,2279][230] Additionally, incomplete penetrance of ALPS in individuals with inherited heterozygous TNFRSF6 (encoding FAS) mutations can at least partially be explained by stochastic loss of heterozygosity or acquisition of mutations in the second allele. 231For ALPS caused by somatic events, different distributions of TNFRSF6 mutations have been reported for different patients, ranging from widespread presence in different germ layers to selective mutations or loss of heterozygosity (LOH) in DN T cells. 230,231In patients with mutations distributed more widely across the hematopoietic system, the frequency of mutated cells was found to increase from HSPCs to all major myeloid and lymphoid lineages, consistent with the role of FAS in regulating cell death of various cell types, but was by far most pronounced in the expanded DN T cells, highlighting cell-type and maturationstate dependent selection pressures. 2303][234][235][236] Mechanistically, these mutations render LGL cells independent of the cytokines that normally induce JAK/STAT signaling, promoting their uncontrolled survival and proliferation. 237Sustained expression of the pro-survival factor MCL1 prevents cell death, for example, in response to FAS engagement 238 or after activation. 239Additionally, STAT3 GOF mutations were connected to increased expression levels of DNA methyltransferases, both in T-LGL ex vivo and after overexpression in the KAI3 NK cell line, leading to global DNA hypermethylation. 240terestingly, similar mutations in STAT3 can be frequently detected in the general population, albeit at very low variant allele frequencies (VAFs), suggesting they might undergo a certain extent of positive selection in healthy individuals but are not sufficient to induce extensive lymphoproliferation with clinical manifestations. 241though recurrently mutated, in particular in T-LGL, only a minor fraction of NK-LGL actually harbors mutations with a well-defined impact on components of the JAK/STAT pathway. 235,242PI3K and MAPK signaling are other pathways that are recurrently affected, consistent with their well-defined role in leukemogenesis and oncogenic transformation. 235,243Interestingly, LGL patients often carry several mutations at subclonal VAFs, which in some cases evolve independently of each other, suggesting the co-existence of several clones. 235,243,244The clonal composition has been studied closely for T-LGL, with widely varying numbers of clones and clonal sizes between patients. 244A widely accepted model potentially explaining this co-existence of several clones postulates that chronic exposure to activating stimuli promotes the continuous proliferation of T and/or NK cells, enabling the acquisition and selection of advantageous mutations. 245This concept has been supported by early studies on IL-15, whose overexpression is sufficient to induce fatal LGL leukemia in transgenic mice, characterized by chromosomal aberrations. 246,247Similarly, overlap in TCR sequence motifs between leukemic and non-leukemic T-cell clones suggests antigenic stimulation as one driver behind LGL proliferation. 244In this context, HCMV has been suggested as a possible source of antigens recognized by T-LGL.Indeed, CD4 + T-LGL from a large fraction of patients responded to stimulation with an HCMV lysate or peptide. 248 another study, only a small fraction of TCRs was predicted to be HCMV-specific, 244 although it has to be noted that this analysis was limited to three epitopes from the HCMV pp65 protein, and prediction of TCR specificities is still a challenging endeavor.Therefore, whether HCMV infection plays a major role in the etiology of LGL remains an open question that could be of particular interest for NK-LGL, as the extensive clonal expansion of NK cells in response to HCMV might provide ample opportunities for mutations to arise and selection to act.Future studies could further address the clonal composition of NK-LGL, utilizing tools such as mitochondrial or somatic mutations and assess how these populations evolve over time, as dynamic shifts in clonal composition have been reported for T-LGL. 244,249,250Connecting those dynamics to the mutational spectrum might afford insights into their individual competitive advantages.Apart from better characterizing the genetic landscape of NK-LGL, such insights might be useful to better understand the clonal dynamics of NK cells from healthy donors, as NK-LGL cells often have a normal phenotype but an apparent selective advantage over their genetically intact neighbors.Therefore, closely studying their genetic makeup could identify candidate driver genes giving NK cells a selective advantage.
