TCRβ Allelic Exclusion in Dynamical Models of V(D)J Recombination Based on Allele Independence
This information is current as Etienne Farcot, Marie Bonnet, Sébastien Jaeger, Salvatore of October 1, 2021. Spicuglia, Bastien Fernandez and Pierre Ferrier J Immunol 2010; 185:1622-1632; Prepublished online 28 June 2010; doi: 10.4049/jimmunol.0904182 http://www.jimmunol.org/content/185/3/1622 Downloaded from
Supplementary http://www.jimmunol.org/content/suppl/2010/06/28/jimmunol.090418 Material 2.DC1 http://www.jimmunol.org/ References This article cites 63 articles, 20 of which you can access for free at: http://www.jimmunol.org/content/185/3/1622.full#ref-list-1
Why The JI? Submit online.
• Rapid Reviews! 30 days* from submission to initial decision
• No Triage! Every submission reviewed by practicing scientists by guest on October 1, 2021
• Fast Publication! 4 weeks from acceptance to publication
*average
Subscription Information about subscribing to The Journal of Immunology is online at: http://jimmunol.org/subscription Permissions Submit copyright permission requests at: http://www.aai.org/About/Publications/JI/copyright.html Email Alerts Receive free email-alerts when new articles cite this article. Sign up at: http://jimmunol.org/alerts
The Journal of Immunology is published twice each month by The American Association of Immunologists, Inc., 1451 Rockville Pike, Suite 650, Rockville, MD 20852 Copyright © 2010 by The American Association of Immunologists, Inc. All rights reserved. Print ISSN: 0022-1767 Online ISSN: 1550-6606. The Journal of Immunology
TCRb Allelic Exclusion in Dynamical Models of V(D)J Recombination Based on Allele Independence
Etienne Farcot,*,1 Marie Bonnet,† Se´bastien Jaeger,† Salvatore Spicuglia,† Bastien Fernandez,* and Pierre Ferrier†
Allelic exclusion represents a major aspect of TCRb gene assembly by V(D)J recombination in developing T lymphocytes. Despite recent progress, its comprehension remains problematic when confronted with experimental data. Existing models fall short in terms of incorporating into a unique distribution all the cell subsets emerging from the TCRb assembly process. To revise this issue, we propose dynamical, continuous-time Markov chain-based modeling whereby essential steps in the biological procedure (D-J and V-DJ rearrangements and feedback inhibition) evolve independently on the two TCRb alleles in every single cell while displaying random modes of initiation and duration. By selecting parameters via fitting procedures, we demonstrate the capacity of the model to offer accurate fractions of all distinct TCRb genotypes observed in studies using developing and mature T cells Downloaded from from wild-type or mutant mice. Selected parameters in turn afford relative duration for each given step, hence updating TCRb recombination distinctive timings. Overall, our dynamical modeling integrating allele independence and noise in recombination and feedback-inhibition events illustrates how the combination of these ingredients alone may enforce allelic exclusion at the TCRb locus. The Journal of Immunology, 2010, 185: 1622–1632.
n developing T and B lymphocytes, allelic exclusion typically allelic exclusion continues to puzzle immunologists. For instance, restricts the assembly by V(D)J recombination of a pro- the mere fact that a small group (#5%) of allelically included http://www.jimmunol.org/ I ductively rearranged (in-frame) variable exon to only one al- abT cells carrying two productively rearranged (VDJ+/VDJ+) lele of, respectively, TCR and Ig genes. The resulting allelically TCRb alleles eventually develop alongside the overwhelming mass excluded (e.g., abT cells) commonly display a TCRb genotype ($95%) of allelically excluded cells still eludes a comprehensive made of a combination of one in-frame assembled allele (hence- explanation (2). + forth denoted VDJ ) and either one unrearranged germline (GL) Two types of modeling theories have prevailed in an attempt to 2 or partially (DJ)-rearranged allele or one out-of-frame (VDJ ) tackle this conundrum (3). The so-called stochastic models com- rearranged allele (1). Consequently, at the phenotypic level, the monly considered a low probability p for an in-frame joint to occur b TCR proteins expressed by these cells are encoded at a single due to inefficiency in either the onset or achievement of recombi- by guest on October 1, 2021 chromosome, a feature that may contribute in preserving the nation (4–6). With the probability for allelic inclusion given by the working of an adaptive immune system founded on clonal-cell square p2, the models were able to account for the rare appearance selection procedures. Despite years of efforts, the phenomenon of of allelically included cells. Soon afterward, however, these sim- plistic views appeared incoherent with a mass of experimental *Centre de Physique The´orique, Centre National de la Recherche Scientifique Unite´ findings, including those of a vast majority of T cells carrying Mixte de Recherche 6207, Universite´ de la Me´diterrane´e-Universite´ de Provence- b Universite´ Sud Toulon Var, Centre National de la Recherche Scientifique Luminy dually rearranged TCR alleles, and B cells carrying dually rear- + Case 907; and †Centre d’Immunologie de Marseille-Luminy, Centre National de la ranged Ig H chain alleles as well, to