GPA33: A Marker to Identify Stable Human Regulatory T Cells

Forkhead box protein 3-expressing CD4⁺ regulatory T (Treg) cells inhibit undesirable immune responses through a variety of mechanisms and thereby limit immunopathology (1). Two lineages of Treg cells have been described, which differ in their developmental origin. So-called thymic Treg (tTreg) cells develop from immature thymocytes in response to recognition of self-antigen (2). Mature conventional CD4⁺ T cells can, however, also differentiate into Treg cells, known as peripheral Treg (pTreg) cells, in response to recognition of Ags derived from commensal organisms or food (3–5). tTreg and pTreg cells are both needed for full protection from (auto)inflammatory disease. Still, major functional differences between these cell types have not been found, suggesting that their complementarity may stem from their distinct Ag specificities. One important difference between these Treg cell types has, however, been identified. Whereas tTreg cells are thought to be irreversibly committed to the Treg cell lineage, pTreg cells may have the capacity to revert into conventional T (Tconv) cells and can cause pathologic conditions upon adoptive transfer (6–8).

Adoptive cell therapy (ACT) with Treg cells is a powerful method to attenuate inflammation and promote acceptance of organ allografts in mice (9–14). Clinical trials are ongoing to mitigate graft-versus-host disease (15–22), autoimmune diseases like type 1 diabetes mellitus (23, 24), and to prevent rejection of organ grafts (25). Despite encouraging preliminary results, mouse studies suggest that greater efficacy must be possible (24), especially through the use of Treg cells with AgRs specific for peptides presented in the tissues that must be protected (26–34).

There are several hurdles for ACT with Treg cells (35). First, pure populations of human Treg cells are difficult to obtain. CD4⁺ Treg cells express CD25 and lack CD127 (36–38), but these markers do not allow rigorous separation between Treg and Tconv cells. To limit the presence of contaminating Tconv cells, CD4⁺CD25⁺CD127^– cells are often expanded in vitro in the presence of the drug rapamycin. This drug inhibits Tconv cells, whereas Treg cells are relatively tolerant to it (39–42). Nonetheless, rapamycin does reduce the Treg cell proliferative capacity and may affect their function (43–45). Moreover, rapamycin induces conversion of Tconv into Treg cells, which, however, readily lose their Treg cell properties upon withdrawal of this drug (44). Finally, rapamycin presumably does not exclude pTreg cells from populations of Treg cells used for ACT. After adoptive transfer, unstable (rapamycin-induced or peripheral) Treg cells can lose expression of FOXP3, acquire the ability to produce proinflammatory cytokines (like IL-2, IFN-γ, and IL-17), and cause pathology (6–8, 46–49). This is a particular risk for next generation treatments using Treg cells carrying tissue-specific AgRs (26, 27, 29–35, 50). It is important, therefore, to develop methods to isolate pure Treg cell populations devoid of Tconv cells and unstable Treg cells.

Because of their superior lineage fidelity, tTreg cells are probably the cell type of choice for ACT (51–53). Reliable markers to identify human tTreg cells have, however, not been found. One candidate, neuropilin 1, is preferentially expressed on tTreg cells in mice, but is not present on human Treg cells in blood (49, 54, 55). Murine tTreg cells characteristically coexpress the transcription factors Foxp3 and Helios. Although FOXP3 can transiently be expressed by activated human Tconv cells, Treg cells express this transcription factor regardless of activation status (56, 57), and this is essential for maintenance of Treg cell identity and function (58, 59). Helios is also not a perfect marker (60), as Helios^– Treg cells and recent thymic emigrant (RTE) Tconv can convert into Helios⁺ Treg cells under certain conditions (61, 62). Nonetheless, tTreg cells reportedly never lose expression of Helios (63). Importantly, Helios helps maintain Treg cell stability (64–66) and the production of Tconv cell–specific effector cytokines in FOXP3⁺ T cells is predominantly found in cells lacking expression of Helios, both in mice and humans (67–70). Because transcription factors cannot be used for isolation of viable cells, there is a great need for markers that identify viable stable human Treg cells. Through combined transcriptomic and proteomic analyses (71), we found that the transmembrane protein GPA33 (72) is expressed by a subset of human Treg cells that exhibit the hallmarks of tTreg cells. These cells can efficiently be expanded in vitro in the absence of rapamycin without loss of salient Treg cell properties, allowing their study as well as yielding a pure product for Treg cell–based ACT.