Clinically more favorable examples putting the spotlight on the positive selection of somatic variants in lymphocytes have been observed in the context of primary immunodeficiencies, where somatic reversion of mutant to wildtype alleles or second-site mutations can rescue immune cell development, in some cases resulting in milder phenotypes. 251Such reverse mosaicism has been described for various primary immunodeficiencies such as adenosine deaminase (ADA) deficiency caused by mutations in ADA, 252,253 X-linked severe combined immunodeficiency (SCID) caused by mutations in IL2RG, 254 Wiskott-Aldrich syndrome (WAS) caused by mutations in WAS, [255][256][257] SCID caused by mutations in CD247, [258][259][260] or leukocyte adhesion deficiency type-1 (LAD-1) caused by mutations in ITGB2. 261Similar to the disease-associated mutations in lymphoproliferative conditions, the cell types carrying such reversions at detectable frequencies vary widely between diseases and patients.
The revertant cell fraction has probably been studied most closely in WAS, where reversion occurs relatively frequently.6][257] Interestingly, in some cases, even multiple independent reversions or second-site mutations were detected in single individuals, 256 underlining how strong selection pressure combined with the extensive proliferative capacity of lymphocytes can amplify even such presumably rare events.Furthermore, special cases of purifying selection in lymphocytes were recently described in the context of mitochondrial disorders caused by heteroplasmic mitochondrial DNA (mtDNA) mutations. 262,263Patients suffering from mitochondrial encephalomyopathy, lactic acidosis, and stroke-like episodes (MELAS) caused by heteroplasmic mutations in the mitochondrial leucyl-tRNA gene MT-TL1 present with a broad spectrum of clinical phenotypes and severity, potentially due to differing distribution of the mutation to tissues. 264Intriguingly, T cells but also NK and B cells show reduced heteroplasmy of the damaging mutation. 262Similar negative selection occurs against single large-scale mtDNA deletions causing congenital disorders such as Pearson syndrome (PS). 263In this study, the authors provide a more nuanced picture by demonstrating a gradient in the frequency of damaging variants across T-cell subsets, from the highest heteroplasmy in naïve to the lowest in CD8 T EM and mucosalassociated invariant (MAIT) T cells, with CD4 T CM falling in between, and confirm that this also applies to MELAS patients.Consistent with this more penetrant depletion of pathogenic mtDNA in the more differentiated T-cell subsets, in vitro activation of T cells from a PS patient reveals purifying selection in expanding T cells.These findings underscore the unique ability of lymphocytes to undergo massive clonal expansion, creating specific selection pressures and abundant opportunities for the forces of selection to operate.
Overall, as somatic variants are one of the intrinsic mechanisms underlying the selection of the cells causing various hematological conditions, as well as their precursor lesions, they might also drive the heterogeneous behavior of normal lymphocytes under general, tissue-specific, or context-induced selection pressures and thereby shape their clonal composition.

| Selection of somatic variants in healthy individuals
Clonal hematopoiesis, lymphoproliferative disorders, and revertant mosaicism are rather obvious examples that magnify how the positive selection of somatic variants shapes the population dynamics of the hematopoietic compartment.However, they likely represent extreme cases in which cells acquire large gains in fitness, manifesting in macroscopic dominance over their unmutated competitors.
Understanding and quantifying the impact of the selection of somatic variants globally, including those with more subtle effects on cellular competition, is complicated by the stochasticity of the mutational processes and the ensuing high degree of genetic diversityeven for tumors.This complexity also makes differentiating somatic mutations into neutral passengers, positively selected drivers, and detrimental variants under negative selection pressure a challenging task that requires large sample sizes and models that accurately estimate background mutation rates.6][267][268] The main assumption for dN/dS as a selection measure is that synonymous mutations act selectively neutral and can, therefore, serve as a proxy for the mutation density to test for over-or underrepresentation of coding mutations.Depending on the question, dN/dS can be applied to quantify the overall effect of mutations (or different types thereof) on selection or to identify the role of mutations in individual genes for the selective process.Along these lines, the application of dN/dS models to the large resource of The Cancer Genome Atlas (TCGA) has demonstrated ratios slightly above 1 for both missense and truncating substitutions, indicating a higher abundance of mutations under positive than negative selection pressure. 269Indeed, extending the analysis to the gene level identified nearly 200 genes under positive selection in cancer, recapitulating many well-defined oncogenes and matching current estimates of the number of cancer genes.Conversely, only a few tens of genes are estimated to exhibit negative selection, suggesting a weak effect on cancer genome evolution.Together, these numbers also imply that the vast majority of coding substitutions appear to accumulate as neutral passengers, drifting along with the few driver mutations they co-occur with.