a ratio of ∼60% VDJ /DJ to Recherche Scientifique Unite´ Mixte de Recherche 6102-Institut National de la Sante´ 40% VDJ+/VDJ2 cells (not mentioning the few VDJ+/VDJ+- et de la Recherche Me´dicale U631-Universite´ de la Me´diterrane´e, Campus de Luminy Case 906, Marseille Cedex 9, France equipped cells; see Refs. 1, 2, 7). Thus, purely stochastic models 1Current address: Institut National de Recherche en Informatique et en Automatique, were found to insufficiently explain allelic exclusion. Virtual Plants Team, Cooperation Centre for Agronomic Research in Development/ Feedback inhibition is the hallmark of the current regulated Unite´ Mixte de Recherche De´veloppement et Ame´lioration des Plantes, Montpellier models of allelic exclusion. These models support the notion that Cedex 5, France. V(D)J recombination initiates at one allele at a time due to Received for publication December 29, 2009. Accepted for publication May 7, 2010. a specific yet ill-defined molecular control(s) (8–10) or as the result This work was supported by the Agence Nationale de la Recherche program BioSys number 06-135161. Work in P.F.’s laboratory is also supported by Institut National de of a stochastic, low-probability onset (11, 12). Whatever the la Sante´ et de la Recherche Me´dicale, Centre National de la Recherche Scientifique, cause, a VDJ+ outcome, one in every three (e.g., Vb-to-DJb) the Association pour la Recherche sur le Cancer, the Institut National du Cancer, the joints on average (13), eventually leads to the prohibition of Fondation Princesse Grace de Monaco, and the Commission of the European Com- munities. E.F. was supported by a fellowship from Agence Nationale de la Recherche further rearrangement via a signal conveyed from the immature BioSys, number 06-135161. M.B. was supported by fellowships from the Marseille- receptor (the so-called pre-TCR) built from the newly synthesized Nice Genopole and Association pour la Recherche sur le Cancer. TCRb polypeptide (14–17). This concept agrees with the 60:40 Address correspondence and reprint requests to Dr. Pierre Ferrier, Centre d’Immuno- logie Marseille-Luminy, Marseille, Parc Scientifique de Luminy, Case 906, Marseille ratio mentioned above. However, it remains unclear as to how Cedex 09, 13288 Paris, France (P.F.), or Bastien Fernandez, Centre de Physique The´- recombination proceeds to the opposite allele in the relatively orique, CNRS Luminy, Case 907, 13288 Marseille Cedex 09, France (B.F). E-mail frequent cases of an out-of-frame VDJ2 initial assembly. More- addresses: [email protected] (P.F.) or [email protected] (B.F.). over, these models again fall short in terms of properly depicting The online version of this article contains supplemental material. the production of VDJ+/VDJ+ cells, unless we assume that a Abbreviations used in this paper: DN, double-negative; GL, germline; WT, wild-type. loosened control sporadically tolerates a synchronization of re- Copyright Ó 2010 by The American Association of Immunologists, Inc. 0022-1767/10/$16.00 combination at the two homologous TCRb alleles. These concerns www.jimmunol.org/cgi/doi/10.4049/jimmunol.0904182 The Journal of Immunology 1623 led us to consider an alternative scenario that is not tied into possible statuses as depicted in Fig. 1), at time t in a differentiating T cell a strictly sequential mode of interallelic activation for V(D)J re- population, the recombination time window of which started at t0. The combination. Markov transition matrix Q = Q(tDJ, tVDJ, tf) (Table I), composed of the probabilities 1/tDJ, 1/3tVDJ, 2/3tVDJ, and 1/tf for transitions through the In this study, we aimed to use a dynamical approach to model corresponding cell states (for a definition of the genomic statuses, re- TCRb gene recombination in an effort to comprehend the sto- combination time window, and transition rates, see Results, Formulation of chastic and regulated premises of allelic exclusion within a de- the model: overview and Formulation of the model: basic features) is upper velopmental scheme that would integrate all the observed cell triangular due to the feed-forward structure of the transition graph. The Markov chain was exploited using the relation subsets at once. Principally, successive D-to-J and V-to-DJ rear- t0 t0 rangements at TCRb alleles of individual T cells and ensuing x t ¼ x t0 exp t2t0 Q ;t t0; ð1Þ feedback inhibition are seen throughout as independent, possibly t t0 14 t in which x 0 ðtÞ represents the vector ðxi ðtÞÞ and, likewise, x 0 ðt0Þ¼ t 14 i¼1 concurrent and mostly not stringently simultaneous biochemical ðx 0 ðt ÞÞ i 0 i¼1 denotes the initial distribution at the window origin. transactions, with fluctuating modes of initiation and duration. In this study, we were first concerned with data from TCRb+ T cells Indeed, evidence that developing T cells exhibit uniformed, bial- (in the form of hybridomas) that corresponded to final states in the model + lelic transcription of a Vb gene before recombination (18) makes (denoted sVDJ /…; see below). Based on relation (1), we made explicit allele autonomy in the conduct of TCRb rearrangements a plausible assumptions as follows: the distribution remains constant from the end of the time window onwards [meaning in particular homogeneous hypothesis. Besides, the variety of rearranging sequences (19) and