The following Abs were used: anti-CD1a (clone HI149), anti-CD226 (clone DX11), anti-CCR7 (clone 150503), anti-CD4 (clone RPA-T4), anti-CD8 (clone RPA-T8), anti-CD127 (clone HIL-7R-M21), anti-CD25 (clone 2A3), and anti-CD45RA (clone HI100) from BD Biosciences; anti-CD27 (clone 0323), anti-TIGIT (clone A15153G), anti-CD127 (clone A019D5), anti–IFN-γ (clone B27), and anti–IL-17A (clone BL168) from BioLegend; anti-CD31 (clone WM59), anti-CD3 (clone OKT3), anti–IL-2 (clone MQ1-17H12), anti-FOXP3 (236A/E7), and anti-Helios (clone 22F6) from eBioscience; anti-GPA33 (clone 402104) from R&D Systems; anti-GPA33 from the Olivia Newton-John Cancer Research Institute, Heidelberg, AU (73), labeled with Alexa Fluor 647 Succinimidyl Ester (Thermo Fisher Scientific); and an in house anti-GPA33 (SQ-GPA33—available upon request).

Human materials were obtained in accordance with the Declaration of Helsinki and the Dutch rules with respect to the use of human materials from volunteer donors. Umbilical cord blood was obtained from the Sanquin cord blood bank, according to the Eurocord guidelines, with informed consent from the mothers. Donors were full-term infants of healthy mothers after uncomplicated pregnancy. Buffy coats from healthy anonymized donors were obtained after their written informed consent, as approved by Sanquin internal ethical board. PBMCs were isolated up to 72 h after blood harvesting, using a Ficoll-Paque Plus (GE Healthcare) gradient. CD4⁺ cells were extracted through magnetic sorting using CD4 microbeads (Miltenyi Biotec) according to the manufacturer’s instructions. For long-term storage in liquid nitrogen, samples were frozen in FCS with 10% DMSO. Sorting of viable CD4⁺ T cell subsets was done on a FACSAria III (BD Biosciences), based on surface expression of a live/dead marker, CD127, CD25, CD45RA, and/or GPA33.

Anonymous human thymus tissue was obtained from surgical specimens of children up to 3 y of age undergoing open heart surgery at the Leiden University Medical Centre. Informed consent was gathered in accordance with the Declaration of Helsinki and was approved by the Medical Ethical Committee of the Leiden University Medical Centre. The tissue was disrupted mechanically and pressed through a 100-μm mesh filter to make single-cell suspensions.

T cells were cultured at 10⁴ or 2 × 10⁴ cells per well in 96-well U-bottom plates (Greiner Bio-One) in IMDM + 10% FCS + 1% L-glutamine + 1% penicillin/streptomycin (T cell medium) for up to 14 d with or without 100 nM rapamycin (Sigma-Aldrich) and 300 IU/ml IL-2 (proleukin; Novartis). On day 0, cells were stimulated with soluble anti-CD3 mAb (PeliCluster, 0.1 μg/ml) and anti-CD28 mAb (0.1 μg/ml; eBioscience). Fresh medium with IL-2 with or without rapamycin was added on day 4 and 11 of culture. On days 7 and 14, cells were counted, harvested, partly used for experiments, partly split to the original starting concentration, and restimulated as above. For iTreg cell cultures, 10⁴ Tconv cells were cultured with anti-CD3/CD28 mAb and 10 ng/ml recombinant human TGF-β (Peprotech). On day 4, fresh T cell medium supplemented with TGF-β was added. Cells were harvested at day 7.