Recent studies started addressing the impact of these mechanisms under homeostatic conditions in healthy individuals at different ages.Interestingly, somatic variants seem to have a similar effect on the selection in healthy tissues as for tumors, 269 with dN/dS values slightly above 1.More specifically, a recent large-scale genome-wide analysis of HSPCs revealed a net positive selection of coding mutations with dN/dS at 1.06 (95% confidence interval 1.03-1.09),equating to 1 in 34 to 1 in 12 coding mutations being drivers for clonal expansion. 196Based on the dN/dS ratios, each adult was estimated to have more than 100 driver mutations among the sequenced HSPC colonies. 196Bayesian models applied to emulate HSPC family trees inferred that most drivers have moderate effects, with growth rates of 5%-10% per year but with a heavy tail of rare drivers conferring a greater selective advantage.For example, some HSPC clades with TET2 and DNMT3A mutations exhibited growth rates of >20%-30%, and there were numerous clades with similar fitness effects without known driver mutations.This suggests a substantial number of drivers with strong and even more with weak to moderate effects remain unknown, requiring even larger studies or alternative approaches for their identification.Similarly, mutations in lymphocytes from healthy individuals are under net positive selection with a recent dN/dS ratio estimate of 1.12 (95% confidence interval 1.06-1.19). 195Intriguingly, focusing the analysis on naïve T or B cells yields similar signals of positive selection, whereas the dN/dS ratios of memory cells seem to be closer to 1, potentially suggesting a more pervasive selection of somatic variants in naïve than memory lymphocytes, although these subset-specific estimates might be limited due to their smaller sample size.Nevertheless, it is tempting to speculate that the recruitment via specific antigen receptors might represent a bottleneck narrow enough to restrict the number of somatic variants, below the point at which a sizeable number would overcome stochastic differences in lymphocyte expansion to measurably influence cellular competition.
Innate lymphocytes might differ in this regard, as their recruitment via germline-encoded receptors likely lets significantly more somatic variants enter the competition.Hence, studies on the mutational landscape of innate lymphocytes, in particular comparisons of conventional and clonally expanded adaptive NK cells, might be a promising avenue that could shed light on how these mechanisms shape their clonal composition (Figure 4).

| Random monoallelic expression in the diversification of receptor repertoires and beyond
Apart from genetic differences, stably imprinted decisions on gene activity by regulation in cis might constitute another important source of heterogeneity on which selection might act.This is particularly important for different types of immune cells, as stable allelic expression is the basis for diversifying receptor repertoires, including those of NK cells.Diploid organisms inherit two copies of each gene, one from each parent, which can either be expressed (biallelic expression), expressed from one allele (monoallelic expression), or be silent.Imprinting is a prominent example of allelic expression observed many years ago based on parental-origin effects, resulting from the constitutive expression from the same parental allele across most cells. 270However, more prevalent is the regulation of gene expression in the somatic process of random monoallelic expression (RME).Specifically in females, one X chromosome is transcriptionally silenced in each cell in the process of X chromosome inactivation (XCI), balancing gene dosage between the sexes. 271Interestingly, also a fraction of autosomal genes have been found to undergo autosomal RME (aRME).Interestingly, aRME genes are dispersed over the genome, not silenced by the inactivation of whole chromosomes, and therefore largely regulated independently from each other. 272[274][275][276] Initial studies relying on single nucleotide polymorphism arrays to detect allelic expression of polymorphic genes found widespread allelic expression in both mice and humans; approximately 8%-16% of all genes were extrapolated to have allelic expression bias in clonal lymphoblastoid cell lines. 273,274Similarly, a recent study identified more than 2000 genes with some degree of aRME, by analyzing bulk RNA-seq data of paired samples from different human tissues for differences in allele ratios across tissues. 276The first scRNA-seq study addressing aRME also identified large numbers of genes undergoing allelic expression at different early stages of embryonic development, as well as in mature liver cells and fibroblasts. 275wever, none of these studies could analyze the clonal and only to a certain extent, the temporal stability of aRME in these settings.Indeed, pooling cells from the same embryo removed most aRME genes, suggesting their allelic expression is highly dynamic and might represent stochastic transcriptional bursting rather than stably propagated aRME.In a follow-up study, the clonal stability of aRME was more closely analyzed in murine fibroblasts expanded in vitro, as well as human ex vivo T-cell clones induced by YFV. 277Also in these settings, a large fraction of genes showed dynamic aRME in fibroblasts and especially in YFV-epitope-specific T cells, where even the majority of genes expressed were transcribed with significant allelic imbalance.However, when taking clonal identity into account, only few of these genes were consistently expressed from the same allele, indeed confirming that the majority of allelic imbalance can be attributed to dynamic transcription from one or the other allele.
