The peripheral differentiation of human natural killer T cells

Abstract The peripheral maturation of human CD1d‐restricted natural killer T (NKT) cells has not been well described. In this study, we identified four major subsets of NKT cells in adults, distinguished by the expression of CD4, CD8 and CCR5. Phenotypic analysis suggested a hierarchical pattern of differentiation, whereby immature CD4+ CD8− CCR5− cells progressed to an intermediate CD4+ CD8− CCR5+ stage, which remained less differentiated than the CD4− CD8− and CD4− CD8+ subsets, both of which expressed CCR5. This interpretation was supported by functional data, including clonogenic potential and cytokine secretion profiles, as well as T‐cell receptor (TCR) excision circle analysis. Moreover, conventional and high‐throughput sequencing of the corresponding TCR repertoires demonstrated significant clonotypic overlap within individuals, especially between the more differentiated CD4− CD8− and CD4− CD8+ subsets. Collectively, these results mapped a linear differentiation pathway across the post‐thymic landscape of human CD1d‐restricted NKT cells.


INTRODUCTION
Classical natural killer T (NKT) cells express a highly biased T-cell receptor (TCR) repertoire and serve as a bridge between the innate and adaptive immune systems. 1,2 The semi-invariant human NKT cell TCR comprises a fixed TRAV10/TRAJ18 gene-encoded a-chain with a canonical CDR3a loop 3 and a restricted TRBV25 gene-encoded b-chain, which together enable the recognition of glycolipids presented in the context of CD1d. 4 NKT cells deploy a range of effector functions in response to antigen encounter and contribute in various ways to the immune processes that mediate pathogen control, tumor surveillance, allergic phenomena and autoimmune disorders. 5 Although initially considered to be homogeneous, later studies revealed considerable phenotypic and functional diversity within the peripheral NKT cell compartment. Two subsets, CD4 + CD8 À and CD4 À CD8 À , have been described in mice, and a third subset, CD4 À CD8 + , has been described in humans. 6 These patterns of coreceptor use segregate with functionally distinct effector programs. 7,8 The development of murine NKT cells is thought to comprise four stages, based on the expression of CD24, CD44 and NK1.1. 9,10 Further nuances are suggested by the existence of mature PLZF high Tbet low RORct low interleukin (IL)-4-producing and PLZF high Tbet low RORct high IL-17-producing subsets in the thymus that resemble NK1.1 À NKT cells. 11,12 It is also likely that peripheral CD4 + and CD4 À NKT cells in mice represent distinct lineages that emigrate independently from the thymus. 13 However, the extent to which human NKT cells follow an equivalent differentiation pathway remains unclear, despite close parallels in the TCR-mediated antigen recognition process and the highly conserved nature of CD1d.
In this study, we combined phenotypic, functional and molecular techniques to characterize the post-thymic differentiation of human NKT cells. Our data supported the notion of a single lineage compartment and outlined a maturation pathway compatible with the reported heterogeneity among circulating subsets of CD1drestricted NKT cells.

Phenotypic analysis of NKT cells
To define the properties of PBS57-hCD1d multimerbinding cells, we initially measured several proteins typically expressed by T and/or NK cells (Figure 1b and Table 1). The NK cell markers CD56 and CD161 were found at much higher frequencies on NKT cells compared with conventional T cells. In contrast, NKT cells almost completely lacked CD16 and infrequently expressed the NK recognition molecules NKB1 and NKAT2. These results largely concurred with a previous report that used the monoclonal antibody 6B11 to identify NKT cells. 19 However, there were two notable exceptions. First, we found that nearly all NKT cells expressed CD95, consistent with a predominant memory phenotype. 20 Second, we found that a higher fraction of NKT cells expressed CD56. These differences likely reflected greater sensitivity in our flow cytometry panel for these dimly expressed proteins.
Most NKT cells expressed CCR5 but lacked CD45RA and CD62L, a phenotype reminiscent of effector T cells. However, similarly high frequencies of NKT cells expressed CD27, CD28 and CD127, markers typically associated with central memory T cells. The expression of cytolytic enzymes was also consistent with a central memory phenotype, in that most NKT cells expressed granzyme A, whereas only a few expressed granzyme B, and virtually none expressed perforin. NKT cells therefore occupied a unique phenotypic niche most akin to central memory T cells.

