Feedforward regulation of Myc coordinates lineage-specific with housekeeping gene expression during B cell progenitor cell differentiation

The differentiation of self-renewing progenitor cells requires not only the regulation of lineage- and developmental stage–specific genes but also the coordinated adaptation of housekeeping functions from a metabolically active, proliferative state toward quiescence. How metabolic and cell-cycle states are coordinated with the regulation of cell type–specific genes is an important question, because dissociation between differentiation, cell cycle, and metabolic states is a hallmark of cancer. Here, we use a model system to systematically identify key transcriptional regulators of Ikaros-dependent B cell–progenitor differentiation. We find that the coordinated regulation of housekeeping functions and tissue-specific gene expression requires a feedforward circuit whereby Ikaros down-regulates the expression of Myc. Our findings show how coordination between differentiation and housekeeping states can be achieved by interconnected regulators. Similar principles likely coordinate differentiation and housekeeping functions during progenitor cell differentiation in other cell lineages.


Introduction
Cell proliferation, metabolic state, and differentiation are linked: proliferating progenitor cells exit the cell cycle and adjust their metabolism as they differentiate [1][2][3]. Mechanistically, this coordination is thought to involve mutual antagonism between cyclin-dependent kinases that promote cell-cycle entry and transcription factors that induce tissue-specific gene expression [1,2].
A detailed inventory of differentiation stages is available for mammalian haematopoiesis. In B cell differentiation, discrete stages are defined by CD markers [4], gene expression profiles [5] (www.immgen.org), transcription factor binding [6][7][8], and cell-cycle states [9][10][11]. A critical step is the transition of proliferating B cell progenitors towards cell-cycle arrest and differentiation. We refer to proliferating B cell progenitors as Fr.C following Hardy's nomenclature [4]; this stage is also known as the pro-B, pre-B1, or large pre-B cell stage (Fig 1A). We refer to quiescent, differentiating B cell progenitors as Fr.D following Hardy's nomenclature [4]; this stage is also known as the pre-B, pre-B2, or small pre-B cell stage (Fig 1A). Transcriptional regulators of the Ikaros family of zinc finger proteins are up-regulated at this transition [12] and are required for B cell progenitor differentiation in vivo [13][14][15]. IKZF1, the gene encoding Ikaros, is recurrently mutated in human B cell progenitor acute lymphoblastic leukemias (B-ALLs) with translocations between the IGH locus and the ABL1 proto-oncogene (BCR-ABL1) [16,17].
Here, we employ an inducible system in a B cell progenitor line that models the transition from B cell progenitor proliferation toward cell-cycle arrest and differentiation upon the regulated delivery of the transcription factor Ikaros from the cytoplasm into the nucleus [7,18]. The availability of this in vitro model, combined with access to primary B cell progenitors for validation experiments, makes B cell progenitor differentiation an attractive system to study progenitor differentiation and the mechanisms that link differentiation with changes in cell cycle and metabolism. To understand the regulatory control of B cell progenitor differentiation, we developed an algorithm that examines the temporal correlation between the expression of transcription factors and their target genes over the course of progenitor differentiation and scores the relative contribution of different transcription factors This approach, which is transferable to other cell-state transitions, highlighted that the Ikaros targets Foxo1 and Myc as high-scoring transcription factors. Perturbation experiments showed that the transcriptional repression of Myc was critical for the coordinated regulation of lineage-specific, cell-cycle, and metabolic genes. Ikaros-mediated repression of Myc connects B cell differentiation to cellcycle exit and metabolic adaptation. Similar principles may coordinate differentiation and housekeeping functions during progenitor differentiation in other cell lineages.

Temporal dissection of gene expression changes during B cell progenitor differentiation
We controlled the dosage of nuclear Ikaros-ERt2 with temporal precision by the addition of 4-hydroxy-tamoxifen (4-OHT) in the pre-B cell line B3 [7,18]. This model is designed to approximate the B cell linker (BLNK)-dependent increase of Ikzf3 expression in primary B cell progenitors in response to pre-B cell receptor signaling [12], and the nuclear translocation of Ikaros recapitulates the majority of gene expression changes that distinguishes proliferating (Fr.C) from differentiating (Fr.D) B cell progenitors in vivo [7,18]. We performed timeresolved RNA sequencing (RNA-seq) of 4-OHT-treated B3 cells expressing Ikaros-ERt2 or control vector (Fig 1B) at 6 time points after 4-OHT. We combined pairwise comparison between time points by limma [19] with time-course analysis by maSigPro [20] to identify 5,865 differentially expressed genes (S1 Table). Pseudotemporal ordering [21] of single cell RNA-seq (scRNA-seq) data showed an unbranched path over the experimental differentiation time course (Fig 1C).

Dynamics and validation of the experimental system
Gaussian mixture modeling for model-based clustering [22] resolved up-and down-regulated genes, subtrends of immediate and delayed regulation, and 2 nonmonotonic up-down-and down-up-regulated groups ( Fig 1D). Functional characterization of these trends by gene set analysis showed that up-regulated genes were enriched in Janus kinase/signal transducers and activators of transcription (JAK-STAT) signaling, cell adhesion, and B cell receptor and interleukin signalling, while metabolic functions, RNA metabolism, and ribosome biogenesis were enriched among down-regulated genes. This analysis indicates a clear distinction between induced (signaling and cell-cell communication) and repressed (mainly metabolism-related) processes during B cell progenitor differentiation (Fig 1D).
To ask how well this experimental system models B cell progenitor differentiation in vivo, we compared dynamic gene expression at each time point with static, developmental stagespecific gene expression by primary B cell progenitors [5]. Up to 6 h after 4-OHT-induced nuclear translocation of Ikaros, gene expression correlated positively with the less mature Fr.C and negatively with the more mature Fr.D states in vivo (Fig 1E). The 12 h time point marked a tipping point at which the positive correlation with Fr.C and the negative correlation with Fr.D was lost (Fig 1E). After that, gene expression correlated positively with the more mature Fr.D and negatively with the less mature Fr.C (Fig 1E). Therefore, gene expression in B3 cells showed a transition from an Fr.C-like state to an Fr.D-like state within a 24-h time frame. The experimental system allowed the capture not only of the start and end points but also the dynamics that occurred between them.
To examine the ability of this model system to pinpoint Ikaros target genes relevant to human disease, we assembled gene expression profiles of 1,404 B-ALL samples with and without IKZF1 mutations (S1 Fig and associated text). Genes that were differentially expressed between Fr.C and Fr.D were preferentially deregulated in IKZF1-mutated B-ALL [23,24], including current and potential therapeutic targets [25][26][27] and prognostically relevant gene signatures in B-ALL [17,28] (S1 Fig). There was significant overlap between differential gene expression in IKZF1-mutated B-ALL and early gene expression changes in B3 cells at 0 to 2 h after Ikaros induction ( Fig 1F, odds ratio = 2.53, adjusted [adj.] P = 0.02 for the 200 top differentially expressed genes). Because early gene expression changes are likely direct, this finding indicates that our in vitro system identifies gene expression changes in IKZF1-mutated B-ALL that may result directly from the loss of Ikaros function.