With this stricter definition, only a small fraction (<1%) of expressed genes were identified as undergoing clonally stable aRME.These which they discuss technical aspects such as thresholds for allelic expression but also important semantic differences in their respective definitions of prevalence; while Reinius and Sandberg argue that only genes that are recurrently expressed from one allele in several clones should be considered, Gimelbrant et al. extrapolate the results to higher numbers because they suggest that each clone will only express a fraction of aRME genes. 278,279e central question for the purpose of this review is whether aRME might have an effect on clonal selection and expansion, which might be the case even if few genes are regulated in that manner.A striking result of the studies identifying frequent aRME was a significant enrichment of transmembrane receptor genes. 2731][282][283][284] However, mechanistically, these represent rather special cases, since their monoallelic expression is mainly the result of productive genetic rearrangement being restricted to one of the two alleles. 285,286Instead, transcriptional regulation that results in aRME is one of the key mechanisms determining NK cell receptor expression.9][290][291] In these studies, Ly49 receptor expression was clonally stable and mostly monoallelic.Similarly, KIRs are predominantly transcribed from one allele, and their expression correlates inversely with DNA methylation at proximal promoter CpG sites. 292,293Consistent with their probabilistic regulation, expression of Ly49s and KIRs roughly follows the "product role," resulting in co-expression frequencies close to the product of the individual receptor frequencies, 294 although deviations increase if more than two KIRs are expressed by a cell. 295Recently, other receptors such as CD45, CD8a, and Thy1 were demonstrated to be expressed in a stable monoallelic fashion, albeit by an extremely low frequency of cells (~0.01%), whereas most cells express both alleles. 290Despite these low frequencies of monoallelic expression, the extensive ability of lymphocytes to proliferate might expand such rare cells selectively if one of the allelic expression states provides a competitive advantage.
Intriguingly, the only confidently identified clonal aRME genes by scRNA-seq of YFV-specific T cells encoded cell surface receptors 277 : KLRB1 (encoding CD161), CD7, and IL18RAP, which are all expressed at lower levels by human memory NK cells. 65Given that the number of active alleles scales with transcript levels, the selective expansion of cells with monoallelic expression might be one of the mechanisms by which this downregulation on the population level is achieved. 275nversely, for other genes that are subject to aRME, biallelic expression might provide an advantage, since it is associated with ~twofold higher transcript levels. 275One such example might be KLRC2 (encoding NKG2C).Consistent with the allelic expression of Klrc1 in mice, 291 the KLRC locus was found to be among aRME genes identified in the recently analyzed human tissue atlas. 276Frequencies of NKG2C + NK cells, as well as its per-cell expression levels, are dependent on the presence of two functional alleles, supporting a gene-dosage effect. 92,296Functionally, engagement of NKG2C enables more efficient signaling and proliferation in homozygous compared to hemizygous individuals, suggesting a similar difference might exist between individual cells expressing different levels of the receptor. 92While there must be additional mechanisms enabling increased expression levels of NKG2C on memory NK cells, as also hemizygous individuals show higher per-cell expression after HCMV infection, 92 analyses of monovs biallelic expression frequencies in NKG2C + naïve and memory NK cells could be an interesting starting point to assess whether selection based on aRME occurs during HCMV infection (Figure 4).