Functional analysis of NKT cells
To determine the functional characteristics of NKT cells, we stimulated PBMCs with aGalCer, anti-CD3 or PBS57-hCD1d in the presence of brefeldin A and measured the expression of intracellular cytokines, namely IL-2, IL-4, TNF and IFNc. The PBS57-hCD1d multimer emerged as the most potent stimulant (data not shown). In line with the division-linked heterogeneity of memory T cells, functional differences were evident among phenotypically defined subsets of NKT cells (Figure 2b). In particular, we found that CD4 + and DN NKT cells were significant sources of IL-4, as reported previously. 7,8,19,[21][22][23] Large fractions of CD4 + NKT cells also expressed high levels of IL-2 and TNF compared with CD4 À NKT cells. In contrast, CD4 À NKT cells more frequently expressed IFNc. These patterns of cytokine secretion suggested that CD4 + NKT cells were less differentiated than CD4 À NKT cells.
In further experiments, we sorted NKT cell subsets to purity and quantified extracellular cytokine secretion in response to stimulation with the PBS57-hCD1d multimer ( Figure 2c). These measurements were largely concordant with the intracellular cytokine profiles, barring a few minor discrepancies likely attributable to methodological differences. One notable exception was the finding that DN NKT cells produced high levels of TNF, which suggested greater output on a per cell basis relative to CD4 + or CD8 + NKT cells.

Differentiation of NKT cells
On the basis of these functional traits, we hypothesized that CD4 + NKT cells were less differentiated than CD4 À NKT cells. This inference was substantiated by comparing markers of differentiation across NKT cell subsets ( Table 2). Fewer CD4 + NKT cells expressed CD11a, CCR5, granzymes A and B, and NK-associated markers compared with CD4 À NKT cells. In contrast, CD4 + NKT cells more frequently expressed CD62L. Careful analysis further revealed distinct subsets of CD4 + NKT cells. The smaller population, representing about 5-10% of CD4 + NKT cells, expressed CD62L but lacked CCR5, CD161 and granzyme A (Figure 3a and Supplementary figure 1). This phenotype was reminiscent of na€ ıve T cells, with the exception that NKT cells rarely expressed CD45RA. The CCR5 À subset therefore appeared to be less differentiated than the CCR5 + subset within the CD4 + NKT cell compartment.
To confirm these findings, we undertook several independent lines of experimentation. First, we characterized NKT cells in cord blood (Figure 3b). These progenitors were found to be uniformly CD4 + CCR5 À (and still CD45RA À ), with a functional profile similar to   Table 1.
that of adult CD4 + NKT cells (high expression of TNF and relatively low expression of IFNc). Second, we quantified T-cell receptor excision circles (TRECs) in NKT cell subsets. TREC levels decline geometrically with the number of divisions since rearrangement of the TCR. 24 As shown in Figure 3c, the highest TREC levels were detected in adult and cord blood CD4 + CCR5 À NKT cells. This subset therefore emerged during the initial stages of development. Progressively lower TREC levels were present in CD4 + CCR5 + and CD4 À NKT cells. Third, we cloned phenotypically defined NKT cells. In line with previous studies of in vitro proliferation, 25,26 we found that CD4 + NKT cells were significantly more amenable to clonogenic expansion compared with CD4 À NKT cells ( Figure 3d). It was also notable that CD4 persisted on the surface of all expanded CD4 + NKT cell clones (data not shown). This finding suggested that any transition to the CD4 À state was either rare or contingent on additional stimuli, such as further proliferation or an unknown in vivo signal. Similarly, both DN and CD8 + NKT cells largely retained their phenotypes in culture, although there was some plasticity in the expression of CD8a. Most CD8 + clones became heterogeneous in this respect, and the occasional DN clone acquired CD8a.