Surface staining of cells was done in PBS containing 0.5% FCS for 15 min at room temperature. To exclude dead cells from the analysis, Near IR (Life Technologies) was used as a live/dead marker. For intracellular staining of FOXP3, Helios, and cytokines, cells were fixed and permeabilized after surface staining with the FOXP3/transcription factor fixation/permeabilization buffers (eBioscience) following the manufacturer’s instructions. Data were acquired on an LSR II Cytometer (BD Biosciences) and analyzed using the FlowJo software (version 10; Tree Star). To measure cytokine production, cells were washed with culture medium and stimulated with 20 ng/ml PMA (Sigma-Aldrich) and 1 μM ionomycin (IM; Sigma-Aldrich) in the presence of brefeldin A solution (eBioscience) for 4 h at 37°C.

PBMCs were washed in PBS and labeled with 5 μM CellTrace Violet (Thermo Fisher Scientific/Molecular Probes). After 8 min, an equivalent volume of FCS was added, and cells were washed twice in IMDM/8% FCS. Labeled PBMCs were cocultured with expanded, and rested Treg cells and stimulated with anti-CD3 mAb (PeliCluster; 0.05 μg/ml). Proliferation was analyzed by flow cytometry after 4 d of coculture.

Sorted T cells were snap frozen in liquid nitrogen. Bisulfite conversion of DNA was performed using the EZ DNA Methylation-Direct Kit according to the manufacturer’s protocol (Zymo Research). Cell pellets were taken up in PBS, and 1 × 10⁴ cells were used for proteinase K digestion. Methylation-specific (MS) quantitative PCR (qPCR) was performed using iQ SYBR Green Supermix (Bio-Rad Diagnostics) (74). MS-qPCR was performed in 8 μl of reactions, containing 2 μl of the eluted bisulfite-treated DNA solution, 0.5 μM of MS or demethylation-specific primers for the FOXP3 Treg cell–specific demethylated region (TSDR) and 4 μl SYBR Green reagent. MS-qPCR was conducted at 98°C for 10 min, 40 cycles of 98°C for 15 s and 60°C for 1 min, followed by melt curve analysis. The methylation rate (in percentage) was calculated using the following formula: 100/(1 + 2^{Ct CG−Ct TG}), where Ct CG is defined as cycle threshold (Ct) values using MS primers and Ct TG defined as Ct values using demethylation-specific primers.

Cellular proteomes and transcriptomes were measured as described (71). Absolute protein abundance was estimated using the intensity-based absolute protein quantification (iBAQ) (75) approach, with a few modifications described elsewhere (76). Briefly, peptide intensities were summed up and divided by the number of observable peptides per protein. The iBAQ values were then normalized based on the total sum of all protein intensities to be able to compare abundances across samples. The iBAQ values of three technical replicates were averaged based on the median. Raw files, MaxQuant output files, and normalized iBAQ values are deposited at the Proteomics Identification Database repository (PXD005477; http://proteomecentral.proteomexchange.org). RNA sequencing data can be accessed with accession number GSE90600 (https://www.ncbi.nlm.nih.gov/geo).

Statistical tests were performed using Prism 6.0 (GraphPad Software). Group differences were assessed using Row-matched or normal one-way ANOVA, followed by Tukey honestly significant difference (HSD) multiple comparison test. A p < 0.05 was considered statistically significant at a 95% confidence level. Data are presented as mean ± SD. Significance levels are as follows: *p < 0.05, **p < 0.01, and ***p < 0.001, unless otherwise indicated in the figure legends.