Apart from cell surface receptors, some of the earliest examples of aRME were described for various cytokines such as Il4, Il5, Il13, [297][298][299] or Il2 300 Although different studies have yielded variable results regarding the quantitative aspects of aRME of these cytokine genes, which again underline the methodological difficulties in faithfully detecting and quantifying aRME, 272 these findings suggest aRME is a mechanism that might enable probabilistic diversification of effector functions.More recently, independent expression of the two alleles of the T cell-lineage-defining transcription factor Bcl11b has been demonstrated using reporter mice. 301Murine NK cells express Bcl11b only in immature stages, and their development is largely independent of Bcl11b. 302In fact, its deletion results in the differentiation of thymocytes into an NK-cell-like population.
In humans, BCL11B expression increases with differentiation and is most highly expressed in adaptive NK cells, where it is associated with the upregulation of T-cell-related genes such as CD3 subunits or CD5. 68In this context, it would be interesting to study whether there is a difference in the number of Bcl11b alleles expressed by conventional and memory NK cells.
Overall, aRME is an important mechanism generating the variegated, binary expression patterns of NK cell receptors that define the naïve NK cell repertoire from which memory cells are selected.
In addition, quantitative differences in gene expression based on mono-vs.biallelic expression might be a mechanism more subtly affecting NK cell fitness during clonal expansion and could explain some of the phenotypic characteristics of memory NK cells.Recent methodological developments that enable high-throughput singlecell sequencing with haplotype phasing will enable us to shed light on the relevance of these mechanisms on a global scale.

| Selection of transcriptional and epigenetic states
Fluctuating transcriptional and epigenetic cell states might represent another, more transient factor influencing clonal selection and expansion (Figure 4).The first description of such non-genetic heterogeneity as a substrate for Darwinian selection was probably based on experiments demonstrating a small proportion of penicillin-naïve bacteria able to withstand drug treatment. 303terestingly, their daughter cells were mostly sensitive to repeated drug challenges, containing similar frequencies of resistant bacteria as their drug-naïve parents, pointing toward the reversibility of the drug-resistant state.Since then, this concept has been revisited in mammals in the context of cancer cell resistance to drug treatment, identifying different non-genetic mechanisms that enable survival and re-expansion of a small proportion of therapy-refractory cells. 304An early study demonstrated a consistent fraction of resistant cells originating from an EGFR-mutant non-small-cell lung cancer cell line persisting through treatment with tyrosine-kinase inhibitors. 305Different clones of this and other cell lines generated similar frequencies of resistant cells that were too high (0.3%-5%) to arise spontaneously by genetic mechanisms.Re-acquisition of sensitivity under drug-free culture conditions further supported that resistance was non-genetic, while differences in gene expression between parental and resistant cells suggested these might play a role in establishing drug resistance.Similar observations in several settings were associated with dormant, less differentiated "cancer stem cells," which have been extensively studied in recent years, as discussed elsewhere. 304However, whether resistance is achieved by an active cell state transition in response to drug exposure-rather resembling Lamarckian adaption-or passive, Darwinian selection of a pre-existing state is still a topic of active debate and might, in fact, be context-dependent.Recent work analyzing the behavior of single cells and their progeny is starting to shed light on the underlying mechanisms, stability, and clonal inheritance of drug-resistant states.A series of elegant studies from Arjun Raj's laboratory identified and characterized rare "pre-resistant" cells in a treatmentnaive population derived from a single melanoma cell. 306Marker genes overlapping with the transcriptional program of fully resistant cells enriched a population that generated ~8-fold more resistant colonies than the bulk of cells, suggesting this program predisposes cells to become resistant.Pre-resistant cells displayed only a fraction of the transcriptional features of fully resistant cells.They also showed only minor changes in chromatin accessibility, losing their growth advantage within a few cell divisions under drug-free conditions.In contrast, full resistance acquired under drug treatment was stable even after "drug holidays" and encompassed a much larger number of changes in gene expression and chromatin accessibility.Systematic analysis of the genome-wide heritability of gene expression in this system identified hundreds of genes with heritable transitions into low or high-expression states that were typically stable for up to 5-10 divisions under drug-free conditions and significantly overlapped with the resistance gene module. 