proliferation of cells harboring the distinct (sVDJ+/…) statuses] t0 t0 multiplicity of molecular factors/mechanisms involved in TCRb (i.e., x ðtÞ¼x ðt0 þ 1Þ if t $ t0 + 1) and in line with biological evidence gene recombination and feedback signaling (16, 20) strongly favor that thymus seeding by T cell progenitors (of status denoted GL/GL in the a widespread noise in, and randomness of, all these events. Fur- model) proceeds via gated intervals of receptiveness followed by longer refractory periods (26), we postulated that such cells enter the process on Downloaded from thermore, because none of the shaping events leading up to a sig- a cyclic basis (instead of a constant flow), with every single cell com- naling launch (from TCRb gene transcription to protein synthesis pleting its window within the seeding cycle (i.e., for all cells, t0 +1# end and formation of the pre-TCR) nor the resulting feedback block in of the refractory periods). Accordingly, the distribution resulting from any rearrangement are likely to take place instantly, the notion of feed- such cycle does not depend formally on t0. It reads back control implies a deferred outcome(s) with the possible pro- xð1Þ¼xð0ÞexpðQÞ; ð2Þ longation of recombination after a VDJ+ has been made. Hence, 14
xð Þ¼ðx ð ÞÞ http://www.jimmunol.org/ dynamical aspects are expected to impinge in a general sense on with the initial distribution 0 i 0 i¼1 containing GL/GL cells only (i.e., xGL/GL(0) = 1 and xi(0) = 0 for all i GL/GL). Steady-state TCRb gene recombination and, potentially, allelic exclusion. recombination takes place during consecutive seeding cycles. Thus, the + With this framework in mind and building on a recent study by fractions xi(1) for all four final states in the model i 2{sVDJ /GL, Sepulveda et al. (21), which illustrated stochastic modeling of sVDJ+/DJ, sVDJ+/VDJ2, sVDJ+/VDJ+} (see Results, Formulation of the model: basic features) actually proffer a formal expression for the distri- TCRg1/g4 gene rearrangement, we depicted TCRb gene re- + combination as a continuous-time Markov process contingent on bution of TCRb T lymphocytes. Given the triangular structure of the matrix Q, analytic formulas for the variables x (1), which, via Q, explicitly adjustable parameters. Using available data from wild-type (WT) i depend on the parameters (tDJ, tVDJ, tf), may be obtained by solving in and mutant mice, we obtained quantitative values for these param- a recursive procedure their corresponding linear differential equations. In eters by standard fitting procedures. In these settings, stochastic practice, for simplicity, we used in this study the matrix exponential tool of by guest on October 1, 2021 dynamics combined with feedback control thoroughly accounted Mathematica (Wolfram Research, Champaign, IL) to directly deduce these variables from expression (2). for the generation, in experimentally compatible ratios, of cell From relation (1), we also computed cell fractions from preselected DN cohorts comprised of a large proportion of allelically excluded data in studies by Aifantis et al. (24, 25) using another series of specific cells along with small subsets of VDJ+/VDJ+ cells and those har- statements. In this case, TCRb+ DN thymocytes prior to b-selection were + boring a VDJ+/GL genotype. The parameter values further prof- identifiedwithtransientstates(VDJ/…) in the model. Practically, we fered estimates for the mean length of the joining and inhibiting further presumed that cells ending their recombination time window and not in the (sVDJ+/. . .) form, including (VDJ+/. . .), have a limited intervals, thus providing a scale of relative timings tailored to lifetime before committing to apoptosis or diverging toward another TCRb gene recombination. Our modeling reveals concepts award- lineage. Thus, differing from above where the computations relied on ing a discrete chromosomal system with the property to display cell fractions integrated over a stationary regimen comprised of multiple robust allelic exclusion at minimal regulatory cost while keeping cell-seeding cycles, we then had to consider statistical numbers cumu- lated over a limited interval. As an integration period, we chose the the opportunity to maximize genetic diversity via dual allele usage. nonstationary section of the model corresponding to the timing window T = 1. Therefore, the sampled fractions (supposedly collected at an in- Materials and Methods stant t independent of the cell-differentiation course) had to be fitted by the following expression: Source of TCRb+ experimental distributions ðt + to Experimental distributions of mouse TCRb T cells used were obtained XiðtÞ¼ xi ðtÞdt0; ð3Þ from published studies and compiled in Tables II and III. Numbers of t21 b-selected T lymphocytes well-defined in terms of TCRb genotype (Table which integrates the cell populations over the time window that began after II) were derived from data sets available in studies by Khor and Sleckman t21. A simpler expression was achieved by further assuming that experimen- (22) and Senoo et al. (23), which used peripheral T cell-derived, cloned tal sampling is effected once the flood of cells initiating TCRb recombination hybridomas from WT (22, 23) and engineered mutant bLD/LD (23) mice, has reached a steady-state course (i.e., with the proportions of