T Cell Lineage Commitment: Identity and Renunciation

Multipotent hematopoietic cells gain access to the T cell developmental pathway and then confirm that choice of fate by “burning their bridges” to other pathways. The process involves a continuing dialogue between the differentiating cell and the signals coming from its environment. The best understood of these signals are effects of Notch1 engagement with Delta ligands provided in the environment of the thymus, which are needed for T cell specification throughout the lineage-choice process and only become dispensable after T cell fate is confirmed (1–3). In vivo, expression of Notch ligand with supportive cytokines, such as IL-7 and Kit ligand, gives the thymus its T cell inductive activity (3). However, the cells that begin the T cell program initially have access to other options. These can be revealed if the cells are removed from the thymic environment and challenged with different environmental signals. Only after they lose the ability to make these alternative responses are the cells committed.

Commitment is the cell-intrinsic regulatory transformation through which a cell’s alternative potentials are eliminated. “Potential” is not the same as the default fate that a cell adopts in an undisturbed condition. “Fate” is the intersection between the cell’s potential and the permissive or nonpermissive circumstances in which a cell finds itself. For a typical uncommitted cell, the fates that its progeny actually adopt in vivo may be only a slice of the uncommitted cell’s full potential, because environmental conditions are limiting. As the cell progresses toward commitment, its potential shrinks. At the point where potential, rather than environment, becomes limiting for fate, the cell is committed. Interestingly, one of the few cases where a “fate” assay approaches a “potential” assay is in the case of hematopoietic stem cells, where extremely long assay times are conventionally used, extremely large numbers of progeny are generated before the end point, and the intermediates include highly motile cells that can sample an unusually large number of in vivo environments before differentiating as scored at the assay end point. However, this exception proves the rule that multipotent cells can only show the subset of their potentials that their environments support.

If fate is different from potential, then why study potential at all (4, 5)? “Potential” reveals what the regulatory machinery of the cell itself, transcription factors and signal receptors, contributes to its identity. The commitment process establishes an irreversible, portable, unconditional identity by actively narrowing a cell’s range of potentials. Exclusion of any particular fate can occur by repression of a key transcription factor or of a signaling receptor that would be needed to induce that pathway in permissive conditions. For example, T cell potential itself depends on Notch1 expression. During B-lineage commitment, T cell potential is excluded by Pax5’s silencing of the expression of Notch1 (6). Ultimately, commitment to a lineage is the readout for the gene-regulatory network that links positive differentiation in one pathway with the controlled silencing of particular transcription factors and signaling receptors needed for other pathways.

Normally, developmental potentials of two populations are compared by testing cells under uniform conditions that are clearly permissive for the fates being assayed. However, there are caveats. When hematopoietic precursors are adoptively transferred i.v., the protocol, in effect, “hands over” to the grafted cells themselves the ultimate choice of the environment into which they will lodge. The choice is mediated by adhesion molecules and chemokine receptors on the cells that can bias migration to particular microenvironments, and these “traffic control” molecules are themselves developmentally regulated. For example, the chemokine receptors CCR9 and/or CCR7 are needed for hematopoietic precursors to enter the thymus; double mutants are dramatically impaired as T cell precursors in vivo (7–9). The abilities of cells to home predictably to particular environments in vivo are biologically important and especially relevant for therapeutic use of uncommitted cells in clinical practice (5). However, these “traffic control” molecules are not needed under permissive conditions to program T cell identity itself (10).

Thus, changes in cell potential need to be explained by mechanisms operating at three distinct levels. At the cell-intrinsic level, potential is defined by transcription factors and possible epigenetic constraints. At the opposite extreme, in cases where the environment is not experimentally controlled, the cell’s opportunities may be biased by its migration preferences. Between these levels, at the cell/environment interface, triggering of potential depends on expression of signaling receptors that modify transcription factor activity in response to inductive environmental signals.