307As a result, clonally related cells that divided 3-4 times and were then separated for drug treatment showed largely concordant fates, underlining the cell-intrinsic and clonally inherited nature of this complex phenotype. 308Interestingly, even within the fully resistant population, different clones showed diverse transcriptional features, whereas clonal twins largely overlapped in their transcriptomes after selection, suggesting these diverse fates are predetermined.When treated with different doses or inhibitors, clonal twins acquired different transcriptional programs, potentially switching fates to adapt to these altered environmental conditions.Together, these studies suggest a two-step process in which rare pre-resistant cells are initially selected and then undergo cellular reprogramming, stabilizing the resistant phenotype.They further highlight how the interplay between transcriptional predetermination and the flexibility to adapt to different environmental conditions shapes the composition and phenotypic heterogeneity of isogenic cell populations.
How are these findings relevant to the clonal expansion of innate lymphocytes?In addition to the heterogeneity composed of the binary expression states of, for example, different receptors, there is dramatic variability in per-cell expression levels.As discussed above, these differences in receptor expression have important consequences for clonal expansion, resulting in avidity selection of Ly49H hi cells during MCMV infection, which also seems to occur for human memory NK cells expressing high levels of NKG2C. 8,98Although differences in Ly49H levels of sorted populations persisted after MCMV infection, receptor levels of the Ly49H lo population increased, suggesting some cells fluctuate into the high-expression state, conferring them with a selective advantage.Importantly, this selection of gene expression states might not only apply to the main receptors involved in the recognition of infected cells but could potentially be one of the mechanisms underlying other transcriptional and epigenetic characteristics of memory NK cells (Figure 4).In support of this hypothesis, we believe that we make a few observations in our study on clonally expanded memory NK cells that draw important parallels to the clonal inheritance of cell states observed in the context of drug resistance.While there is a characteristic signature in chromatin accessibility shared by all memory NK cells, individual subpopulations corresponding to one or a few clonotypes differed markedly from each other.Differences between subpopulations were largely donor-specific, and many regions were significantly associated with individual clones. 9Conversely, cells belonging to one clonotype were strikingly concordant in their epigenetic profiles, revealing that the diversified picture of the memory population is the result of its assembly from epigenetically distinct clones.Raj et al. note similar differences between virtually every resistant clone, leading to the hypothesis that some general adaptive process might be at work that "'burns in' all intrinsic differences in the initial state, even minor ones, regardless of their functional consequence". 309In further support of partial inheritance of the initial cell state, we found parts of the signatures defining the maturation gradient from conventional CD56 bright to CD56 dim NK cells to be reflected by different memory NK cell subpopulations and clones.Similar stability of transcriptionally distinct states of NK cell subsets has been observed in the murine system after MCMV infection, 171 and clonal concordance of gene expression has recently been described for T cells induced by yellow fever vaccination (YFV), 310 suggesting this might be a general mechanism impacting clonal identity of lymphocytes.Although the mechanisms underlying this epigenetic and transcriptional memory are only beginning to be unraveled, an enrichment of AP-1 family TF motifs in the regions associated with drug resistance, including in the proximity of genes that showed clonally coherent expression points toward a role of these TFs in the adaptive process. 306,307Indeed, knockout of AP-1 family TFs such as FOSL1, JUNB, and JUN affected those genes more strongly than expression-level matched control genes. 307As AP-1 family TF motifs were also strongly enriched in the chromatin specifically accessible in memory NK cells and AP-1 TF activity induced by combined peptide and pro-inflammatory cytokine stimulation, their potential involvement in "burning in" the unique chromatin accessibility associated with individual NK cell clones might be another interesting parallel between these adaptive processes. 9Therefore, the experimental and analytical approaches utilized to identify pre-resistant states of cancer cells might be a promising toolkit for identifying cell states in the naïve NK cell compartment that predispose individual cells to clonal expansion.