GL/GL precur- respectively. Distributed numbers of early-developing double-negative sors entering the recombination window being constant in time). (DN) thymocytes prior to b-selection (Table III) were derived from data t0 t21 sets in studies by Aifantis et al. (24, 25), which used FACS-sorted intra- Explicitly: x ðt0Þ¼x ðt21Þ for t21 t0 t; cytoplasmic TCRb+, cell-surface CD25+ small DN cells from WT and ð 2 2 2 2 1 ð4Þ a a / / t21 pT - or SPL-76-deficient animals (pT and SPL-76 , respectively). implying XiðtÞ¼ xi ðt21 þ t0Þdt0; 0 Simulation of TCRb+ T cell distributions and analytic xt21ðt2 þ t Þ¼xt21ðt2 Þ ðt QÞ expressions in which i 1 0 1 exp 0 with, as before, the initial distribution xt21(t21) = x(0) containing only GL/GL cells. Again, we To modelize the random dynamics of TCRb gene recombination in early used the matrix exponential and formal integration tools in Mathematica developing T lymphocytes, we generated a Markov chain based on the (Wolfram Research) to obtain an explicit expression of Xi(t)=Xi, which t0 transition graph shown in Fig. 1, in which xi ðtÞ2½0; 1 symbolizes the likewiseÐ depends on (tDJ, tVDJ, tf), by using the t-independent relations X ¼ xð Þ 1 ðt QÞdt X ¼ðX Þ14 fraction of individual cells harboring the same genomic status i (among 14 0 0exp 0 0 and i i¼1. 1624 DYNAMICAL MODELING OF TCRb ALLELIC EXCLUSION
Least-squares fits and parameter estimation addition, to ascertain that solutions were method-independent, we also verified that the same curves of parameter triples were obtained when To fit the analytic predictions of our statistical model to the experimental applying the maximum likelihood method, which determines the most data sets, we primary used the least-square method, a standard technique in likely parameters by maximizing the sum numerical analysis (27). The method is based on an algorithm designed to tune parameters so as to minimize the Euclidean distance between the Ni xið1Þ predicted distributions and the experimental distributions, relying on the + ðusing data sets for selected T lymphocytesÞ fact that the distance vanishes if (and only if) the two distributions co- i2S +x + incide. In this study, given a collection of TCRb cell numbers {Ni}i2S (in or which S denotes any subset of cells harboring one potential TCRb+ geno- mic status as depicted in Fig. 1 and in Tables II and III (e.g., sVDJ+/DJ; + 2 + + + 2 + + Ni sVDJ /VDJ ; sVDJ /GL; or VDJ /DJ; VDJ /VDJ ; VDJ /VDJ ), the Xi + ðusing data sets for preselected DN thymocytesÞ: quantities to minimize would be the distances i2S +X N x ð Þ 2 + i 2 i 1 Least-squares fitting procedure and parameter estimation: + + i2S N x worked example ðin fitting data sets for selected T lymphocytesÞ Using data from WT mice in the study by Senoo et al. (23), the distance to or minimize was 2 N þ x þ ð Þ N X 2 sV DJ =GL sV DJ =GL 1 + i 2 i 2 + + 76 Sx i2S N X 2 N þ x þ ð Þ
sV DJ =DJ sV DJ =DJ 1 Downloaded from ðin fitting data sets for preselected DN thymocytesÞ þ 2 76 Sx 2 NsV DJþ=V DJ2 xsV DJþ=V DJ2 ð1Þ Ni þ 2 ; between the experimental proportions + (in which +N ¼ + Ni is N i 2 S i2S 76 Sx a normalization factor) retrieved from Tables II or III, respectively, and the in which Sx ¼ xsV DJþ=GLð1Þ þ xsV DJþ=DJ ð1Þ þ xsV DJþ=V DJ2 ð1Þ. xið1Þ To perform minimization using standard fitting algorithms [including corresponding relative fractions + (in which, likewise, x i 2 S when applying the widely used Nealder-Mead algorithm (28) for parameter http://www.jimmunol.org/ Xi validations via the routine “optim” in R] always firstly requires assigning +x ¼ + xið1Þ), or + , computed as described in the previous i2S X i2 S arbitrary values to all the parameters and iteratively traversing the param- paragraph. eter space through a succession of small displacements along each param- In practice, numerical calculations consisted of applying built-in proce- eter axis. dures and functions (especially function FindMinimum) in Mathematica In practice, to explore the predicted continuum of parameters for the distribution ðNsV DJþ=GL; NsV DJþ=DJ ; NsV DJþ=V DJ2 Þ = (1, 39, 36), we (Wolfram Research) that, starting from arbitrarily chosen values of (tDJ, therefore initially ran the least-square algorithm for 500 parameter triples tVDJ, tf), searched for minimizing the above mentioned distances and, from there, directly yielded the matching, resultant parameter triples. Of equidistributed on a cubic grid. In applying the validating criterion that note, due to finite computation time, a perfectly vanishing distance could we adopted in this study [closest integers for the computed cell numbers + not be reached by numerical means. Therefore, to validate the parameter x ð Þ N i by guest on October 1, 2021 1 + coinciding with experimental figures (1, 39, 36)], the pro- values, we adopted the following criterion: we only retained parameter x i 2S triples for which the closest integers to the predicted cell numbers cedure returned .450 solutions (parameter triples) that aligned on