Hematopoietic stem cells begin with access to ≥10 distinguishable fates, including erythroid and megakaryocytic, macrophage, neutrophilic granulocyte, eosinophil, basophil, mast cell, conventional dendritic cell (DC), plasmacytoid DC, NK cell, T lymphocyte, and B lymphocyte, as well as their subtypes. In the T cell pathway, there is an ordered loss of access to these fates. Thus, commitment emerges as the end of a sequential process of lineage exclusions, summarized in Fig. 1. Before arrival in the thymus, precursors have lost the ability to make erythroid cells and megakaryocytes, but the most immature cells in the thymus clearly retain access to DC, NK cell, macrophage, and granulocyte, as well as probably mast cell and B cell potential, in addition to T cell potential. The B cell potential is rapidly extinguished (11). In fetal mouse T cell development, B potential is lost even before the cells arrive in the thymus (12, 13). However, NK, DC, and even macrophage and granulocyte (“myeloid”) potential remain evident until a later intrathymic stage, in fetal and postnatal cases, if the cells are tested under permissive conditions. These NK and DC potentials disappear at a distinct, later transition between the double negative (DN)2 (c-Kit⁺ CD44⁺ CD25⁺) and DN3 (c-Kit^low CD44⁻ CD25⁺) (14–18) stages, more precisely mapping between the DN2a (c-Kit⁺⁺) and DN2b (c-Kit⁺) stages (19, 20). Single cells in the DN2 (DN2a) stage can still give rise to clones, including T cells, as well as NK cells, macrophages, DCs, or even granulocytes (15, 16, 18, 21, 22); in contrast, cells one stage later (DN2b stage) can no longer do this (19, 20). This last step of alternative fate exclusion completes T-lineage commitment.

T-lineage gene expression is fully activated by the DN3 stage (20). The genes required for TCR expression and function are turned on asynchronously, depending on gene-specific combinations of inputs from the transcription factors that are needed for early T cell development, including GATA-3, TCF-1, basic helix–loop–helix E proteins E2A and HEB (Tcf12), Runx/CBFβ complexes, and the Notch-activated transcription factor RBPJκ (CSL) (23). In detail, the induction of T cell genes remains poorly explained. None of the essential factors seems dominant in the way that GATA-1 is for erythroid genes (24). However, these factors also participate individually in aspects of lineage exclusion, as described below.

As long as immature thymocytes remain undisturbed in the thymus, their prevalent fate will normally be to generate T cell progeny, with few NK, dendritic, or myeloid progeny and no B cell progeny (25, 26). Notch signaling makes this in vivo environment nonpermissive for display of B, myeloid, and NK alternative potentials. Notch can block these alternative pathways, even if the cells are forced to express ectopic transcription factors of other lineages that would otherwise impose lineage conversion (18, 27–29). Thus, it is only when cells are removed from Notch signaling that distinct levels of restriction are seen. B-lineage repression becomes independent of Notch signaling within a few cell divisions after thymic entry (11), when the cells are still in the initial early T cell precursor (ETP/DN1, c-Kit⁺⁺ CD44⁺ CD25⁻) compartment (Fig. 1), although Notch is still needed at that stage to keep myeloid potential at bay. Myeloid- and dendritic-lineage exclusions only become Notch independent later, at the DN2b stage (19, 20) during commitment.

Hematopoietic precursors, in the bone marrow and in the early intrathymic stages, simultaneously express some levels of transcription factors that participate in disparate fates, as well as many of the target genes associated with these contradictory pathways (30, 31). Indeed, ETPs share certain transcription factors with the ensembles that guide differentiation of B and myeloid cells. T cells share with B cells the use of E proteins, Myb, Runx1/CBFβ, and Ikaros, and both initially share with myeloid cells the use of the avian erythroblastosis virus E26 oncogene homolog (ETS)-family factor PU.1. In the myeloid case, potential seems to be controlled through effects on the shared factors that make a “bridge” from the T cell pathway.

Myeloid potential reflects activity of the bZIP transcription factor C/EBPα, together with the ETS-family factor PU.1 (32), which is also essential for generation of lymphoid-primed multipotent precursors (33, 34). These play roles in virtually all myeloid cells, can upregulate each other (35), and together can convert even fibroblasts directly into macrophages (36). Myeloid growth factor receptor signaling from the environment can promote myeloid differentiation, in part by posttranslationally enhancing PU.1 activity (37–39). Because the myeloid growth factor receptors themselves are among the major target genes of PU.1 activation (40, 41), PU.1 enables environmental myelopoietic growth factors to activate a positive-feedback circuit to drive myeloid development. Both myeloid factors are present, or inducible, at stages when myeloid potential of thymocytes is highest. Prethymic precursors, as well as ETP and DN2a pro-T cells, express moderately high levels of PU.1 (20, 34). As the cells undergo commitment, PU.1 is downregulated, with expression decreased by ∼100-fold by the time cells reach the DN3-stage checkpoint (20). Although C/EBPα expression is low in ETPs and silenced soon afterward, lineage-tracing experiments confirm that many T cells are indeed generated from C/EBPα⁺ precursors as well (42). Adding back PU.1 or C/EBPα to committed, DN3-stage thymocytes restores myeloid potential; in the absence of Notch signals, either factor can promote the reactivation of the other (27, 28, 43, 44). Thus, these factors provide a clear way to understand why the ETP and DN2a thymocytes can still differentiate into myeloid cells, whereas DN2b and later-stage cells normally cannot.