Clonotypic analysis of NKT cell subsets
To probe these lineage relationships in more detail, we performed an unbiased molecular analysis of all expressed TR gene products in sort-purified (> 98%) subsets of NKT cells. The flow cytometric sorting strategy is shown in Supplementary figure 2.
In a cross-sectional analysis of three healthy subjects, we found that the canonical TRAV10/CVVSDRGSTLGRLY/ TRAJ18 sequence 14,15,27 was ubiquitous and highly conserved at the nucleotide level among CD4 + , DN and CD8 + NKT cells (Figure 4a, b). Some additional TCRa sequences were detected, especially in subject 4, presumably reflecting a lack of allelic exclusion. In line with previous reports, 28 the corresponding TCRb sequences were substantially more diverse and predominantly TRBV25-1 + (Figure 4c, d). Importantly, we found nucleotide-identical TCRb clonotypes within all three phenotypically defined subsets from subject 4 and subject 7, thereby providing direct evidence that CD4 + , DN and CD8 + NKT cells were related by ancestry and/or interconversion. Our data were significant in this context. Assuming a null hypothesis that each subset arose independently, equivalent sharing of TCRb sequences would have been expected among CD4 + , DN and CD8 + NKT cells both within and between subjects. This scenario was rejected (P < 10 À4 ; Fisher's exact test).
To confirm and extend these findings, we used highthroughput sequencing to analyze TCRb sharing across NKT cell subsets in three additional healthy subjects. In each case, two samples of peripheral blood were taken 6 months apart, thereby enabling an assessment of clonotype distribution over time. Moreover, the CD4 + NKT cell subset was further sort-purified on the basis of CCR5 expression. In line with our conventional sequencing analysis, we found shared clonotypes at individual time points across all four phenotypically defined subsets of NKT cells (Figure 5a). The most extensive overlap was detected between DN and CD8 + NKT cells. This pattern was recapitulated in the longitudinal dataset, suggesting a high degree of fluidity between these phenotypes (Figure 5b). Relatively few TCRb sequences were shared over time, however, especially between the CD4 + and CD4 À NKT cell subsets. A few CD4 + CCR5 + clonotypes nonetheless appeared to seed the DN and CD8 + populations. In contrast, no transitions were observed back to the CD4 + CCR5 À state. Collectively, these data provided evidence for linear differentiation within a highly dynamic system.

DISCUSSION
Human NKT cells were classified previously into three broad subsets, defined on the basis of coreceptor Percentages are given as mean AE one standard deviation for 5-12 subjects.
expression (CD4 + , DN and CD8 + ). Earlier work also suggested that NKT precursors expressed CD4 but not CD161. This view was upheld by the observation that CD4 + NKT cells predominated both in cord blood and in the thymus. [29][30][31] Moreover, the expression of CD161 was shown to increase with age in young humans, 30 consistent with a progressive maturation of the NKT cell compartment. In line with previous studies, 8,32,33 we found that CD4 + NKT cells could be subdivided according to the expression of CCR5. Our collated phenotypic, functional and molecular data further indicated that these subsets represented distinct stages of maturation within the post-thymic NKT cell pool. CD4 + CCR5 À NKT cells likely emerged first in the periphery. These cells expressed many markers commonly associated with a na€ ıve phenotype among conventional T cells (CD27, CD28, CD62L, CD127 and low CD11a) and lacked cytolytic enzymes (granzymes A and B, and perforin). They also harbored TREC levels comparable to those found in cord blood NKT cells, which shared all of the same phenotypes. This finding suggested that CD4 + CCR5 À NKT cells were produced continuously throughout adulthood. However, these cells were clearly not na€ ıve in the classical sense, because they did not express CD45RA and could be triggered in some instances to produce IFNc.
CD4 + CCR5 + NKT cells appeared to be more differentiated. They had lower TREC levels, consistent with an average of two additional divisions from the CCR5 À stage, and readily produced effector cytokines upon restimulation. During in vitro clonal expansion, however,  (c) CD4 + , DN and CD8 + NKT cells (n = 10 000 per subset) were stimulated with the PBS57-hCD1d multimer. The amount of each cytokine released into the supernatant was measured using cytokine bead array. Results are shown as the mean AE one standard deviation from two independent experiments (n = 3-5 subjects). *P < 0.05 (Kruskal-Wallis test with Steel-Dwass correction).
these cells largely maintained their phenotype, suggesting that further transition was either rare or dependent on other stimuli, such as additional division cycles or a signal that was not recapitulated in culture. In line with this interpretation, distinct tissue-specific microenvironments were shown previously to influence both the differentiation 34 and function [35][36][37] of murine NKT cells.
CD4 À NKT cells were either CD8 À or CD8aa + (none expressed CD8b). Our data suggested that the expression of CD8aa on CD4 À NKT cells was reversible, potentially reflecting the promiscuity of the a-chain promoter upon Tcell stimulation. On this basis, it was conceivable that CD8 + NKT cells arose as a consequence of in vivo activation rather than programmed differentiation, with functional differences mirroring the chronology of antigen encounter on a cell-bycell basis. This notion was supported by the clonotypic relationship between DN and CD8 + NKT cells. In comparison with the CD4 + subsets, CD4 À NKT cells displayed lower clonogenic potential and greater functionality, particularly with regard to the expression of cytolytic enzymes. These features indicated a more differentiated phenotype, substantiated by the relative absence of TRECs.
DN and CD8 + NKT cells were rarely terminally differentiated in the classical sense, instead maintaining a "central memory-like" phenotype in relation to conventional T cells (CD27 + CD28 + CD127 + ). However, the latter seldom expressed these molecules alongside CCR5. This key difference potentially reflected contrasting selection processes in the thymus and/or divergent programming at the transcriptional level. 13 In practical terms, the expression of CCR5 indicated that differentiated NKT cells were poised to mobilize rapidly and participate in early immune responses triggered by the release of CC-chemokines. 38 Collectively, these results provided evidence for a linear differentiation pathway, as depicted in Figure 6. The most likely alternative scenario was that CD4 + and CD4 À NKT cells arose separately as distinct lineages in the thymus. However, this latter model failed to explain why CD4 À NKT cells divided and differentiated in the periphery to a much greater extent than CD4 + NKT cells. It was also difficult to reconcile independent recombination events with the contemporaneous occurrence and longitudinal sharing of identical TCRb clonotypes between the CD4 À and CD4 + subsets. Accordingly, our findings were most consistent with a single lineage compartment and offered a working hypothesis to clarify the peripheral ontogeny of human NKT cells.