The progression from proliferating to resting stages of pre-B cell differentiation is marked by "reverse" metabolic reprogramming
To follow up on the down-regulation of genes related to metabolism (Fig 1D), we compared the expression of glycolysis and tricarboxylic acid (TCA) cycle genes during B3 cell differentiation in vitro (Fig 2A and 2B, left) and B cell progenitor differentiation in vivo (Fig 2A and 2B, right). Key glycolysis and TCA cycle genes were down-regulated in vitro and in vivo (Fig 2A  and 2B; the transition from Fr.C to Fr.D is indicated with a bracket). Analysis of Ikaros chromatin immunoprecipitation sequencing (ChIP-seq) data [7] showed that numerous core metabolic genes were directly bound by Ikaros, as illustrated by the glucose transporter Slc2a1 and the TCA cycle gene Fh1 (Fig 2C; see figure legend for a list of Ikaros-bound core metabolic genes).
In extracellular flux assays, nuclear translocation of Ikaros triggered a pronounced reduction (55%-65%) in extracellular acidification (ECAR) as a measure of lactate production ( Fig  2D, left). The oxygen consumption rate (OCR) was reduced by 45%-50% (Fig 2D, right). This was accompanied by reduced mechanistic target of rapamycin complex 1 (mTORC1) activity, as read out by phosphorylation of S6 ribosomal protein (S3A Fig)  To validate these findings, we transduced primary Fr.C-like B cell progenitors with Ikaros or Ikzf3 (Aiolos) and determined the resulting changes in ECAR and the OCR. Ikaros and Aiolos reduced the ECAR and the OCR of primary B cell progenitors (Fig 2E).
In the experiments described above, changes in metabolic gene expression and metabolic activity were induced by the expression of Ikzf1 or Ikzf3. During B cell progenitor differentiation, Ikzf3 expression is initiated by signaling through the pre-B cell receptor via a pathway that requires BLNK [12]. To determine whether pre-B cell receptor signaling is linked to analysis of gene expression trends in B3 differentiation. Each box contains the total number of genes and the major functional terms (q < 0.01) for each of 4 expression trends (up-regulated, down-regulated, first up-then down-regulated, and first down-then up-regulated). Up-and down-regulated trends are further separated into 2 subtrends, constant and delayed. (E) Comparison of the transcriptomes of the B3 in vitro differentiation time course with in vivo B cell progenitor differentiation stages. The 0 to 6 h time points resemble Fr.C, whereas 18 and 24 h are similar to Fr.D, and the 12 h time point represents a tipping point. The source of the numerical data underlying this figure is listed in S1 Data. (F) Comparison of genes differentially expressed in IKZF1 mutated human B-ALL and the B3 differentiation. Numbers denote common genes that are differentially expressed in IKZF1-mutated versus IKZF1 wild-type human B-ALL and the top 200 genes regulated at the indicated times after Ikaros induction in B3 cells. The color scale indicates the odds ratio for the enrichment of Ikarosregulated gene during B3 cell differentiation among differentially expressed genes in human B-ALL with and without IKZF1 mutations. B-ALL, B cell progenitor acute lymphoblastic leukemia; ERt2; Fr.C, proliferating B cell progenitor; Fr.D, differentiating B cell progenitor; HA, haemagglutinin; JAK-STAT, Janus kinase/ signal transducers and activators of transcription; RNA-seq, RNA sequencing; scRNA-seq, single cell RNA sequencing; TCA, tricarboxylic acid cycle; 4-OHT, 4-hydroxy-tamoxifen.  . The source of the numerical data underlying this figure is listed in S1 Data. (C) Ikaros ChIP-seq shows that Ikaros targets metabolic genes. The metabolic regulation, we used an experimental system in which BLNK activity can be inducibly restored in BLNK-deficient B cells by means of a 4-OHT-inducible BLNK-ERt2 fusion protein [29]. As expected, restoration of BLNK activity induced Ikzf3 and repressed Igll1, a known target of Ikzf1 and Ikzf3 during the the Fr.C to Fr.D transition [12,18] (Fig 2F, top). Restoration of BLNK activity led to the repression of the metabolic genes Slc7a5, Hk2, and Ldha (Fig 2F,  bottom), indicating that pre-B cell receptor signaling controls the expression of metabolic genes in B cell progenitors.
These data demonstrate altered metabolic gene expression, reduced glycolytic flux, and a drop in oxygen consumption at the transition of B cell progenitors from proliferation to quiescence in vitro and in vivo. This indicates "reverse" metabolic reprogramming towards a less glycolytic state [30,31]. The expression of B cell genes such as Igll1 and Foxo1 was correlated with the expression of housekeeping genes, such as Myc, the metabolic gene Hk2, and the cellcycle gene Ccnd2, not only at the level of cell populations but also in individual cells (Fig 2G). We conclude that the regulation of B cell-specific genes is coordinated with the regulation of housekeeping pathways during B cell progenitor differentiation.