+N +N a smooth curve (i.e., the red dots depicted in Fig. 2; also see Supplemental xið1Þ or Xi all coincided with the experimental figures + + Table I). x i2 S X i2 S {Ni}i2S. Continuum of fitting parameters Results Formulation of the model: overview As described above, the least-squares approach was a comprehensive and practical way to numerically solve the equations We implemented a stochastic modeling of TCRb gene re- combination based on the Markov process formalism (29). Thus, xið1Þ Ni Xið1Þ ¼ ;i2S ðor TCRb dual-allele rearrangement status in a collection of individ- Sx SN SX Ni ual single cells is represented via a time-varying random variable ¼ ; i2SÞ for the parameters ðtDJ; tVDJ; tf Þ: SN distributed into a finite set of states. Holding to the Markov prop- erty, this variable evolves with time from one state to the next at Incorporating in this scheme as many equations as the number s of analyzed states in S, there are indeed only s21 independent conditions to a rate specific to the considered transition, independently of all comply with, because one equation is always ascertained by the normal- past statuses. Permitted transitions and associated rates are x ð Þ i 1 ¼ Ni ¼ encoded into the so-called transition matrix (30), the structure of ization + + + + 1. i2S x i2S N which is determined by: 1) biological constraints on TCRb gene For all data sets in Tables II and III, s = 3 (i.e., states sVDJ+/DJ; + 2 + + + 2 + + recombination (D-J occurring first, prior to V-DJ; no direct V-J); sVDJ /VDJ ; sVDJ /GL; and VDJ /DJ; VDJ /VDJ ; VDJ /VDJ , re- + spectively). Accordingly, in each case, there were only two independent and 2) feedback inhibition following a VDJ assembly (see Ref. 1, equations to resolve the three parameters tDJ, tVDJ, and tf. Therefore, Fig. 1, Table I). Transition rates represent adjustable parameters precisely defining those parameters was possible aside from one degree of expressed in terms of the mean duration of the given rearrange- freedom, and, consequently, the fitting procedure, when feasible, yielded ment step and average time lapse to achieve inhibition (covering a continuum of triple solutions in the parameter space. the whole period from TCRb gene expression and protein synthe- sis to trans-allelic block in recombination), respectively. Collec- Additional validation of parameter triples tively, the distribution of cell subpopulations harboring distinct To further access the effectiveness of parameter evaluations, we sought to TCRb gene configurations (b-genomic statuses) evolves in a de- confirm the above calculations using additional tools and methods. We thus terministic way along the linear flow generated by the transition applied the routine optim in R, another minimizing procedure that relies on the well-known Nelder-Mead (downhill simplex) algorithm (28). To be matrix. Provided an initial distribution is known for t = 0, the agreed, the proposed estimations had to yield indistinguishable or, if not, fraction of cells included in each subpopulation at any one time similar (according to the above-defined criterion) triple solutions. In t . 0 can be calculated. The Journal of Immunology 1625
VDJ+/DJ are seen as one identical status). In this way, the model only features general aspects regarding the structural organization and use of gene segments at the TCRb locus, without dwelling on which particular cluster(s) of gene segments is being used for recombination. Upgrading prospects (e.g., by taking into account the regular usage of separate Db–Jb clusters) are considered in the Discussion. TCRb gene recombination in jawed vertebrates takes place in minor subpopulations of thymic cells at the very initial stages of T cell development between thymus seeding by lymphoid progeni- tors and pre-TCR–driven selection of thymocytes committed into the abT cell lineage (also known as b-selection). In the mouse, FIGURE 1. Transition graph of the Markov process modeling TCRb these cell subsets, delineated via cell-surface expression of discrete gene recombination and feedback inhibition. The graph also depicts or- molecular markers, are commonly named DN1–3 cells, with DN2 dered rearrangement events at the TCRb locus (Db-to-Jb joining and DN3 displaying predominant DJb- and VDJb-rearranged occurring first, prior to Vb gene recombination); arrows represent products, respectively. Allelic exclusion/feedback inhibition is authorized transitions in TCRb gene rearrangement/cell genomic status. tightly coupled to b-selection prior to the DN3–DN4 stage transi- Transition rates (in red) are computed from those at a single allele (e.g., the tion, with cells not passing this checkpoint doomed to apoptosis or, → first GL DJ transition occurs at rate 2/tDJ as a Db-to-Jb rearrangement possibly, diverted toward a distinct gdT cell lineage (31). Hetero- may occur at each individual allele). Notice that, at the TCRb locus, there Downloaded from geneity in DN1–3 cell distribution, despite evidence of cycled is only one potential open reading frame through the various Jb gene segments and Cb exons and no stop codon within the Db gene segments seeding by T cell precursors (26), suggests an uncoordinated mode (Ref 33; S. Jaeger and P. Ferrier, unpublished observations). Accordingly, of activation and course of DN cell differentiation and/or recom- in a situation in which there is no evidence that DJb joints encode bination programs. To account for all these features in our model- functionally relevant, truncated TCR b-chains, it is thus the final recom- ing (notice that developmental stages are not explicitly represented bination outcome (the Vb-to-DJb joining event) that is assumed to impinge