| CON CLUDING REMARK S
The ability of innate lymphocytes to clonally expand and persist introduces a new perspective on immune memory.Importantly, we have a breadth of knowledge on the mechanisms of clonal selection and expansion of adaptive lymphocytes at our hands that the immunological community has built over the last decades.To truly understand how these mechanisms apply to the innate immune system, it will be crucial to get a quantitative understanding of the dynamics shaping innate memory compartments.Conversely, insights into the mechanisms of clonal success within the innate immune system might apply to T and B cells, further broadening the scope to bottlenecks other than selection via antigen-specific receptors.
Although massive clonal expansion in the innate immune system has so far only been described for NK cells in the context of CMV infection, it seems likely that innate lymphocytes have this ability in other contexts, and CMV is extraordinary mainly because it induces large clones with characteristic phenotypes that are relatively easy to identify.Recent studies demonstrating immune memory of innate lymphoid cells 311,312 warrant further investigations into their ability to clonally expand and persist.Their localization to confined tissue niches might already constitute a major clonal bottleneck, as might also be the case for tumor infiltration.Additionally, chronic exposure to inflammatory cues might result in persistent remodeling of the innate lymphoid compartment, potentially perpetuating chronic inflammatory or autoimmune diseases.Therefore, identifying settings of innate clonal expansions and the mechanisms underlying their clonal success and deletion will be crucial to understanding how they contribute to the control of pathogens, tumors, but also immune-related diseases.

ACK N OWLED G M ENTS
In addition to the TCR, serving as Signal 1, T cells primarily depend on two supplementary signals: Signal 2 involves co-stimulation through receptors like CD28, and Signal 3 encompasses pro-inflammatory cytokines such as IL-12 or type I IFN, which further enhance the response. 94Similarly, different signals are required for the efficient induction of effector functions and memory differentiation of NK cells during CMV infection (Figure 2).While the antigen receptors transmitting Signal 1 for NK cells are invariable and therefore do not differ in affinity for their cognate antigen, there are different mechanisms resulting in different signal strengths with important consequences for NK cell expansion.

F I G U R E 2
Cooperative signals induce differentiation of NK cells during HCMV infection.Recognition of HCMV gpUL40-derived peptides presented on HLA-E is mediated by NKG2C (Signal 1) inducing the activity of IRF4, NFAT, AP-1, and NF-κB TFs.Co-stimulation via CD2 (Signal 2) supports activation, especially when NKG2C is engaged by low-affinity peptides.Pro-inflammatory cytokines IL-12 and IL-18 (Signal 3) induce AP-1, NF-κB, and STAT4 activity.Mainly IRF4, AP-1, and STAT4 have been demonstrated to cooperatively induce chromatin remodeling required for memory NK cell differentiation.HLA class I downregulation enables missing-self responses due to lack of inhibition via self-MHC-specific KIR and ILT2.Created with BioRe nder.com.
Future studies addressing the clonality of NKG2C − NK cell expansions might enable insights into the role of NKG2C and other receptors in clonal competition.As Signal 2, mainly two different receptors have been demonstrated to support expansion in mice and humans.The murine Ly49H + NK cell response is especially dependent on signaling through DNAM-1.Its ligands CD155 and CD112 are rapidly upregulated on dendritic cells (DCs) and macrophages during MCMV infection and similarly contribute to the recognition of HCMV-infected DCs by human NK cells. 101,102While acute IFNγ production is equally efficient without DNAM-1, expansion and memory formation are severely diminished, translating into impaired viral control in primary and especially secondary responses after adoptive transfer. 101Paradoxically, DNAM-1 expression is progressively lost by Ly49H + NK cells during MCMV infection in an m157-dependent manner, resulting in a mostly DNAM-1 − memory compartment.