Systematic identification of key transcription factors in B cell progenitor differentiation
B cell progenitor differentiation is marked by the differential expression of numerous transcription factors. To define the key transcription factors and the regulatory pathways involved in this process, we first identified all transcription factor genes that showed a significant and robust (log 2 fold change > 1.5) change in expression between consecutive time points. We considered transcription factors that were differentially expressed (adj. P < 0.01) between Fr.C and Fr.D in vivo ( [5]; www.immgen.org). We included transcription factors that showed updown-or down-up-regulation over the in vitro time course, based on the consideration that genes with nonmonotonic expression may be important for B cell progenitor differentiation even if they were not differentially expressed between the start and the end of the transition. This resulted in a total of 23 candidate transcriptional regulators ( Fig 3A, Table 1). To evaluate the potential importance (or "weight") of transcriptional regulators during cell-state transitions, we developed an approach that numerically integrates the differential expression over time of transcription factors and their target genes, which we refer to as transition weight matrix (TWM). For each transcription factor, we identified potential target genes based on transcription factor ChIP-seq peaks in gene promoters. We then determined the enrichment of transcription factor binding over differentially expressed genes and multiplied this enrichment with the log 2 fold change in transcription factor mRNA expression for each time interval.
The resulting values were summed over the time series to yield a score for transcription factor expression and target enrichment over time. We next examined to what extent the expression of each transcription factor correlated with mRNA levels of its target genes in a consistent fashion over time, which we term "coherence." Coherence between the expression of each embryonic stem cell gene Esrrb is a negative control. Other Ikaros-bound core metabolic genes are Dlat, Hk2, Idh3g, Slc2a3, Dlst, Sdhb, Pgam1, Mdh2, Pfkl, Dld, Pfkp, Sdha, Pcx, Idh1, Fh1, Aco1, Slc2a1, Slc2a5, Aldoa, Pdha1, Pkm, Pfkm, Gpi1, Pck2, Eno3, Suclg, and Pgk1. (D) Ikaros reduces ECAR, a measure of glycolysis (left) and the OCR (right) in B3 cells after 24 h (mean ± SEM of 3 biological replicates, �� P < 0.005, two-tailed t test). The numerical data underlying this figure are included in S1 Data. (E) Ikaros and the Ikaros family member Aiolos reduce ECAR and OCR in primary B cell progenitors (mean ± SEM of 6 to 7 biological replicates, � P < 0.05, �� P < 0.005, ��� P < 0.0001, one-tailed t test). The numerical data underlying this figure are included in S1 Data. (F) Impact of BLNK-mediated induction of Ikzf3 on the expression of metabolic genes. The numerical data underlying this figure are included in S1 Data.  transcription factor and its target genes was determined for each time interval and summed over the time series to yield a score for coherence. Finally, the score for transcription factor expression and target enrichment over time was multiplied with the score for coherence to yield a TWM score for each transcription factor. Details of this approach as well as a comprehensive mathematical description are provided in S3 Fig and in the "Methods" section.
We validated the TWM approach using the paradigm of T helper 17 (Th17) T-cell differentiation, in which the key transcriptional drivers are known [32,33] and high-quality transcription factor ChIP-seq data are available (S2 Table). TWM correctly identified RORC (TWM score = 0.17), the signature transcription factor of Th17 cells, and IRF4 (TWM score = 0.14), ChIP-seq identified RUNX1, FOXO1, and Myc as high-scoring regulators. (D) Progressive up-regulation of Foxo1 mRNA and down-regulation of Myc mRNA with time after Ikaros induction. Shown is the mean of 3 independent RNA-seq replicates. (E) Up-regulation of Foxo1 mRNA during in vivo B cell progenitor differentiation ( [5]; www.immgen.org). Down-regulation of Myc mRNA during in vivo B cell progenitor differentiation ( [5]; www.immgen.org). (F) Ikaros ChIPseq [7] shows Ikaros binding to the promoter regions of Foxo1 and Myc. The established Ikaros target gene Igll1 is shown for reference. ChIP-seq, chromatin immunoprecipitation sequencing; FOXO1, Forkhead Box O1; Fr.C, proliferating B cell progenitor; Fr.D, differentiating B cell progenitor; Myc, MYC proto-oncogene; RNA-seq, RNA sequencing; RUNX1, RUNT-related transcription factor; Th17, T helper 17; TWM, transition weight matrix.  [5]. "In vitro" denotes differentially expressed during the time course of B3 cell differentiation. "Direction" denotes up-or down-regulation of the transcription factor. � With the exception of Hoxb3, Sp7, and Sp100, the promoters of all 23 transcription factors are directly bound by Ikaros ChIP-seq peaks [7]. which is required for Rorc expression, as transcription factors with the highest TWM score ( Fig 3B). c-Maf (Maf; TWM score = 0.14), a transcription factor with an established role in Rorc induction and Th17 differentiation [34], and Hif1-alpha (Hif1a; TWM score = 0.12), which controls the balance between Th17 and regulatory T (Treg) cell differentiation [35,36], also scored highly. Therefore, TWM successfully identified key regulators of Th17 cell differentiation.

Abbreviations
To apply TWM to B cell progenitor differentiation, we retrieved published ChIP-seq data sets for transcription factors in B cell progenitors. ChIP-seq data for the transcription factors Runx1, Foxo1, Myc, Irf4, Spi1, and Jund, but ChIP-seq data for Jund did not contain statistically significant peaks by CLCbio Peak Finder [37]. This left Runx1, Foxo1, Myc, Irf4, and Spi1 as potential regulators for which potential target genes could be identified based on promoterproximal ChIPseq peaks (Table 1; proximal genes were identified by RGMatch [38]). We included EBF1 and Pax5 as key factors of known importance for B cell differentiation and CTCF as a negative control. Runx1 (TWM score = 1.75), Myc (TWM score = 1.20), and Foxo1 (TWM score = 1.14) were identified as highly ranked transcription factors (Fig 3C).
Runx1 showed transient down-regulation followed by up-regulation. RUNX1 ChIP-seq target genes showed enrichment for differential expression (P < 0.0005) and good coherence ( Fig 3C). Foxo1 was progressively up-regulated and showed good enrichment for differential expression of its ChIP-seq target genes (P < 0.001). Promoters bound by FOXO1 were mainly up-regulated (P < 10 × 10 −36 ). Myc was progressively down-regulated and also showed strong enrichment for differential expression of its ChIP-seq target genes (P < 0.001). Promoters bound by Myc were mainly down-regulated (P < 10 × 10 −9 ). The relevance of FOXO1 and Myc for primary B cell progenitor differentiation is supported by the up-regulation of Foxo1 mRNA and the down-regulation of Myc mRNA during in vivo B cell progenitor differentiation ( Fig 3E) and direct binding of Ikaros to the Foxo1 and Myc promoters in B cell progenitors ( Fig 3F).
For cell-state transitions for which sufficient ChIP-seq data are not available, TWM can be implemented based on promoter accessibility and the presence of transcription factor motifs in target gene promoters. We used DNase I hypersensitive sites sequencing (DNase-seq) to assess promoter accessibility at each time interval and the presence of transcription factor motifs for all 23 differentially expressed transcription factors (Table 1, Fig 3A). TWM identified Spib (TWM score = 2.62), Arid5a (TWM score = 1.81), Foxo1 (TWM score = 1.67), and the Myc antagonist Mxd4 (TWM score = 1.58) as top-scoring transcription factors (Table 1).
For validation of the TWM results, we focused on Foxo1 and Myc for two reasons. First, Foxo1 and the Myc pathway scored highly in both ChIP-seq and DNase-seq-based TWM approaches. Second, Foxo1 and Myc show monotonic changes in expression (up-and downregulation, respectively) during the B3 cell differentiation time course, which facilitates the analysis of their impact on B cell progenitor differentiation.