CD25⁺CD127^– Treg cells in human blood can be subdivided into CD45RA⁺ naive and CD45RA^– effector Treg (eTreg) cells (77). Because pTreg cells derive from Tconv cells that have been activated under tolerogenic conditions, one would expect that such cells lack expression of CD45RA, the prototypical marker of naive T cells. The CD45RA⁺ Treg cell population might thus theoretically only contain tTreg cells (78, 79). However, in vitro–expanded cultures of CD45RA⁺ Treg cells (gating strategy in Fig. 1A) still contained a sizable population of FOXP3⁺Helios^– single positive (SP) cells and a small FOXP3^–Helios^– double negative (DN) population (Fig. 1B). Importantly, unlike FOXP3⁺Helios⁺ double positive (DP) Treg cells, a substantial proportion of these FOXP3⁺Helios^– SP cells produced IL-2, IL-17, and IFN-γ when stimulated with PMA + IM (Fig. 1C). The FOXP3⁺Helios^– SP cells could be activated Tconv that have upregulated FOXP3 (57) or genuine FOXP3⁺Helios^– SP Treg cells. To distinguish between these possibilities, we mixed Tconv cells from one donor with Treg cells from another donor (using an HLA-mismatch for identification) and examined their phenotype after coculture. Although some Tconv cells did acquire expression of FOXP3, its expression level was much lower than that found in the FOXP3⁺Helios^– SP cells derived from the Treg cell population (Fig. 1D). The CD45RA⁺ Treg cell population thus includes Treg cells that do not (stably) express Helios, possibly representing peripherally induced eTreg cells that have upregulated expression of CD45RA, as has been shown for some conventional terminally differentiated effector cells (80).

To select live FOXP3⁺Helios⁺ DP Treg cells, better cell surface markers are necessary. Such markers might be more abundant in CD45RA⁺ Treg cells, as this population is probably still more enriched for tTreg cells than the CD45RA^– Treg cell population. We therefore searched a dataset that we recently generated (PXD005477), in which we compared whole cell proteomes of five CD4⁺ T cell populations from human blood (71). These included naive Treg (nTreg) (CD4⁺CD25^intCD45RA⁺), eTreg (CD4⁺CD25^hiCD45RA^–), naive Tconv (nTconv) (CD4⁺CD25^–CD45RA⁺), memory Tconv (mTconv) (CD4⁺CD25^–CD45RA^–), and a population of CD4⁺CD45RA^– T cells with intermediate expression of CD25 containing a mixture of eTreg and mTconv cells similar to the previously described fraction 3 (Fr. III) (Fig. 2A) (71, 77). We found that a molecule called GPA33 is prominently enriched in nTreg cells, at both the mRNA and protein level (Fig. 2B). GPA33 is a transmembrane receptor of unknown function, belonging to the Ig-domain family, originally identified in human gastrointestinal tissue and tumors (72, 73). It is preferentially expressed in the nTreg cell population at levels that fall within the range of well-known T cell surface molecules such as CD27 and CD28 (Fig. 2C). Although the other CD4⁺ T cell populations also express this receptor, their GPA33 protein levels are much lower than those found in the nTreg cell population (Fig. 2C), and a similar pattern was found at the mRNA level (Fig. 2D).

Flow cytometry confirmed that GPA33 is highly expressed on most nTreg cells (Fig. 3A). Subsets of cells in the eTreg cell population as well as in Fr. III also expressed high levels of GPA33, whereas a lower level of expression was found in nTconv cells and some mTconv cells. In all Treg cell subsets, GPA33^high cells lacked CD127 and expressed FOXP3 and Helios (Fig. 3B). Selection for cells expressing GPA33 strongly enriched the FOXP3⁺Helios⁺ DP subpopulation from the CD4⁺CD25⁺ Treg cell population or from the CD4⁺CD25^intCD45RA⁺ nTreg cell population (Fig. 3C). In fact, the combination of just CD4, CD25, and GPA33 yielded a FOXP3⁺Helios⁺ DP Treg cell population of superior purity than obtained by the combination of CD4, CD25, and absence of CD127, which is commonly used to identify Treg cells (81). Nearly pure FOXP3⁺Helios⁺ DP could be obtained by selecting for high expression of GPA33 within the Treg cell gate (Fig. 3B, Supplemental Fig. 1). This enrichment was also greater than obtained by selection for CD4⁺CD25⁺CD127^– Treg cells expressing TIGIT or lacking expression of CD226 (Supplemental Fig. 1), two markers described to improve the purity of the FOXP3⁺Helios⁺ DP population (82, 83). Finally, absence of GPA33 identified CD4⁺CD25⁺CD127^– T cells able to produce IL-2, IFN-γ, and IL-17, whereas Treg cells expressing the highest levels of GPA33 mostly lacked this ability (Fig. 3D, Supplemental Fig. 2). Expression of GPA33 even identified cells unable to produce IL-2, IFN-γ, and IL-17 in Fr. III, the FOXP3⁺ population with the highest frequency of cells producing proinflammatory cytokines (71, 77). High expression of GPA33 thus marks CD4⁺CD25⁺CD127^–FOXP3⁺Helios⁺ Treg cells that are unable to produce Tconv effector cytokines. As such, selection for GPA33 may be useful for ACT. Treg cells from patients with type 1 diabetes do express GPA33, both immediately ex vivo and after 2 wk of expansion with a clinical grade protocol (Supplemental Fig. 3A, 3B). This shows that, in principle, GPA33 can be used to purify autologous Treg cells for treatment of such patients (23).