as separate variables in this study), we assumed that, for each and on in-frame/out-of-frame readability [i.e., at this later transition, the pos- every single cell of initial GL/GL status, all requirements for b http://www.jimmunol.org/ sible occurrence of a rearrangement-generated premature stop codon(s) sequential rearrangements (a nonexhaustive list including notably along the unique VbDJbCb open reading frame actually becomes relevant availability of discrete transcription/coactivation factors, recombi- in terms of productivity]. nase expression and activity, and changes in chromatin structure and in chromosomal organization/positioning; see Ref. 32) could be met during a time window of length T (without loss of general- Formulation of the model: basic features ity, we take T = 1). The window begins at variable time points for The model features a standardized compartmentalization of TCRb each individual cell depending on both intrinsic (e.g., gene/ gene recombination such that, at the single-cell level, each TCRb protein expression landscapes, all items in the list mentioned by guest on October 1, 2021 allele displays one of the following configurations: 1) GL; 2) above) and extrinsic (e.g., thymic environment) fluctuation fea- partially D-J rearranged (DJ); or 3) completely V-DJ rearranged, tures. Within the window, successive D–J and V–DJ rearrange- either productively (VDJ+) or not (VDJ2). The genomic status at ments proceed concurrently on opposite alleles such that the b this locus is defined by the configuration of the two TCRb alleles, status evolves according to the Markov process based on the tran- regardless of how the allelic pairs are ordered (e.g., DJ/VDJ+ and sition graph featured in Fig. 1. At individual alleles, the GL→DJ
Table I. Transition matrix Q =(qrs)r,s = {1,…,14} associated with the Markov model 0 1 3 2 000000000000 B tDJ C B 0 3 2 1 1 00 000 0 000C B 3tVDJ tDJ 3tVDJ C B 00 3 0001 0000000C B tDJ C B 00 0 3 004 2 000000C B 3tVDJ 3tVDJ C B 00003 1 0 1 000000C B tf tDJ C B 00000000000000C B C B 0000003 002 1 000C B 3tVDJ 3tVDJ C B 0000000 3 1 0 2 1 00C B tf 3tVDJ 3tVDJ C B C B 00000000000000C B C B 00000000000000C B 0000000 000 3 0 1 0 C B tf C B 3 1 C B 0000000 000 0 0 tf C @ 00000000000000A 00000000000000
The order of rows and columns follows the numbering in transition graph shown in Fig. 1. In each row, the symbol 3 14 denotes the opposite of the sum of row entries, namely the quantity 2+ qrs. s¼1 1626 DYNAMICAL MODELING OF TCRb ALLELIC EXCLUSION
2 and subsequent DJ→VDJ+ or DJ→VDJ transitions occur as in- Table II. Experimental numbers of cells harboring the sVDJ+/GL, + + 2 dependent Poisson processes, with rates equating to, respectively, sVDJ /DJ, or sVDJ /VDJ genomic status 1/tDJ, 1/3tVDJ, and 2/3 tVDJ, in which tDJ and tVDJ represent the T Cell Type (Data Source) N VDJþ =GL NsV DJþ =DJ NsV DJþ =V DJ2 average lengths of time of the corresponding transitions, and the V- s to-DJ transition alone bears susceptibility to the 1/3 rule of in- WT (22) 0–2a 118–120a 92b frame recombination, as it is thought to be the case at the TCRb WT (23) 0–2a 38–40a 36 LD/LD locus (1, 33) (Fig. 1; the figure shows the actual rates for TCRb Mutant b (23) 14 32 36 dual allele usage). Should a VDJ+ be completed and the resulting The distinct genomic statuses are indicated and were obtained from data in Khor and Sleckman (22) and Senoo et al. (23). pre-TCR produced and made operational, the given cell can then a + Numerical variability results from formatting the experimental data so as to fit switch to one final state in the model (symbolized sVDJ /GL, the genomic statuses in the model [especially the status of TCRb alleles carrying an 2 sVDJ+/DJ, sVDJ+/VDJ , or sVDJ+/VDJ+). The later transitions unrearranged Db1 gene segment is uncertain (i.e., could be GL or DJ depending on the configuration of the cis-linked Db2–Jb2 cluster)]. all arise at rate 1/tf, in which, similar to above, the parameter tf bIn this study, the productive versus nonproductive outcome in cells harboring + denotes the mean time interval between completion of a VDJ and two complete VDJ rearrangements was specified for only 44 out of 92 hybridomas, 2 effectiveness of pre-TCR–mediated allelic exclusion. As time runs and all 44 displayed a VDJ+/VDJ genotype. Accordingly, in the corresponding N þ ¼ N þ 2 þ fitting procedure, we assumed that 92 = sV DJ =V DJ s VDJ =V DJ out, the end of the recombination window eventually puts a stop to N þ þ s VDJ =V DJ . We note that, in all cases of fitted parameters, the model did predict any further change in the status of lingering cells [strictly speaking, at least 44 out of 92 sVDJ+/VDJ cells that harbor 1 VDJ+ only. the latter statement only pertains to the b genomic status; a more elaborate model would be needed to also integrate the outcome— death, commitment to a distinct lineage, etc.