Transcriptional up-regulation of Foxo1 is linked to FOXO1 target gene expression during B cell progenitor differentiation
FOXO1 is essential for B cell development, and its role in the progression from B cell progenitor proliferation to B cell progenitor differentiation (i.e., the FR.C to Fr.D transition) has been characterized in exquisite detail [8,9,11,39]. According to current models, Fr.C cells proliferate in response to interleukin-7 (IL-7) receptor signaling, which inactivates FOXO1 via the phosphatidylinositol 3 kinase/Akt kinase/mechanistic target of rapamycin complex 1 (PI3K/Akt/ mTOR) axis (Fig 4Ai). Changes in signaling through the IL-7 receptor and/or pre-B cell receptor lead to the post-translational activation and stabilization of FOXO1 protein (Fig 4Aii). (C) Analysis of scRNA-seq time series data for a correlation between Foxo1 mRNA level and the expression of FOXO1 target genes. FOXO1 target genes were defined by ChIP-seq peaks in promoters as described above (black ORs and P values). Alternatively, FOXO1 target genes were defined FOXO1 then induces the expression of target genes that are critical for B cell progenitor differentiation and include B cell receptor signaling components, the recombinase activating genes Rag1 and Rag2, and the immunoglobulin light chain loci Igk and Igl (Fig 4Aiii) [8,9,11]. We confirmed that FOXO1 protein expression progressive increased during Ikaros-induced B3 cell differentiation (Fig 4B).
Our finding that Foxo1 mRNA increases over the time course of B3 cell differentiation and also between the Fr.C to Fr.D B cell progenitor stages in vivo raise a hitherto unanswered question about the role of FOXO1, namely, whether the transcriptional up-regulation we have uncovered contributes to the regulation of FOXO1 target genes in B cell progenitor differentiation. This is difficult to address by population RNA-seq data, because FOXO1 target genes were differentially expressed during the Fr.C to Fr.D transition, concomitant with the increase in Foxo1 mRNA (P < 10 −13 , for FOXO1 ChIP-seq targets versus nontargets, chi-squared test). To address this conundrum, we interrogated scRNA-seq data. We asked whether Foxo1 mRNA was positively correlated with the expression of FOXO1 target gene transcripts at the single-cell level and to what extent Foxo1 mRNA was a predictor of FOXO1 target gene expression independently of time. FOXO1 target genes were defined either by ChIP-seq peaks in promoters as described above or, alternatively, as FOXO1-dependent genes that were deregulated after genetic deletion of Foxo1 in early lymphoid/B cell progenitor cells [40]. We found significant correlations between We next used linear models to compute the extent to which the level of each differentially expressed target gene was explained by time after 4-OHT addition, by Foxo1 mRNA level, or a combination of time and Foxo1 mRNA level. Adding Foxo1 mRNA levels to the models significantly increased the fraction of the variance in FOXO1 target gene expression that could be explained by time after 4-OHT addition alone (Fig 4E). The scRNA-seq analysis indicates that the level of Foxo1 mRNA expression correlates with the expression of FOXO1 target genes in the same cells.

Repression of Myc is required for the regulation of a subset of Ikaros target genes in B cell progenitors
We next examined the regulatory relationship between Ikaros and Myc during B cell progenitor differentiation. We transduced B3 cells with Myc-ERt2 or control vector (Fig 5A). In an attempt to capture immediate Myc target genes in B3 cells, we isolated chromatin-associated RNA [41] for high-throughput sequencing 30, 60, and 180 min after induction with 4-OHT. Analysis over the time series (as described for Fig 1) identified 2,809 differentially expressed based on the deregulation of genes after genetic deletion of Foxo1 in early lymphoid/B cell progenitor cells ( [40] blue ORs and P values). B3 cells with low, medium, and high Foxo1 mRNA level were analyzed separately (left), cells with low and medium Foxo1 mRNA level were pooled (middle), or cells with medium and high Foxo1 mRNA level were pooled (right). Genes were defined as significantly associated with Foxo1 if P < 0.01 (Fisher test). We conducted an enrichment analysis comparing targets and nontargets. P values and ORs are shown. (D) Examples for the relationship between Foxo1 mRNA and mRNA expression of selected FOXO1 targets in scRNA-seq time series data. The source of the numerical data underlying this figure is listed in S1 Data. (E) The bars indicate the amount of variance (r 2 values calculated based on the difference between linear models) of FOXO1 target gene expression [40] after Ikaros induction by 4-OHT treatment is accounted for by time, Foxo1 mRNA expression, and the combination of time and Foxo1 mRNA expression in scRNA-seq. The source of the numerical data underlying this figure is listed in S1 Data. ChIPseq, chromatin immunoprecipitation sequencing; FOXO1, Forkhead Box O1; Fr.C, proliferating B cell progenitor; Fr. D, differentiating B cell progenitor; OR, odds ratio; scRNA-seq, single cell RNA sequencing; 4-OHT, 4-hydroxytamoxifen. https://doi.org/10.1371/journal.pbio.2006506.g004 Feedforward regulation of Myc genes after Myc induction (adj. P < 0.05; 1,485 up, 1,241 down). As a reference, we transduced B3 cells with Ikaros-ERt2 for chromatin-associated RNA-seq as described for Myc-ERt2. Analysis over the time series identified 1,354 differentially expressed genes after Ikaros induction (adj. P < 0.05; 662 up, 692 down) (S4 Table and Fig 5B). Gene ontology (GO) analysis of Ikaros-regulated genes showed enrichment of a spectrum of functional terms dominated by adhesion, differentiation, signaling, kinase activity, phosphorylation, and metabolism ( Fig 5B). GO term analysis of Myc-regulated genes showed enrichment mainly of metabolism-related terms within the time frame of these experiments ( Fig 5B); 387 of the 1,354 Ikaros-regulated genes identified by sequencing of chromatin-associated RNA were also differentially expressed in response to Myc induction, and thus were responsive to regulatory inputs from Myc as well as Ikaros. GO term analysis of these genes showed predominant enrichment of metabolic genes (Fig 5B), similar to Myc-regulated genes. The remaining 967 Ikaros-regulated genes were not differentially expressed in response to Myc induction. GO term analysis of these genes showed enrichment mainly of adhesion, differentiation, the immune system, signaling, kinase activity and phosphorylation (Fig 5B). Therefore, responsiveness to Myc separated Ikaros target genes broadly into those related to metabolism (Myc-responsive) and differentiation (Myc-unresponsive).