For ACT, Treg cells must be expanded in vitro. Among the various subsets, the CD45RA⁺ Treg cells possess the greatest capacity for in vitro expansion, far exceeding that of CD45RA⁻ Treg cells (77, 84). We reasoned that selective culture of GPA33^high cells (all of which are FOXP3⁺Helios⁺) from the CD45RA⁺ Treg cell population should eliminate the undesirable FOXP3⁺Helios^– and FOXP3^–Helios^– cells that are present after expansion of the entire CD45RA⁺ Treg cell population (Fig. 1B). To test this idea, nTreg cells were isolated based on different expression levels of GPA33 (Fig. 4A, top) and expanded for 2 wk with anti-CD3/CD28 mAb and IL-2. Cells expressing the highest level of GPA33 remained almost exclusively FOXP3⁺Helios⁺ (Fig. 4A, bottom). Cells expressing lower levels of GPA33 yielded progressively more FOXP3⁺Helios^– SP and even DN cells (Fig. 4A, bottom). A similar pattern was observed after culture of Fr. III and eTreg cells that had been sorted based on different GPA33 expression levels (Supplemental Fig. 2). Furthermore, the ability to produce effector cytokines, such as IL-2 and IFN-γ, after expansion, correlated inversely with the GPA33 expression level of the original nTreg cell population used (Fig. 4C). The GPA33^high nTreg cell population therefore appears stably committed to the FOXP3⁺Helios⁺ Treg cell fate, at least during in vitro culture. Interestingly, GPA33 itself was also maintained during expansion of GPA33^high nTreg cells (Fig. 4D), GPA33^high eTreg cells, and Fr. III cells (Supplemental Fig. 2). GPA33 persisted on GPA33^high nTreg cell clones for as long as monitored (90 d; data not shown). In contrast, Tconv cells lost GPA33 during culture (Fig. 4D). Moreover, expression of GPA33 did not appear on TGF-β–induced Treg (iTreg) cells (Fig. 4D). GPA33^high nTreg cells stably remained FOXP3⁺Helios⁺ and were unable to produce proinflammatory cytokines, despite the fact that no rapamycin was added to the cultures. Importantly, these expanded GPA33^high nTreg cells suppressed proliferation of CD4⁺ or CD8⁺ Tconv cells as potently as unfractionated CD4⁺CD25⁺CD127^– Treg cells that had been cultured in the presence of rapamycin (Fig. 5A), as is standard procedure for clinical grade Treg cell expansion (40). Similarly, their suppressive capacity was comparable to that of other Treg cell populations cultured without rapamycin but better than that of iTreg cells (Supplemental Fig. 4). GPA33^high nTreg cells also retained full CpG demethylation of the FOXP3 CNS2 TSDR (Fig. 5B), which is a hallmark of stable Treg cells (24, 85), again without a requirement for rapamycin.

Stringent selection for a GPA33^high nTreg cell population from adult blood yields fewer cells than isolation of the unfractionated CD4⁺CD25⁺CD127^– population. This disadvantage is, however, largely compensated by the much greater expansion that is possible in the absence of rapamycin (Fig. 5C). Selection for nTreg cells expressing high levels of GPA33 thus allows the isolation of Treg cells that can robustly be expanded in vitro and retain their salient properties without a need for rapamycin.