PU.1 is also crucial for development of almost all types of DCs, in part because of its regulation of the growth factor receptor Flt3 (41). Although conventional and plasmacytoid DCs develop from different programs that diverge sharply in their use of E proteins, both depend on PU.1 for generation (41) and require expression thereafter of PU.1-subfamily ETS factors, PU.1 or SpiB, respectively (45). The ability of PU.1 and SpiB to support DC development can even be detected in the presence of some continuing Notch signaling (43) (M.M. Del Real and E.V. Rothenberg, unpublished observations), when C/EBP factors are not activated. Thus, it is not surprising that DC potential in thymocytes also closely tracks the natural expression of PU.1.

C/EBPα expression in thymocytes is limited by Notch signaling (46), possibly through the Notch-induced repressor Hes1 (47). It probably begins to be inhibited soon after thymic entry, because Notch targets, including Hes1, are active in pro-T cells throughout the ETP to DN3a stages. Recent data from our laboratory’s genome-wide epigenetic mapping studies confirm that repression machinery has already begun to be recruited to the Cebpa gene, based on H3K27me3 histone marks, as early as the ETP/DN1 stage (J. Zhang, A. Mortazavi, B. Williams, B. J. Wold, E.V. Rothenberg, unpublished observations); however, as noted above, it can still be reactivated at a later stage if PU.1 levels are high and Notch signals are removed. Under normal conditions, its expression limit may be set by the increasing dependence of pro-T cells on Notch signals for viability as they pass the DN2a to DN2b transition (see later discussion) (20).

The mechanism that represses PU.1 between the DN2a and DN3 stages is different. Two cis-regulatory regions of the Sfpi1 (PU.1) gene seem to mediate repression: a multifunctional Upstream Regulatory Element (48, 49) and a recently described, cell-type–specific silencer (50). Runx1/CBFβ complexes seem to repress at both elements. Two other factors may also contribute to PU.1 repression: TCF-1 (encoded by the Tcf7 gene) and Gfi1, both of which can work through sites near the Upstream Regulatory Element (51, 52). Finally, GATA-3 seems to antagonize PU.1 expression with a very sensitive dose response, although the mechanism may not be direct (29) (D.D. Scripture-Adams, A.M. Arias, K.J. Elihu, M. Zarnegar, and E.V. Rothenberg, unpublished observations). The details of the mechanism explaining the timing, magnitude, and near-permanence of the repression need further study, but it is notable that GATA-3, Runx1/CBFβ, and TCF-1 play direct roles in the positive regulation of T cell genes as well.

It is revealing to compare these lineage-choice mechanisms in early T cells with the gene-regulatory networks that have been proposed to explain other cell fate choices, where each cell type requires PU.1 but at different levels (52, 53). In myeloid- versus B-lineage fate choice, access to the B cell fate seems to depend on PU.1 restrained by Ikaros plus E2A (52). The switch function at the heart of this network consists of “myeloid” Id2 and Egr2 on one side, induced by high-level PU.1, with Gfi1 on the other, which is turned on by E2A and Ikaros and itself helps to limit PU.1 expression. Superficially, the early pro-T cell regulatory environment seems like that of the early B cell, with strong expression of Ikaros and Gfi1, steady activity of E proteins, and little C/EBPα, Egr2, or Id2. However, although early B and early T cells need to restrain PU.1 using shared lymphoid factors, in one respect their two networks give quite different results as PU.1’s activity is antagonized. In early T cells, the B cell option has already been closed off irreversibly, even before Runx1, TCF-1, and Gfi1 help to silence PU.1.