Feedforward regulation of Myc
Ikaros and Aiolos directly bind to the Myc promoter and repress Myc at the transcriptional level [7,18,42]. Whether repression of Myc is essential for coordinating the regulatory roles of Ikaros and Myc in B cell progenitors is currently unknown, although it has been shown that Myc can override the down-regulation of Ccnd3 (cyclin D3) and the up-regulation of the cellcycle inhibitor Cdkn1b (p27) by Ikaros and Aiolos [42]. These examples indicate that the regulation of at least some Ikaros target genes requires the repression of Myc. This raises the question of whether the regulation of target genes by Ikaros generally requires the down-regulation of Myc or whether there are classes of Ikaros target genes that do and do not require Myc repression. As an experimental system that removes Myc from the direct control of Ikaros, we simultaneously transduced B3 cells with both Ikaros-ERt2 and Myc-ERt2 and sequenced chromatin-associated RNA as described above. Of the 1,354 Ikaros target genes identified by induction of Ikaros-ERt2 alone, 512 differentially expressed also in the presence of Myc (240 were up-regulated, and 272 were down-regulated). We refer to these Ikaros target genes as "Mycresistant" (Fig 6A). The remaining 842 Ikaros target genes were only differentially expressed (adj. P < 0.05) in response to Ikaros alone (422 up, 420 down) but were not differentially expressed when Myc was expressed alongside Ikaros (Fig 6A and 6B). We refer to these Ikaros target genes as "Myc-sensitive" (Fig 6A). This analysis defined distinct sets of Ikaros target genes that do or do not require the down-regulation of Myc: coexpression of Myc neutralized the impact of Ikaros on Myc-sensitive target genes but not on Myc-resistant target genes ( Fig 6B).
GO analysis demonstrated that Myc-resistant Ikaros target genes were enriched for functional terms related to signaling, adhesion, differentiation and development, and the immune system ( Fig 6A). By contrast, Myc expression did interfere with Ikaros regulation of target genes related to metabolism, proliferation, and mRNA translation (Fig 6A). To ask whether these differences separated B cell-specific from housekeeping genes, we classified Ikaros-regulated genes during B cell progenitor differentiation as B cell specific or ubiquitously expressed. As a measure for how broadly genes are expressed, we used tau-values compiled from mouse Encyclopedia of DNA Elements (ENCODE) RNA-seq data across 22 tissues [43]. We assembled a panel of broadly expressed genes with tau-values < 0.25 that were expressed > 0.5 fragments per kilobase of exon model per million reads mapped (FPKM) in B3 cells as well as in spleen (an organ rich in B cells), and a panel of tissue-specific genes with tau-values > 0.70 (also expressed > 0.5 FPKM in B3 cells and spleen).
We found that the distribution of tissue-specific and housekeeping genes was skewed between Myc-sensitive and Myc-resistant Ikaros target genes. Housekeeping genes were enriched among Myc-sensitive Ikaros targets (P = 0.019, odds ratio = 1.64), and B3 tissuespecific genes were depleted among Myc-sensitive genes (P = 2.122 × 10 −5 , odds ratio = 0.42). Conversely, Myc-resistant Ikaros target genes were enriched for B3-specific genes over housekeeping genes (P value = 1.59× 10 −7 , odds ratio = 3.57). B3 tissue-specific genes were enriched, and housekeeping genes were depleted among Myc-resistant Ikaros targets (Fig 6C). Validation experiments showed that Myc was able to override Ikaros in the regulation of most glycolysis and glutaminolysis genes ( Analysis of Ikaros ChIP-seq data showed that promoters that were up-or down-regulated by Ikaros irrespective of Myc showed strong enrichment for Ikaros binding (P < 2.2 × 10 −16 , odds ratio = 6.30 and P < 2.2 × 10 −16 , odds ratio = 4.05, respectively; Fig 6D). Myc-sensitive Ikaros-regulated genes were also significantly enriched for Ikaros binding (P < 2.2 × 10 −16 , odds ratio = 5.92 for down-regulated and P < 1.7 × 10 −13 , odds ratio = 2.08 for up-regulated genes compared to all expressed genes; Fig 6D), suggesting that many promoters may receive regulatory inputs from both Ikaros and Myc.
The picture emerging from this analysis is that Ikaros regulates both ubiquitous and tissuespecific genes. Regulation of most ubiquitous genes by Ikaros is broadly sensitive to Myc dosage, whereas regulation of many tissue-specific genes by Ikaros is resistant to Myc dosage. These data are consistent with the existence of a regulatory circuit whereby Ikaros down-regulates Myc, and target gene expression reflects the combined effect of Ikaros expression and Myc down-regulation. The data suggest a model of progenitor cell differentiation in which the regulation of lineage-specific differentiation genes is coordinated with that of housekeeping genes. In B cell progenitor cell differentiation, such coordination is achieved by feedforward regulation of Myc by Ikaros; Fig 6E, left). Continued Myc activity interferes with the regulation of housekeeping functions and abolishes the coordinated regulation of housekeeping and lineage-specific differentiation genes (Fig 6E, right). This model is consistent with the known role of Myc as a major regulator of metabolism, the cell cycle, and RNA transcription and translation [44] and provides a framework for how Ikaros-mediated Myc repression contributes to differential gene expression during B cell progenitor differentiation.