—of those cells that specifying the model so as to account for the given data set). Even did not reach a (sVDJ+/. . .) closing stage during their recombination though parameter solutions are depicted for a specific T cell dis- Downloaded from window; see Ref. 34]. Given all of the above statements, we relied tribution in the particular study (Fig. 2, see legend), fits using on standard results from the Markov process theory and on different distributions within the limits of those indicated in Table biologically relevant formulations to explicitly compute parameter- II led to similar solutions with identical properties (data not dependent estimations of TCRb+ cell distributions; notably, expres- shown). Notably, the solution curves obtained for WT mice in sions (2) and (4) proffered analytic predictions on statistical values of the two studies investigated in this paper displayed sound consis- developing T cells harboring the distinct b genomic statuses (for tency (Fig. 2, blue and red dots; slight deviation between the two http://www.jimmunol.org/ details, see Materials and Methods). curves likely reflects statistical variations inherent to relatively small-sized biological samples). Overall, these results thus imply Fitting statistics from b-selected T lymphocytes that, provided a proper calibration of the parameters has been To calibrate parameters in the model and, ultimately, assess the performed, a Markov process simulating the basic steps of TCRb statistical relevance of modeling predictions, we relied on available locus assembly in developing T cells reproduces in silico the data from experimental analyses of TCRb+ cell distributions and the application of standard fitting procedures. Briefly, we applied least-squares minimizing methods to fractions ascertained from by guest on October 1, 2021 published numbers of T cell-derived, cloned hybridomas well characterized in terms of TCRb genotype (for details on the bi- ological approach, see, for example, Ref. 22). The mathematical procedures were implemented using Mathematica (Wolfram Re- search), function FindMinimum. The parameter triples (tDJ, tVDJ, tf) obtained in this way were further validated using the maximum likelihood method (see Materials and Methods, Least-squares fits and parameter estimation and Additional validation of parameter triples; all methods also described in Ref. 27). In practice, studies in the literature yielded quantitative in- formation on cell populations displaying only three out of the four potential TCRb+ final states outlined above, namely sVDJ+/GL, sVDJ+/DJ, and sVDJ+/VDJ2 (22, 23) (Table II). Hence, resolving the three parameters from these fragmentary statistics was possi- ble aside from one degree of freedom (see Materials and Methods, Continuum of fitting parameters). Consequently, for each data set in Table II, the fitting process, if doable, was predicted to yield a continuum of triple solutions situated on a curve in the param- eter space. To draft the curves, we ran the minimizing procedures for 500 initial parameter triples, equidistributed along each pa- rameter axis. The validated solutions (full range of numerical values) are displayed in Supplemental Table I and the definitive FIGURE 2. T lymphocyte data fitting. Plots of parameter triples curves shown in Fig. 2 (for a practical application of the global ap- obtained in fitting the experimental data sets summarized in Table II proach, also see Materials and Methods, Least-squares fitting pro- [three-dimensional drawing and two-dimensional projections (A–C) are shown]. Blue dots: fitting the distribution (NsV DJþ=GL;N VDJþ=DJ ; cedure and parameter estimation: worked example). Thus, fitting s N þ 2 þ N þ þ sV DJ =V DJ sVDJ =V DJ ) = (2, 118, 92) chosen from the experimen- the model to quantitative data from hybridomas generated in two tal data provided for WT T cells in Khor and Sleckman (22); red dots: studies using WT mice (22, 23) and, in one case, also mice car- fitting the distribution ðNsV DJþ=GL;NsV DJþ=DJ ;NsV DJþ=V DJ2Þ = (1, 39, rying a modified TCRb locus impinging on some aspects of gene 36) chosen from data provided for WT T cells in Senoo et al. (23); green recombination (23, 35) yielded parameter triples that aligned onto dots: fitting the distribution ðNsV DJþ=GL;NsV DJþ=DJ ;NsV DJþ=V DJ2Þ = (14, a smooth curve in all cases (Fig. 2; each individual triple 32, 36) provided for the bLD/LD mutant T cells in Senoo et al. (23). The Journal of Immunology 1627 statistical data on TCRb genotype/cell-subset distribution delin- the control of the initiation phase of allelic exclusion (allelic asyn- eated experimentally in independent analyzes of mature T lym- chronism) at the TCRb locus impinges on the Vb-to-DJb phocytes (as discussed in detail below). rearrangement step (1, 2, 20, 42). In addition, the longer GL→DJ and, conversely, slightly shorter DJ→VDJ transitions observed in Parameter estimates and model predictions parallel between bLD/LD mutant and WT T cells represent Examination of the solution curves in Fig. 2 afforded important provocative results as the bLD mutation: 1) specifically enforces information concerning parameter calibration and related pre- the sole usage of the D2–J2 cluster, the segment assembly of dictions relevant to the specific experiments. First and markedly, which is suspected to proceed less efficiently than that involving + + in those situations in which the number of VDJ /VDJ cells was the D1–J1 cluster; and 2) greatly reduces the numbers of 59 Vb ignored, the explicit solutions could in all cases be obtained for gene segments and the intervening distance from the D2–J2 cluster, → ‘ arbitrary values of tf (in fact, unlimited from 0 + ) with, con- possibly increasing their recombination rate (23, 39, 43). + versely, tDJ and