How can the B cell fate be excluded so early? The B cell fate should be connected to the T cell program by an overabundance of “regulatory bridges.” B cells share with T cells the use of RAG-dependent gene rearrangement to assemble the Ig superfamily immune receptors, a common sequence of immune receptor-dependent developmental checkpoints, and extensive analogies in morphology and signal processing, as well as sustained requirements for transcription factors, including E2A and Ikaros, and IL-7Rα/γc signaling. Furthermore, although the thymic stroma does not produce myeloid growth factors normally, it provides a rich source of the IL-7 that B cell precursors need to grow (3). This makes it easy to understand why the thymus fills with B cells when Notch signaling is impaired (26, 54). However, even after just a few days of contact with Notch signals, thymus-settling precursors lose the ability to make B cells, even when switched to B cell conditions, whereas their ability to make myeloid cells, under optimal in vitro conditions, remains intact.

The B-lineage exclusion mechanism must inactivate some crucial B cell function that T and B cells do not share. The B cell-specifically expressed factors EBF1 and Pax5, both essential for B cell development in a mutual positive-feedback loop (55), are likely to be the crucial control points. Neither of these factors is detectably expressed in any early thymocyte or in vitro differentiated pro-T cell population from the ETP stage onward (56–58). Heavy histone H3K27me3 marks, deposited across these loci as early as the ETP stage, also imply that these genes are silenced epigenetically (J. Zhang, A. Mortazavi, B. Williams, B.J. Wold, and E.V. Rothenberg, unpublished observations). This correlates well with the cells’ inability to switch to the B cell fate from the later ETP stage onward.

Nevertheless, it is not yet fully clear how Notch signaling terminates access to the B cell fate. Although there seems to be a mechanism through which Notch signaling temporarily interferes with EBF1 function in EBF1-expressing cells (59), the permanent transcriptional silencing of the Ebf1 and Pax5 loci needs another explanation. This is not an immediate Notch response like Hes1 induction, because several days of Notch signaling are required for full loss of B cell potential (11, 60). It is more coordinated with upregulation of GATA-3 and TCF-1, the first T cell-specific transcription factors to be induced (60). Even low-dose GATA-3 seems to be highly antagonistic to the B cell fate (29, 61–63) (D.D. Scripture-Adams, A.M. Arias, and E.V. Rothenberg, unpublished observations). However, GATA-3 may not antagonize the B-lineage program alone, because Notch signaling can also block B cell development in vitro from GATA-3–deficient cells (64). This GATA-3–independent effect of Notch could indicate the presence of another T-lineage–specific B cell-exclusion system.

The loss of NK potential, as measured by the ability to generate NK1.1⁺, DX-5⁺, perforin⁺ cells lacking T cell markers when supportive cytokines are provided, occurs after the DN2a stage, similarly timed but probably slightly later than the loss of myeloid and dendritic potential (18–20). NK cell fate is now understood as a complex cluster of programs (65), but, in general, they all have a great deal in common with T cell subset programs. Most of the transcription factors that NK cells use are shared with T cells, and there is a thymically derived lineage of NK cells that makes use of GATA-3 and IL-7R, very much like T cells (66). Although genes expressed by NK cells are often used as markers for NK cell fate choice, most are also expressed by invariant NKT cells and some TCRγδ cells after their TCR-dependent selection. Thus, the problem with defining how T cell precursors lose NK potential is that it is uncertain whether access to the NK program is ever fully lost at all or is simply restrained until a later stage when it can be redeployed for cytolytic T cell functions.

For years, the only candidate master regulator for NK cells was the helix–loop–helix antagonist Id2. This was a frustrating candidate. It is also used in later T cell development and is not a DNA-binding factor itself at all, rather just a decoy that blocks the ability of basic helix–loop–helix transcription factors (E2A, HEB) to bind to the DNA. Thus, Id2 itself has no direct regulatory targets that could explain the network of NK differentiation. However, a great advance has been the discovery of a true sequence-specific DNA-binding factor, Nfil3 (E4bp4), which has a crucial role in the programming of NK cells, including induction of Id2 itself (67, 68; reviewed in Ref. 61). Furthermore, at least one transcription factor was described recently (Zfp105), which may be the most specific NK cell regulator yet (69). These factors are not part of the normal adult T cell precursor regulatory repertoire; therefore, their regulation can be considered a key to NK potential (61).