biased responsiveness to cytokine stimulation might also affect the composition of cytokine-induced memory-like NK cells.In addition to those activating signals that broadly match the signal 1-3 model established for T cells, the skewing of different, mainly inhibitory NK cell receptors on memory NK cells suggests that these receptors further bias clonal selection and expansion.The most important examples are the MHC class I specific receptors from the Ly49 and KIR families.Mechanistically, a survival advantage of KIR-expressing CD8 memory T cells after activation This effect is suppressed by co-expression of the activating receptor Ly49D which recognizes the same H-2D d MHC class I allele.Despite being less proliferative at the peak of infection, this survival advantage shifts the frequencies of the Ly49H + NK cells in favor of the Ly49D + Ly49A − subset, whereas they remain unchanged in C57BL/6 mice without H-2D d .In another study, expression of the self-MHC-specific inhibitory Ly49C/I receptors did not affect clonal sizes at Day 8 after MCMV infection. 8 inhibitory self-MHC-specific receptors might provide an advantage specifically in the memory phase, or murine and human NK cells behave fundamentally different in this regard.Overall, due to this central requirement of a combination of different signals, intrinsic differences in the ability to sense and respond to them might bias clonal selection, expansion, and differentiation F I G U R E 3 NK cells are selected by their combinatorial specificities.The recruitment and competitive expansion of NK cells during HCMV infection can be biased by their receptor expression profiles and manifest in the phenotype of the memory NK cell population.Cells lacking NKG2A, while expressing NKG2C, CD2, and self-MHC-specific KIR (sKIR) preferentially contribute to memory NK cell expansions.Within this subset, high levels of NKG2C and CD2 may provide a further advantage.Created with BioRe nder.com.
While differences in antigen-receptor affinity and avidity, expression of receptors for co-stimulation, and other signals such as proinflammatory cytokines are important factors driving the relative expansion of individual lymphocytes, landmark studies from Dirk Busch's and Ton Schumacher's laboratories established an additional layer of apparent stochasticity that diversifies the T-cell response.By elegantly exploiting transgenic OT-I T cells specific to the Ovalbumin-derived peptide SIINFEKL, thus eliminating TCR specificity as a major cell-intrinsic factor driving recruitment and expansion, they demonstrated stochastic differences in the proliferation and differentiation of individual T cells that propagate into extremely disparate fates on a population level. 159,160As a result, clonal progenies of individual T cells differ in size by several orders of magnitude, producing strongly skewed size distributions in which ~5% of naïve precursors generate over half of the OT-I T-cell pool.While individual T cells can generate diverse progeny encompassing various differentiation states from the spectrum of short-lived effectors to long-lived memory cells, the quantitative models applied by Buchholz et al. provided important insights into the stochastic mechanisms explaining such dramatic differences in size and differentiation state between T-cell clones.Building on these models, in vitro imaging studies following the fate of individual T cells over several generations created a deeper understanding of this stochasticity, complementing the expansion and differentiation models with further kinetics insights.Early after activation, most conflicting estimates of the global prevalence of aRME initiated an extensive debate between the authors of the different studies, in F I G U R E 4 Cell-intrinsic sources of heterogeneity as a substrate for selection.Somatic mutations, autosomal random monoallelic expression (aRME), and fluctuating transcriptional states are well-defined determinants of cell heterogeneity.Combined selection of all these intrinsic aspects might constitute a tight bottleneck through which only rare cells can pass, thus contributing to the large size of memory NK cell clones.Created with BioRe nder.com.
This work was funded by the ERC Advanced Grant "MEM-CLONK" (101055157) to C.R. and Berlin Health Innovations (BHI) Validation Fund to C.R. and T.R. Funded by the European Union.Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or the European Research Council Executive Agency.Neither the European Union nor the granting authority can be held responsible for them.Open Access funding enabled and organized by Projekt DEAL.CO N FLI C T O F I NTE R E S T S TATE M E NT C.R. and T.R. are listed as inventors on a patent series protecting the therapeutic use of HLA-E restricted peptides.