Research ArticleCancer

Targetable genetic alterations of TCF4 (E2-2) drive immunoglobulin expression in diffuse large B cell lymphoma

See allHide authors and affiliations

Science Translational Medicine  19 Jun 2019:
Vol. 11, Issue 497, eaav5599
DOI: 10.1126/scitranslmed.aav5599

Lymphoma’s gain is its loss

Activated B cell (ABC-like) diffuse large B cell lymphoma (DLBCL) can be curable, but therapies are needed for those who relapse or are refractory to current treatments. Jain et al. found that a percentage of ABC-like DLBCL patient tumors had copy gains on part of chromosome 18. The gains increased expression of transcription factor TCF4 (E2-2), which, in turn, activated immunoglobulin μ and MYC. Finding that tumors with this gain were dependent on TCF4 and that TCF4 was regulated by bromodomain protein BRD4, they tested bromodomain inhibition in xenograft models and reported reduced tumor growth and enhanced survival. This work suggests that targeting TCF4 may hold promise against ABC-like DLBCL.

Abstract

The activated B cell (ABC-like) subtype of diffuse large B cell lymphoma (DLBCL) is characterized by chronic activation of signaling initiated by immunoglobulin μ (IgM). By analyzing the DNA copy number profiles of 1000 DLBCL tumors, we identified gains of 18q21.2 as the most frequent genetic alteration in ABC-like DLBCL. Using integrative analysis of matched gene expression profiling data, we found that the TCF4 (E2-2) transcription factor gene was the target of these alterations. Overexpression of TCF4 in ABC-like DLBCL cell lines led to its occupancy on immunoglobulin (IGHM) and MYC gene enhancers and increased expression of these genes at the transcript and protein levels. Inhibition of TCF4 activity with dominant-negative constructs was synthetically lethal to ABC-like DLBCL cell lines harboring TCF4 DNA copy gains, highlighting these gains as an attractive potential therapeutic target. Furthermore, the TCF4 gene was one of the top BRD4-regulated genes in DLBCL cell lines. BET proteolysis-targeting chimera (PROTAC) ARV771 extinguished TCF4, MYC, and IgM expression and killed ABC-like DLBCL cells in vitro. In DLBCL xenograft models, ARV771 treatment reduced tumor growth and prolonged survival. This work highlights a genetic mechanism for promoting immunoglobulin signaling in ABC-like DLBCL and provides a functional rationale for the use of BET inhibitors in this disease.

INTRODUCTION

Diffuse large B cell lymphoma (DLBCL) is the most common form of lymphoma and is curable in about 60% of patients using a combination chemoimmunotherapy regimen, rituximab, cyclophosphamide, doxorubicin, vincristine, and prednisone (R-CHOP) (1, 2). However, those that are refractory to, or relapse following, first-line therapy have a dismal outcome (3). Chimeric antigen receptor (CAR)–T cells are likely to change the landscape of outcomes in relapsed/refractory patients, but a large number of patients are not eligible for CAR-T therapy, and roughly 50% of those that receive CAR-T progress within 12 months (4). Additional rationally targeted therapeutic strategies are therefore needed for DLBCL.

The clinical heterogeneity of DLBCL is underpinned by its molecular heterogeneity, with the major distinction among cases being between the germinal center B cell (GCB)–like and activated B cell (ABC)–like “cell of origin” subtypes identified by gene expression profiling (5). The GCB-like subtype shows transcriptional similarities to normal germinal center B cells, whereas the ABC-like subtype shows transcriptional similarities to CD40-activated B cells or plasmablasts. Patients with ABC-like DLBCL have worse overall survival compared to patients with GCB-like DLBCL when treated with the standard-of-care combination chemotherapy (CHOP) and rituximab (R-CHOP) regimen (6). The ABC-like DLBCL subtype expresses immunoglobulin μ (IgM) (7) in more than 90% of cases, which in association with CD79A and CD79B forms the B cell receptor (BCR) signaling complex that drives chronically active BCR signaling. Several genetic alterations have been shown to promote this signaling, including mutations of the CD79A, CD79B, CARD11, and MYD88 genes (811). However, these mutations only account for about two-thirds of ABC-like DLBCL cases (12), suggesting that one or more genetic drivers remain to be defined.