Surprisingly, it is the NK pathway for which a T-lineage–specific, dedicated negative regulator has emerged first. This is the highly T-lineage–specific transcription factor, Bcl11b (70–72), whose expression is induced immediately before the loss of access to the NK option. Acute deletion of Bcl11b channels ETPs into an NK fate under T cell conditions and allows fully committed T cell precursors to back-differentiate to NK cells or NK-like effectors if deletion is postponed to a later stage (72). If cells delete Bcl11b just during β-selection, there is a severe decrease in their life spans as CD4⁺ CD8⁺ thymocytes, associated with aberrant activation of a number of mature cytolytic effector genes (73), as well as other precociously activated maturation regulators, suggesting an NKT-like fate (74). Bcl11b loss allows the innate-immune cell transcription factors Nfil3 and Zbtb16 (PLZF) to be upregulated, as well as the highly NK-specific regulatory gene Zfp105 (71, 72). Bcl11b also may repress Id2 via direct binding (74). Within the T lineage, Bcl11b is expressed continuously from the DN2b stage onward. Interestingly, its levels of expression vary and are lowest in activated effector CD8 cells and NKT cells. NK cells also express some Bcl11b but at lower levels still (http://www.immgen.org) (75). Thus, Bcl11b seems to be needed as a continuing restraint on the deployment of NK-like functions during early T-lineage commitment and throughout all later T cell development.

Bcl11b’s effects extend more broadly than to repress NK functions. One of the phenotypes that is most striking in Bcl11b knockout pro-T cells is that they can also continue apparent self-renewal in a precommitment DN2a-like state, growing better than wild-type cells in the presence of Notch signals in vitro but retaining DN2-like properties rather than progressing in differentiation (70, 71). Like normal DN2a cells, they retain access to myeloid fates (if myeloid growth factors are supplied) but not B-lineage fates, even after weeks of exposure to strong Notch signals (70, 71). They preserve expression of PU.1 under these circumstances (70, 71), as well as preserve expression of a cluster of stem/progenitor-associated regulatory genes that are expressed in early thymocytes and normally downregulated during commitment (20). These genes are otherwise implicated in aspects of precursor self-renewal (71) and seem to define an initial subprogram for pro-T cell expansion (23). Thus, Bcl11b normally both restrains the NK option and helps to end a discrete, precommitment expansion program (Fig. 1, “phase 1”). Notably, the termination of this phase 1 program is tightly correlated with an abrupt transition to extreme Notch dependence for survival (76), coinciding with commitment at the DN2b stage (20). The shift to acute Notch dependence itself can contribute to the timing of commitment, because it effectively applies a death penalty to any cell, from DN2b through β-selection, which ventures to sample an environment free of the developmental constraints of Notch signals.

The use of distinct repression mechanisms to block different alternative potentials makes the T-lineage program quite versatile in its ability to use input cells from different prethymic commitment states. If the suite of transcription factors activated by Notch signals can separately exclude the B cell fate, whether the myeloid fate has been excluded before or not, and separately block the myeloid fate, whether the B cell fate has been excluded or not, then there is no need to postulate only one true pipeline of prethymic cells feeding the thymus. This is important because there is evidence that a range of different precursors can provide input into the T cell program (reviewed in Ref. 77; 78–80), although in vivo they may reach the thymus differently and undergo differentiation versus self-renewal expansion with different kinetics.

Although these different prethymic precursors share NK and DC potential, and B cell potential if they come from postnatal animals, they differ in their perceived ability to make myeloid cells. Thus, CCR9⁺ lymphoid-primed multipotent precursors (LMPPs) have strong myeloid activity, as well as lymphoid activity, whereas common lymphoid precursors (CLPs) (81) and other lymphoid-biased early lymphoid precursor subsets (82–84) seem to have much stronger B, NK, and T lymphoid potential than granulocyte or macrophage potential, at least in vivo. But this seems to pose a logical contradiction: if many thymus-settling cells have already lost myeloid potential, how can myeloid potential also persist after B cell potential is lost?