GPA33^high nTreg cells fit the criteria for tTreg cells: stable coexpression of FOXP3 and Helios, inability to produce Tconv effector cytokines, and demethylation of FOXP3 CNS2 TSDR. Furthermore, GPA33 is almost universally expressed on Treg cells from neonates, which have not yet been exposed to the commensal microorganisms and food Ags that induce differentiation of pTreg cells (Fig. 6A). To examine whether GPA33 identifies tTreg cells, we determined its expression on human thymocytes. CD3⁺ thymocytes have been divided into four subpopulations based on expression of CD1a and CD27, from least mature (CD1a⁺CD27⁻, gate I) to most mature (CD1a⁻CD27⁺, gate IV) (86) (Fig. 6B). Stages I (CD1a⁺CD27⁻) and II (CD1a⁺CD27⁺) contain mainly late CD4⁺CD8⁺ DP thymocytes. At stages III (CD1a^lowCD27⁺) and IV (CD1a⁻CD27⁺), thymocytes have committed to the CD4 or CD8 lineage (86). FOXP3⁺ Treg cells are found mostly in the more mature stages of development (86) (Fig. 6B). Although some Treg cells express GPA33 already at stage II, the percentage of GPA33⁺ cells gradually increases with maturational stage (Fig. 6B). In the final steps of thymic development, Treg cells acquire CD31 and, just before exit from the thymus, the CD45RA isoform that also marks nTreg cells in blood (86). Importantly, nearly all cells at this late CD31⁺CD45RA⁺ stage express GPA33, whereas the CD45RA⁻ cells are also mostly GPA33⁻ (Fig. 6C). Therefore, GPA33 is upregulated during the final stages of tTreg cell development, just before egress into the peripheral blood. Indeed, GPA33 is universally expressed on CD31⁺CD45RA⁺ RTE Treg cells in blood (Fig. 6D).

ACT with Treg cells can ameliorate autoimmune and pathologic conditions caused by alloreactivity after transplantation (17, 23). Many inflammatory disorders may be amenable to such treatments (35). However, the threshold to allow such therapies is high, given the existence of alternatives such as pharmacological immunosuppressants. ACT with Treg cells holds the potential for more effective mitigation of disease, with fewer side effects, than obtained with available drugs. Nonetheless, broad clinical implementation of ACT with Treg cells will only occur if this therapy is optimally safe. A key step in ensuring safety lies in removal of contaminating Tconv cells and unstable Treg cells (35). This is especially important for development of next generation therapies. The activity of Treg cells is controlled by their TCR (30–33, 87, 88). More potent regulation can be achieved with mono- or oligoclonal Treg cells expressing AgRs specific for the tissues that must be protected than with the currently used polyclonal Treg cell populations (28). Any Tconv cell carrying such a TCR, be it derived from contaminating Tconv cells or from unstable Treg cells, could provoke iatrogenic exacerbation of disease and should therefore be eliminated before infusion into patients.

Current protocols rely on rapamycin to limit the presence of Tconv cells in Treg cell populations cultured for ACT (39, 40). The use of this drug has several caveats. It does not preclude the presence of unstable Treg cells in expanded Treg cell populations and, in fact, induces generation of such cells (44). Furthermore, Treg cell–specific inactivation of mTORC1, the kinase complex that is inhibited by rapamycin, resulted in severe autoimmune disease in mice (43, 45). This suggests that rapamycin may even compromise functional properties of Treg cells. Ideally, therefore, Treg cell populations isolated for ACT should be sufficiently pure to obviate the need to use this drug.