Discussion
To identify transcriptional pathways that drive B cell progenitor differentiation, we developed the TWM algorithm, which identifies coherence between transcription factors downstream of Ikaros induction and their target gene expression over time. This approach pinpointed the transcriptional up-regulation of Foxo1 up-regulation and the repression of Myc as important components of Ikaros-mediated B cell differentiation.
FOXO1 target genes are integral to Fr.D, including immunoglobulin light gene rearrangement [8,9,11]. Previously, several studies have shown that FOXO1 protein is activated and stabilized at the post-translational level during B cell progenitor differentiation [8,9,11]. Here, we demonstrate that the progressive transcriptional up-regulation of Foxo1 correlates with the expression of FOXO1 target genes at the single cell level. Perturbation experiments will be required to establish the extent to which transcriptional regulation of Foxo1 is a functional contributor to FOXO1 target gene expression during B cell progenitor differentiation.
The coordination between proliferation, metabolic state, and differentiation is essential for normal development and homeostasis and has been attributed to antagonism between transcription factors that induce tissue-specific gene expression and cyclin-dependent kinases that promote cell-cycle entry [1][2][3]. By means of perturbation experiments designed to disrupt feedforward repression of Myc by Ikaros, we demonstrate that the regulation of lineage-specific differentiation genes can be dissociated from that of ubiquitously expressed genes, simply by uncoupling pathways by which Ikaros acts as an inducer of B cell progenitor differentiation from the repression of Myc.
Destabilization of one state and implementation of another has been studied extensively in cellular reprogramming [45][46][47], and our analysis introduces a similar idea to progenitor cell differentiation. Repression of Myc extinguishes key features of the undifferentiated state, and up-regulation of Ikaros family transcription factors-Foxo1, and presumably others-promotes a shift to the differentiated state. Consistent with this model, metabolic Myc target genes [48] were mostly repressed (S5B Fig), whereas FOXO1 target genes related to signaling, adhesion, and the immune system were mainly up-regulated (S5B Fig). Reminiscent of the scenario described here, FOXO and Myc control cell proliferation and metabolism in endothelial cells [49]. In B cell progenitors, both Myc and Foxo1 are direct targets of Ikaros, which links the extinction of the Fr.C-like state and establishment of the Fr.D-like state. Integration between cell type-specific and ubiquitous gene expression programs by interconnected regulators may account for other cell-state transitions.

Ethics statement
Mouse work was performed according to the Animals (Scientific Procedures) Act under the authority of project licence PPL70/7556 issued by the Home Office, United Kingdom. All work with mouse cells was in vitro, and the ARRIVE checklist is not applicable.
Cell culture, retroviral transduction, RNA, and protein methods were as described by Ferreirós and colleagues [7]. Isolation of chromatin-associated RNA was done as described by Ma and colleagues [42].

RNA-seq
To control for sources of variability, we implemented a scheme that tracks biological batches (3), conditions (Ikaros or control vector), time points (6), library preparations (6 × 6), bar codes, sequencing runs, and flow cell lanes. Each RNA-seq library was split into two (total of 72) to account for variability associated with sequencing. For sequencing, the 72 libraries were distributed across 4 flow cells with 3 libraries per flow cell. Each lane contained different libraries, batches, time points, and conditions. We aimed for 50 million reads per library × 4 sequencing runs, equaling 100 million reads per sample. Strand-specific libraries were sequenced on an Illumina HiSeq 2500, at 75 nucleotides paired-end. Analysis followed published guidelines [50] using Tophat2 "very-sensitive" mode only allowing a unique best mapping to map sequences to the mm10 reference genome. Trimming was applied to remove Illumina primers and low-quality nucleotides. ht-seq intersection-option was used to assign fragments to genes. We used cqn to correct for GC content and gene-length bias and a nonparametric version of ComBat to correct for RNA-seq library-preparation effects.
We identified differential expression by combining limma [19] to identify sharp changes and maSigPro [20] to identify changes over time. In limma, we computed significant differences between Ikaros (4-OHT and Ikaros effects) and control (4-OHT effects)-between consecutive time points and between every time point and 0 h. To compute a final limma-derived P value for each gene, we combined all contrasts using eBayes function of limma and computed the F-statistic, the associated P value, and the adj. P value (using Benjamini-Hochberg multiple testing correction). In maSigPro, we considered 2 conditions (Ikaros and control) and identified genes whose trends separated over time. We considered a gene to be differentially expressed if any of the following conditions applied: (a) limma FDR < 0.001, (b) maSig-Pro R 2 > 90%, or (c) limma FDR < 0.01 and maSigPro R 2 > 60%.

Gene expression microarray analysis
We used limma to analyze gene expression array data from www.immgen.org.

Gene set analysis
We used Generally Applicable Gene-set Enrichment [51] to perform gene set analysis over ranked lists of genes [52] and Fisher test to perform gene set analysis over clusters of genes. We used default parameters and defined as significant those gene sets with associated Benjamini-Hochberg adj. P < 0.1. We used gene sets included in the GSKB Bioconductor package [53].  (x 1 , x 2 , . . ., x A ) 2. Differences between consecutive times r T ¼ fr Step 1: Log 2 fold change 1. Computing log 2 fold change per contrast: 8 tf i 2 TF; 8 r j 2 r T estimate logFC tf i ;r j 2. Computing aggregated log 2 fold change for consecutive contrasts: Step 2: Odds ratios Step 3: Coherence

Computing direction for all genes and transcription factors for all contrasts:
8 g 2 G; 8 r j 2 r T estimate sign g;r j ¼ signðlogFC g;r j Þ 8 tf i 2 TF; 8 r j 2 r T estimate sign tf i ;r j ¼ signðlogFC tf i ;r j Þ 2. Computing coherence per transcription factor for each gene and for each consecutive contrast: Computing coherence per transcription factor per consecutive contrast: Step 4: Combine 1. Combine log 2 fold change and odds ratio measurements: Cr j 2 Cr logFC tf i ;Cr j �OR tf i ;Cr j 2. Combine all coherence: 3. Combine all measurements:

B-ALL analysis
CEL files for 1404 B-ALL patients were obtained from their respective publications [23,[54][55][56][57]. The raw data were normalized together with the Robust Median Average (RMA) algorithm, whereas systematic variations across studies were eliminated using ComBat. Differentially expressed genes between Ikaros-mutated and Ikaros wild-type subjects were identified with limma, using surrogate Variable Analysis [58] to account for possible latent confounders. Pathway enrichment analysis was performed using (a) mean-rank gene set enrichment [59] on gene sets from the Molecular Signature Database [52] and (b) GSEA [52] on Ikaros targets identified by Chip-seq in mouse B3 cells. Enrichment analysis was repeated on the 10,111 genes expressed in B3 cells.