METHODS Subjects
Adult blood samples were obtained from male and female healthy volunteers aged 23-45 years, recruited via the Blood Bank at the National Institutes of Health (Bethesda, MD). Cord blood samples were obtained from healthy newborn infants delivered full-term at the Shady Grove Hospital Center (Gaithersburg, MD). The use of human samples for this work was approved by the Institutional Review Board of the National Institute of Allergy and Infectious Diseases at the National Institutes of Health.

Culture medium, antibodies and dyes
PBMCs were isolated via standard density gradient centrifugation and maintained in RPMI 1640 medium supplemented with 10% fetal bovine serum, 2 mM Percentages are given as mean AE one standard deviation for 5-7 subjects.

Clonal expansion of NKT cells
Single cells from CD4 + , DN and CD8 + NKT cell subsets were sorted into individual wells of a 96-well plate using a modified FACSAria flow cytometer (BD Biosciences). Allogeneic PBMCs from a healthy subject were loaded with aGalCer (100 ng mL À1 ; Enzo Life Sciences, Farmingdale, NY, USA) and irradiated (4000 rad) for use as antigen-presenting cells.
The cultures were supplemented with IL-2 (100 U mL À1 ) and IL-7 (10 pg mL À1 ) and replenished with fresh medium every 3 days. Active clones were transferred to 24-well plates after 9 days and harvested for phenotypic analysis by flow cytometry after further expansion.

Intracellular cytokine assay
PBMCs were suspended in culture medium with PBS57-hCD1d (0.2 lg mL À1 ), anti-CD28 (1 lg mL À1 ) and anti-CD49d (1 lg mL À1 ) for 12 h at 37°C. Brefeldin A was added at a concentration of 1 lg mL À1 for the final 8 h. The cells were then washed, stained for lineage/differentiation markers, washed again, fixed/permeabilized using Cytofix/Cytoperm (BD Biosciences) and stained for intracellular cytokines. After a final wash step, the cells were fixed and stored at 4°C pending analysis.

Cytokine bead array
Cytokine production was assessed via cytometric bead array using a Human Th1/Th2 Cytokine Kit (BD Biosciences). Briefly, PBMCs were stained with the PBS57-hCD1d multimer and directly conjugated antibodies to identify NKT cell subsets, then sorted in bulk using a modified FACSAria flow cytometer (BD Biosciences). Sorted cells were cultured in a 96-well plate at 10 000 cells/50 lL per well in the presence of anti-CD28 (1 lg mL À1 ) and anti-CD49d (1 lg mL À1 ) with or without PBS57-hCD1d (0.2 lg mL À1 ) for 18 h at 37°C. Supernatants were recovered after incubation and assessed for cytokine production by flow cytometry using capture beads.