Heterogeneity of thymus-immigrating cells can provide one answer: the myeloid potential may be persisting only in descendants of those immigrants that were not CLPs. However, another answer is that the myeloid potential in CLPs is not actually lost but rather is under constraint. In fact, CLPs are also highly efficient at generating myeloid cells, similar to intrathymic ETP and DN2a cells, provided that they are tested similarly in optimal in vitro conditions (5, 25). The largest differences in myeloid potential between CLPs and LMPPs are seen when the cells are assayed in vivo by adoptive transfer, in which they are allowed to determine their own homing to distinct microenvironments. Thus, a strong possibility is that “traffic control” mechanisms specific to CLPs help to make them lymphoid restricted in vivo, by directing them to lymphopoietic environments that are not fully permissive for myeloid development.

This is more plausible than it might appear, because even the bone marrow itself now seems to contain distinct domains that are permissive or restrictive for B lymphoid versus myeloid differentiation (85). In vivo, a G protein-coupled receptor sensitive to pertussis toxin seems to keep CLPs within the lymphopoietic domain and keep them out of the myeloid domain. This localization is important for their behavior as lymphoid-restricted precursors, because these cells generate myeloid progeny instead, if pertussis toxin is used in vivo to block their attraction to B lymphopoietic domains (85). The myeloid growth factor receptor-triggered positive feedback with PU.1 provides a strong candidate mechanism that may be needed for an environment to be myeloid permissive. In contrast, bone marrow lymphopoietic zones that promote B and NK development are likely to protect the cells from these myeloid growth factor receptor-like signals and TLR signals, either of which can readily induce myeloid differentiation from CLP (86–88). Thus, the nature of the mechanisms that provide the first restraint on myeloid differentiation of prethymic lymphoid precursors in vivo may be powerful in vivo, but qualitatively different from the cell-intrinsic, transcriptional regulatory circuits that later extinguish myeloid potential by silencing first EBF1 and Pax5, and then PU.1 and C/EBPα expression during T-lineage commitment.

Lineage commitment in the T cell system might have been a single event driven by a single factor, like erythroid commitment driven by GATA-1, which combines positive- and negative-regulatory activities in a single potent factor (24). But there has been no evidence to support this. Instead, the successive losses of different options for T cell development depend on the operation of molecularly distinct exclusion mechanisms. The environmental effects of Notch signaling are applied only during the commitment process, to activate the T cell program. Then multiple regulators that are likely contributors to commitment, GATA-3, Bcl11b, TCF-1, and members of the Runx family, are turned on and remain active in all T cell subsets at various levels thereafter. The roles of these factors in commitment may each be more to exclude a particular alternative than to cause “T-lineage commitment” in toto. For example, GATA-3, which helps to block B cell development, can even enhance myeloid development from certain precursors (89). T cell commitment overall seems to be a mosaic of separable, though interlocking, parts.

Much debate has been focused on the right ways to consider hierarchies of relatedness in the lymphoid lineages based on the order of lineage exclusion. But mechanisms of exclusion that cause loss of a given developmental potential simply attack the points of vulnerability of the alternative program. They work based on the minimum number of repression targets needed to block access to that fate, not on the number of shared regulatory strands that otherwise make two programs related. Therefore, commitment mechanisms need not operate in order of a hierarchical “relatedness index” between two fates. The large number of shared genes between B and T fates leave points of vulnerability in the B cell-specific need for EBF1 and Pax5, as well as the T cell-specific need for Notch1, and these genes become foci for mutual B/T-lineage exclusion. Conversely, the persistence of myeloid and dendritic options in early thymocytes may not reflect a hierarchical proximity of these fates to the T cell program, but rather a side effect of the versatility of PU.1. Despite providing a regulatory bridge to myeloid fates, PU.1-driven Flt3 expression and other growth-promoting effects are crucial for development of LMPP cells and possibly all prethymic cells. Myeloid potentials, easily checked by Notch signals in vivo, may simply be a temporary price paid for some other useful role for PU.1 in the self-renewing “phase 1” regulatory state of the earliest thymocytes.

If different lineage exclusions are mediated by distinct components of the T cell program, each with its own separate regulation, then the idea that different decisions can be made in nonhierarchical or even variable order (e.g., between fetal and adult precursor cohorts) is not so problematic. We have much to learn about what controls the activation and coordination of different parts of the T cell program, including the mechanisms of action of factors that play a role in lineage exclusion. However, the identity of T cells emerges not from a simple command switch but from the network of these factors’ interactions.

The author apologizes to many colleagues whose work could not be adequately cited due to space limitations.

The author has no financial conflicts of interest.