ChIP-seq data bioinformatics analysis
We analyzed ChIP-seq data for B cell-associated transcription factors (S3 Table). EBF1, PU.1, IRF4 [15], PAX5 [60], CTCF [6], and RUNX1 [61] were from pro-B cells; FOXO1 was from pre-B cells [8]; Myc was from CH12 cells [62]; and Ikaros was from B3 cells [7]. We used Bow-tie2 [63] for mapping to the mm10 reference genome. When fastq files were available, we used "end-to-end" and "very-sensitive" parameters, otherwise we used "local" parameter. In all cases, we filtered duplicated reads and any reads with quality scores below 20. We applied CLCbio Peak Finder tool [37] to identify peaks for each sample using default parameters; we used control libraries when available (S1 Table). We considered peaks with adj. P < 0.01.

DNase-seq profiling
DNase-seq was performed on 20 to 25 million cells with 3 biological replicates for all time points and conditions. Briefly, cells were harvested and washed with cold 1× PBS. Lysis conditions were optimized to ensure >90% recovery of intact nuclei. Enrichment of DNaseI hypersensitive fragments (0-500 bp) was performed using a low-melt gel size selection protocol. Library preparation was performed and sequenced as 43-bp paired-end NextSeq 500 Illumina reads. DNaseI libraries were sequenced at a minimum depth of 20 million reads per each biological replicate and a total of 200 million per time and condition. DNase-seq reads were trimmed to 36 bp and paired-end mapped to the mm10 reference genome using Bowtie2 [63] with the following options: -v 2 -k 1 -m 1-best-strata. We used Wellington to identify the footprinting sites per time and condition [64]. We used MATCH algorithm from TRANSFAC to predict binding sites (BSs) of transcription factor motifs in FP. To minimize the number of false positive BS predictions we defined an FOS-related optimal threshold by using as a goldstandard the ChIP-seq IKZF1 peaks. We identified the optimal level of FOS [64] at which DNase-seq-derived Ikzf1 BS predictions obtained by MATCH were optimally (minimizing false positives) predicting CHIP-seq IKZF1 peaks. Proximal genes for ChIP-seq-and DNaseseq-derived peaks were identified with default parameters in RGmatch [38].

scRNA-seq profiling
We isolated cells using the Fluidigm C1 System. Single-cell C1 runs were completed using the smallest IFC (5-10 μm). Cells were collected at a concentration of 400 cells/μl in a total of 50 μl.
To optimize cell capture rates on the C1, buoyancy estimates were optimized prior to each run. Single-cell capture efficiency was between 75% and 90% across 8 runs. Each C1 capture site was visually inspected for single-cell capture and cell viability. After visualization, the IFC was loaded with Clontech SMARTer kit lysis, RT, and PCR amplification reagents. After harvesting, cDNA was normalized across all libraries from 0.1 to 0.3 ng/μl, and libraries were constructed using Illumina's Nextera XT library prep kit per Fluidigm's protocol. Constructed libraries were multiplexed and purified using AMPure beads. The final multiplexed single-cell library was analyzed on an Agilent 2100 Bioanalyzer for fragment distribution and quantified using Kapa Biosystem's universal library quantification kit. The library was normalized to 2 nM and sequenced as 75-bp paired-end dual indexed reads using Illumina's NextSeq 500 system at a depth of approximately 1.0 to 2.0 million reads per library. Each Ikaros time point was performed once, with the exception of 18-and 24-h time points, in which 2 C1 runs were required in order to achieve approximately 50 single-cells per time point. We mapped 560 scRNA-seq libraries with Tophat2 [65] to the mouse Ensembl gene annotations and mm10 reference genome. We excluded single-cell libraries with a mapping rate less than 50% and less than 450,000 mapped reads; we obtained a total of 324 single cells for all subsequent analysis. Cufflinks [66] version 2.2.1 was used to quantify expression from single-cell libraries using Cuffquant. Gene expression data for each singlecell library were merged and normalized into a single data matrix using Cuffnorm. Monocle [21] was used to compute pseudotime trajectories, in which cells are ordered by their actual progress in the differentiation course rather than by their experimental time point.

Validation of Foxo1 expression and FOXO1 targets in single cells
We filtered for cells with expression greater than 0 and grouped cells based on quantiles 0.33 and 0.66 of Foxo1 expression levels (3 levels: low, middle, and high). Then we did the same for each gene. Using both groupings, we computed a contingency table (Fig 6C, left panel) and a P value for each gene and defined genes as significantly regulated and not significantly regulated. We found that significantly regulated genes were enriched in FOXO1 targets (P = 0.05). We repeated the analysis, first by combining low and middle into a single level (Fig 5C, middle panel, P < 0.05) and second by combining middle and high into a single level (Fig 5C, right panel, P < 0.07). Finally, we combined all significantly regulated definitions and considered a gene to be Foxo1 regulated if the contingency table analysis was significant for any of the 3 analyses. In this case, significantly regulated genes were enriched in FOXO1 targets (P < 0.001). To quantify and compare the role of time and Foxo1 mRNA levels in FOXO1 targets, we first filtered for cells with expression greater than 0 for each gene pair. Then we computed 2 linear models: in the first one, mRNA gene expression was predicted using time; in the second, mRNA gene expression was predicted using a 6-level grouping of Foxo1 mRNA expression. Finally, for each gene we compared the r 2 values derived from each model. Measurement of ECAR and OCR was done using Seahorse XF24 or XF96 extracellular flux analyzers as advised by the manufacturers (Seahorse Bioscience; North Billerica, MA). To assess the impact of Ikaros and Aiolos on ECAR and OCR, primary B cells were transduced with IRES-GFP, Ikzf1-IRES-GFP, or Ikzf3-IRES-GFP; sorted for GFP expression 72 h later; and rested for 3 to 6 h in cultured in the presence of IL-7.