TREC analysis
TREC frequencies were evaluated via quantitative real-time PCR (qRT-PCR) as described previously. 24 Briefly, NKT cell subsets were sorted in bulk as described above and lysed with proteinase K. The solubilized DNA was then used directly as the amplification template. qRT-PCR was performed in duplicate tubes against a standard curve generated from serial dilutions of TREC plasmid DNA.

Conventional TCR sequencing and data analysis
Viable NKT cells (1000-10 000 per subset) were sorted by flow cytometry as described above to > 98% purity. Molecular analysis of expressed TR gene rearrangements was then performed using a template-switch anchored RT-PCR according to reported protocols. [40][41][42] A minimum of 50 functional sequences was analyzed for each sorted NKT cell population. Each TCR sequence was aligned against all reference TRAV and TRAJ or TRBV, TRBD and TRBJ gene sequences from the IMGT database. 43 The V genes were aligned first, followed by the J genes and, for TRB sequences, the D genes. The V and J genes were assigned as the longest, highest percentage identity match with the TCR sequence. Clonotype identity across subsets within individuals was defined by precise matching of nucleotide sequences.  Figure 5. High-throughput analysis of TCR use in NKT cell subsets. CD4 + CCR5 À , CD4 + CCR5 + , DN and CD8 + NKT cells were sortpurified from healthy subjects (n = 3). A high-throughput sequencing approach was used to profile the corresponding TCRb repertoires. The experiment was repeated 6 months later.  Figure 6. The peripheral ontogeny of human NKT cells. Based on the phenotypic, functional and molecular data presented in this study, we propose a linear differentiation model that describes the peripheral ontogeny of human NKT cells. The depicted lineage relationships are consistent with previous reports, as reviewed in detail elsewhere. 13 High-throughput TCR sequencing and data analysis Libraries were constructed as described previously. 44 Briefly, mRNA was extracted from sorted cells using a lMACS mRNA Isolation Kit (Miltenyi Biotec, Sunnyvale, CA, USA). A templateswitch anchored RT-PCR was then performed using a modified version of the SMARTer RACE cDNA Amplification Kit (Clontech, Mountain View, CA, USA). After cDNA synthesis and purification, TCRb templates were amplified using a KAPA Real Time Library Amplification Kit (Kapa Biosystems, Wilmington, MA, USA), with a TRBC gene-specific 3 0 primer (5 0 -GCTTCTGATGGCTCAAACACAGCGACCT-3 0 ) and a standard 5 0 primer (5 Primer II A; Clontech). Amplified samples were separated using an E-Gel Size Select 2% Gel System (Life Technologies). Illumina adaptors were added via a second PCR using the KAPA Real Time Library Amplification Kit (Kapa Biosystems), incorporating 20 lL of each library with one 3 0 primer (5 0 -CAAGCAGAAGACGGCATACGAGATTGCTTCTG ATGGCTCAAACACAGCGACCT-3 0 ) and two 5 0 primers (PE1 FCB ILL 1_2 and the corresponding PE1 ILL barcode 2_2 primer). PCR products were cleaned using an AMPure XP Kit (Beckman Coulter). Libraries were quantified using a KAPA Library Quantification Kit Illumina (Kapa Biosystems) and clustered in the Illumina Cbot according to the manufacturer's instructions (Illumina, San Diego, CA, USA). Sequencing was performed using an Illumina HiSeq instrument with a modified protocol for 150 base paired-end reads. 44 Read 1 was sequenced with the Illumina read 1 primer. Read 2 was sequenced with a mix of 13 TRBJ gene-specific primers. The minimum coverage for each sampled population was approximately 10 reads per input cell. TCRb annotation was performed by combining a custom Java script with the BLAST1 program (National Center for Biotechnology Information, National Institutes of Health). Briefly, the V and J genes were identified first, and the CDR3b was determined by finding the conserved cysteine at the 5 0 end and the conserved phenylalanine at 3 0 end. Unique TCRb combinations (V-D-J) were collapsed to determine the count.

Statistical analysis
Significance testing was implemented using JMP software version 14 (SAS Institute Inc., Cary, NC, USA).