Given their reported superior lineage fidelity (49, 51, 89), tTreg cells seem the most attractive cell type for ACT. A combined proteome/transcriptome database search (71) for markers that might allow their isolation led us to the Ig-domain protein GPA33 (72). A recent study found that FOXP3 binds to a region in the GPA33 gene, already suggesting that it might be expressed in Treg cells (90). We, in this study, showed that this protein is not expressed by all Treg cells but by a subset that has the characteristics of tTreg cells. The GPA33 protein is most closely related to CD2, and has relatives with important functions in T cells, such as CD276 (B7-H3) and VSIG4 [both inhibitors of T cell activation (91, 92)] and several cell adhesion molecules such as CEACAM, ICAM, and NCAM (3D-Protein Basic Local Alignment Search Tool, National Center for Biotechnology Information). GPA33 has been studied as a marker of intestinal epithelium and is highly expressed on some gastric and most colon carcinomas (73, 93). GPA33-deficient mice exhibit increased incidence of colitis and food intolerance (94), and variants at the GPA33 locus are associated with the rare autoimmune disorder eosinophilic granulomatosis with polyangiitis (P. Lyons, J. Peters, F. Alberici, J. Liley, R. Coulson, W. Astle, C. Baldini, F. Bonatti, M. Cid, H. Elding, et al., manuscript posted on bioRxiv). Although these results might reflect a role for GPA33 in epithelial barrier function (94), the expression of this molecule in Treg cells raises the possibility that a more direct role in T cell tolerance exists.

Although the function of GPA33 is not known, we found that it serves as a good marker to select pure populations of Treg cells. The purest FOXP3⁺Helios⁺ DP Treg cell populations are obtained by selecting GPA33^high nTreg cells. Upon expansion without rapamycin, such cells stably retain expression of FOXP3 and Helios as well as full demethylation of the TSDR in the FOXP3 gene. They furthermore abstain from producing effector cytokines and are fully suppressive. Although selection for GPA33^high nTreg cells leads to relatively low yields, this disadvantage is largely offset by their superior expansion because of the absence of rapamycin. Moreover, the need for high cell numbers likely diminishes when Ag-specific Treg cells are used because of their greater potency (28). Nonetheless, when larger populations of Treg cells are required, selecting GPA33⁺ cells from the entire CD4⁺CD25⁺CD127^– or even from the CD4⁺CD25⁺ population combines high yield with increased purity of FOXP3⁺Helios⁺ DP Treg cells.

Our results suggest that GPA33 identifies Treg cells of thymic origin. It is expressed on almost all thymic CD31⁺CD45RA⁺ Treg cells, the population that is about to egress into the periphery (86), as well as on most peripheral nTreg cells, which should, theoretically, predominantly consist of tTreg cells. Moreover, GPA33 is universally expressed on Treg cells in cord blood, which are presumably all thymic-derived, as neonates have not been exposed in utero to the commensals and food Ags that drive differentiation of pTreg cells (95). Under no condition did we observe acquisition of GPA33 expression on Treg cells that lacked this marker initially and GPA33 does not appear on iTreg cells generated in vitro. It should be noted that some Tconv cells do express this receptor. GPA33 is, thus, not a unique Treg cell marker on its own, as is true for any other marker used to identify such cells, including FOXP3 (57) and Helios (61). However, Tconv cells lose expression of GPA33 after TCR-mediated activation, whereas this molecule remains stably expressed on Treg cells. This raises the possibility that GPA33 may be a lineage marker for tTreg cells. Consistent with that idea, GPA33 is found on a subset of cells in the CD45RA⁻ Treg cell populations (eTreg cells and Fr. III), arguably mixtures of pTreg cells and activated tTreg cells (96, 97). The GPA33^high cells in these fractions are universally FOXP3⁺Helios⁺ DP and lack ability to produce effector cytokines.

In conclusion, we have described GPA33 as a surface marker which, when used together with other Treg cell markers, allows isolation of pure and stable Treg cells, displaying the hallmarks of tTreg cells. On the basis of our findings, we consider it attractive to incorporate high expression of this marker as a selection criterion to generate clinical grade Treg cells for ACTs.

We thank Mark Hazekamp, Maaike van Wieringen-Deckers, and the rest of the thoracic surgical staff at the Leiden University Medical Center for providing human thymus material. We would also like to thank Erik Mul, Simon Tol, and Mark Hoogeboezem for helping with FACS, Evert de Vries for fluorescent labelling of the Ab to GPA33, and René van Lier and Mark Mensink for critical discussions and comments on the manuscript.

A patent application for the use of GPA33 to select Treg cells for clinical use has been filed by D.A., J.B., and E.C. The other authors have no financial conflicts of interest.