Categorization of GO terms in broad functional classes
"Immune system" includes the terms "immune," "host defense," "B cell," "T cell," "myeloid," "lymphocyte," "leukocyte," and "hematopoiesis." "Signaling" includes the terms "signal," "signaling," "response," "stimulus," "communication," and "activation." "Adhesion" includes the terms "adhesion" and "integrin." "Differentiation" includes the terms "differentiation" and "development." "Metabolism" includes the terms "metabolic," "metabolism," "biosynthetic," "biosynthesis," and "catabolic." "Translation" includes the terms "translation," "ribosome," and "ribonuclear." "Proliferation" includes the terms "proliferation," "chromatid," "spindle," "mitosis," "mitotic," "cell cycle," "cell division," "DNA synthesis," and "DNA replication." Supporting information S1 Data. Underlying data for figures. (XLSX) S1 Fig. Ikaros target genes identified during B cell progenitor differentiation are deregulated in IKZF1-mutant B-ALL. We assembled gene expression profiles of 1,404 B-ALL samples with and without IKZF1 mutations [23,[54][55][56][57]. IKZF1 was mutated in 406 samples (29%). B-ALL samples with and without IKZF1 mutations differed in 7,222 genes (0.1 FDR; S1A Fig) [67,68] with significant enrichment for 317 of 1,415 gene sets in the molecular signature database [69] (P < 5 × 10 −4 ), including FOXO, Myc, and CXCR4 pathways, adherens junction, cell cycle, integrin-, B cell receptor-, PI3K-, ERK-, MAPK-, and NF-kB signaling pathways, and the PYK2 pathway, which links leukemia with cell adhesion [25] (S5 Table). Among B-ALLs with BCR-ABL1 translocations, 71% had IKZF1 mutations; 1,228 genes were differentially expressed in IKZF1 mutated samples (S1B Fig) with enrichment of FOXO, Myc, CXCR4, chemokine signaling, cell cycle, transcription, adherens junction, focal adhesion kinase, integrin, and B cell receptor signaling as well as the downstream transduction pathways PI3K, ERK, MAPK, NF-kB, and NFAT (P < 5 × 10 −4 , S6 Table). Differentially expressed genes in IKZF1-mutated B-ALL were enriched for Ikaros target genes identified by ChIP-seq in mouse B3 cells (S1C Fig, P = 0.0189 for all samples and P = 0.0017 for BCR-ABL1 samples). There was significant overlap between differentially expressed genes in IKZF1-mutated B-ALL and early Ikaros targets regulated between 0 and 2 h after 4-OHT (Fig 1F, odds ratio = 2.53, adj. P = 0.02 for the 200 most differentially expressed genes). Analysis of the BCR-ABL subset of B-ALL samples identified JAK-STAT (S1D-S1F Fig), G-protein coupled receptor and cytokine signaling (S6 Table). Gene-based prognostic models define subgroups of B-ALL with poor clinical outcome [17,28,70], and a set of 139 asparaginase and vincristine resistance genes [70] was enriched for differential expression during the Fr.C to Fr.D transition (P < 0.05). A 256-probe set "Ph+like" signature indicative of poor prognosis [17] was significantly enriched among genes differentially expressed at 2, 6, and 12 h (all P < 0.05) after nuclear translocation of Ikaros. Combining 2 distinct Ph+like signatures [17,28] resulted in enrichment at all time points (P < 0.05). As a control for the overlap in gene expression between Ikaros-induced B3 cells and IKZF1 mutations in B-ALL, we used recurrent non-IKZF1 genetic lesions in AML, or B-ALL with 4-OHTtreated B3 cells transduced with ERt2 control vector instead of Ikaros-ERt2. Therefore, analysis of B cell progenitor cell state transitions can reveal gene expression signatures with relevance to human disease. (A, B) Differential expression in 1,404 B-ALL samples (A) and of the BCR-ABL1 subset (B). Log 2 fold change between wild-type and IKZF1-mutated samples log10 adj. P values are indicated. Dashed line: log 2 fold change > 0.5; blue: FDR > 0.1. The sources of the numerical data underlying this figure are listed in S1 Data. (C) GSEA of Ikaros-bound genes identified by ChIP-seq in mouse B3 cells in genes differentially expressed in IKZF1-mutated versus nonmutated human B-ALL. The x-axis is the list of genes ordered by magnitude of differential expression, whereas the y-axis represents the enrichment score for the Ikaros tar- Step-by-step account of TWM. TWM ranks TFs by combining 3 sources of information: (i) TF binding to gene promoters in B cell progenitors from publicly available ChIP-seq data, (ii) differential expression within the Ikaros time series between consecutive time points and over the entire time span (0-24 h), and (iii) coherence between the expression of TFs and their targets over time. The approach has 4 steps: (1) For each TF, log 2 fold change in expression is averaged for each consecutive pair of contrasts (red). (2) Enrichment of TF binding over differentially expressed genes is averaged for each consecutive pair of contrasts (yellow).
(3) For each TF, coherence is determined between expression of the TF and its target genes over time (blue). (4) The log 2 fold change and odds ratio for each pair of contrasts are multiplied to generate a combination matrix (gray, center), and the sum of these values is multiplied with the global coherence score to determine the final TWM score (gray, right). ChIPseq, chromatin immunoprecipitation sequencing; TF, transcription factor; TWM, transition weight matrix.  [8], the model incorporates previous [12,42] and current data. Phase 1 is dominated by IL-7 signaling (panel A; blue indicates posttranslational regulation), phase 2 by FOXO1, pre-B cell receptor signaling, and Ikaros (B). Of 21 validated Myc target genes in core metabolism [30], 19 were differentially expressed during the Fr.C to Fr.D transition. Of these, 18 were down-regulated and 1 was up-regulated. Of 2,186 putative FOXO1 target genes defined by FOXO1 promoter binding, 685 were up-and 308 were down-regulated, including genes related to signaling (81 upand 24 down-regulated), adhesion (31 up-and 6 down-regulated), and the immune system (23 up-and 10 down-regulated). The source of the numerical data underlying this figure is listed in S1 Data. BCR; FOXO1; Fr.C, proliferating B cell progenitor; Fr.D, differentiating B cell progenitor; IL-7, interleukin-7; Myc, MYC proto-oncogene. (PNG) S1 Table. Differential expression summary in B3 RNA-seq time series. Limma.P.value (Limma.adj.P.Val) denotes the P values of the moderated F-statistic test using limma [19]. MaSigProPval and R2 denote the P value and the r-squared of the linear model computed using the MaSigpro tool [20]. CONSENSUS_DE is 1 for those genes that were characterized as differentially expressed (see Methods). logFC and Adjp denotes fold change and adjusted P value, respectively, for each contrast. RNA-seq, RNA sequencing. (XLSX) S2 Table. ChIP-seq data sets for Th17 TWM analysis. ChIP-seq, chromatin immunoprecipitation sequencing; Th17, T helper 17; TWM, transition weight matrix. (XLSX) S3 Table. ChIP-seq data sets selected for the B3 TWM analysis. Mapping, filtering, and peak columns describe the methodology used in each case, dependent on the quality of reads and availability of background samples. Cell column denotes the cell type analyzed. ChIP-seq; TWM, transition weight matrix.