tVDJ showing variations of small amplitude. In N xið Þ Finally, in using the numerical expressions 1 + (see Mate- other words, the parameter tf entirely epitomized the anticipated x degree of freedom. Clearly, this parameter (and corresponding rials and Methods, Simulation of TCRb+ T cell distributions and time delays) cannot be inferred from such data sets neglecting the analytic expressions; expressions (1) and (2) and parameter sol- VDJ+/VDJ+ cells. In this context, however, we noticed that the utions from Fig. 2 and Supplemental Table I, we made predictions + parameter tVDJ in general, as well as the parameter tDJ for the concerning the distribution of all TCRb cell populations and, more + + mutant mice (Fig. 2, green dots), displayed either substantial specifically, of the fraction of sVDJ /VDJ T cells not supplied + + 2 variations or saturation for, respectively, small (,0.1) and larger by the given data sets. Strikingly, all sVDJ /DJ, sVDJ /VDJ ,
+ + + Downloaded from (.0.2) tf values (Fig. 2, most evident in A and C, respectively), sVDJ /GL, and sVDJ /VDJ entities were correctly predicted by pointing to more robust estimates of recombination periods using the model, with the first two approximating the 60:40 ratio (pre- , , this later interval. An interval of 0.1 , tf , 0.2 would imply an cisely, for 0.1 tf 0.2, 56 and 39%, respectively; in agreement approximate 0.3–0.5 d (7–12 h) period to achieve pre-TCR syn- with exact measuring; e.g., see Ref. 2) and the latter ones restricted thesis and feedback inhibition assuming that the recombination to a minority (∼5%), as expected (Fig. 3A; similar distributions window is comprised within the 2.5–3 d needed by T cell precur- were obtained using parameters from WT data in Ref. 23; E. Farcot sors to transit through the DN2–DN3 compartments (36, 37). In- and B. Fernandez, unpublished observations). Specifically, the http://www.jimmunol.org/ + + deed, within the 0.1 , tf , 0.2 range, the curves exhibited limited fraction of sVDJ /VDJ cells turned out to be fairly low in all three variations enabling a precise determination of tDJ values [0.036 6 cases, varying monotonically with tf (Fig. 3B). Depending on the + 0.006 and 0.038 6 0.007 for the WT T cells in Khor and Sleck- data set, solutions ranged from 3.6–6.4% of total TCRb cells. man (22) and Senoo et al. (23), respectively; 0.205 6 0.020 for the These estimates are in accordance with assessments on TCRb mutant T cells in Senoo et al. (23)] and fairly consistent estimates allelically included cells based on genomic sequencing of a number of tVDJ values (WT T cells: 0.55 6 0.06 and 0.4 6 0.05; mutant of VbDJb joints (3–10%; even though, at the phenotypic level, dual T cells: 0.33 6 0.03); as already mentioned, slight variations TCRb+ cells represent a smaller percentage due to postrearrange- between numerical figures from WT data sets may reflect statis- ment events, such as competition between Vb gene promoters, tical fluctuations linked to limited-size sampling. Accordingly, in TCR b- and a-chain pairing, etc.) (24, 25, 44, 45). In addition, in by guest on October 1, 2021 LD/LD + WT T cells, tDJ would display smaller values (by one order of the case of the b mice, the predicted proportions of sVDJ / + magnitude) compared with those of tVDJ, implying, in this sce- VDJ cells are similar to those in WT animals and thus corroborate nario, a prompt completion of D–J assembly and a time window experimental observations arguing that the specific mutation does dominated by the V–DJ interval. These results are reminiscent of not impact on TCRb allelic exclusion (23). Overall, the ability of a handful of experimental data implying an increased intricacy of the model to integrate predominant and rare subpopulations and to Vb-to-DJb rearrangement as compared with that of Db-to-Jb, satisfactorily predict their statistical behaviors is testimony to the requiring at least: 1) the partitioning of distant genomic regions power of the approach. via developmentally regulated chromosomal looping (38); and, perhaps, 2) the breaking of Vb-privileged connection to nuclear Fitting statistics from preselected DN thymocytes repressive compartments (12, 38); 3) overriding the relatively in- Besides integrating statistical data on TCRb allelic status in effective Vb recombination signal sequences; and/or 4) supplant- differentiated T cells, potentially accessing such quantitative ing of inter-Vb antagonisms for productive coupling with a DJb figures in early developing DN thymocytes prior to b-selection unit (39–41). In this regard, the tVDJ .. tDJ picture accom- would be an additional advantage. So far, very few experimental modates the assumption generally endorsed in the literature that studies have addressed this issue. Notably, Aifantis et al. (24, 25)
FIGURE 3. Predictions on cell distributions. Pro- ductively rearranged cells related to the parameter tf were computed via the model in using the solution curves shown in Fig. 2. A, Estimations of the four cell populations of the form (sVDJ+/. . .) using parameter triples derived from data in Khor and Sleckman (22). B, Estimations of allelically included cells using pa- rameter triples derived from data sets summarized in Table II. The color code is identical to the one used in Fig. 2. 1628 DYNAMICAL MODELING OF TCRb ALLELIC EXCLUSION