A common mechanism for tumorigenesis is the gain or loss of DNA-encoding oncogenes or tumor suppressor genes, respectively. These copy number alterations (CNAs) perturb a higher fraction of the cancer genome than somatic nucleotide variants or small insertion/deletions (InDels) and are critically important to cancer etiology (13). Here, we integrated multiple datasets, including DNA copy number profiles of tumors, and identified a DNA copy number gain of the E2 transcription factor TCF4 as the most frequent genetic alteration in ABC-like DLBCL. We show that TCF4 is capable of driving IgM expression and is amenable to therapeutic targeting through bromo- and extra-terminal domain (BET) protein inhibition. These data therefore highlight a genetic basis for ABC-like DLBCL with potential implications for future clinical studies.

RESULTS

DNA copy number gains of 18q are the most frequent genetic alteration in ABC-like DLBCL

To identify CNAs in DLBCL, we interrogated the genomic profiles of 1000 DLBCL tumors using the GISTIC2 algorithm (14). These data included high-resolution single-nucleotide polymorphism (SNP) microarrays from 860 previously published cases, in addition to next-generation sequencing (NGS)–derived profiles from our own cohort of 140 cases (tables S1 and S2). The GISTIC analysis revealed 20 loci with significant DNA copy number gain and 21 loci with significant DNA copy number loss (Q < 0.1; Fig. 1A and table S3). As previously observed, 19.1% of tumors (191 of 1000) did not bear any of these DNA CNAs (15). Using a subset of 448 cases with cell of origin subtype available, we identified 9 CNAs that were significantly more frequent in ABC-like DLBCL and 11 that were significantly more frequent in GCB-DLBCL (Fisher, Q < 0.1; Fig. 1B, fig. S1, and table S4). The most frequent genetic alteration in ABC-like DLBCL was gain of 18q21.2, which we observed in 44% of tumors. In line with their enrichment in ABC-like DLBCL, 18q21.2 gains were associated with significantly reduced overall survival in both CHOP- (log-rank, P = 0.0068) and R-CHOP–treated (log-rank, P = 0.0034) patients for whom data were available (Fig. 1, C and D). However, this effect was not independent of cell of origin subtype (fig. S2). Using 199 tumors with matched cell of origin subtype, DNA copy number data, and mutation status for 40 genes, we observed that the frequency of 18q21 gains (23.1% of all tumors; 40.7% of ABC-like tumors) was higher than other ABC-like DLBCL-associated somatic mutations, including mutations in MYD88 (16.6% of all tumors; 33.3% of ABC-like tumors), CD79B (7.5% of all tumors; 18.5% of ABC-like tumors), and other ABC-associated genes (Fig. 1E, fig. S3, and table S5). Because multiple genetic alterations were associated with ABC-like DLBCL, we used the REVEALER algorithm (16) to identify the set of genetic alterations that best explained the ABC-like DLBCL signature. Using a set of 87 CNAs and recurrently mutated genes as the feature set and MYD88 mutations as the seed feature, REVEALER identified an additional four genetic alterations including 18q21.2 gain as those best associating with the ABC-like signature (Fig. 1F). Gains of 18q21 are therefore the most frequent genetic feature of ABC-like DLBCL and are predicted to contribute to this molecular phenotype.

Fig. 1 DNA copy number gains of 18q21.2 are the most frequent genetic alteration in ABC-like DLBCL.

(A) GISTIC analysis of DNA copy number profiles form 1000 DLBCL tumors identified 21 peaks of DNA copy loss (blue, left) and 20 peaks of DNA copy gain (red, right). The green line indicates the significance threshold of Q value = 0.1. (B) The GISTIC peaks from (A) are shown with reference to their frequency in ABC-like (orange) compared to GCB-like (green) cell of origin subtypes (*Q < 0.1). DNA copy gains of 18q21.2 were the most frequent alteration in ABC-like DLBCL cases. (C and D) A Kaplan-Meier plot of overall survival for patients treated with (C) CHOP combination chemotherapy or (D) R-CHOP shows that the presence of 18q21.2 gain is associated with poor outcome. (E) The frequency of 18q21 gains (+18q21) is shown relative to other somatic mutations that are associated with the ABC-like DLBCL subtype. (F) REVEALER analysis was performed to identify the set functionally complementary genetic features that likely contribute to the ABC-like DLBCL molecular phenotype. COO, cell of origin. Mutations of MYD88 were used as the seed feature. Mutations of IRF4, PIM1, and CD79B, and DNA copy gains of 18q21 were selected as additional features that likely also contribute to the phenotype §Seed feature. IC, information coefficient; CIC, conditional information coefficient.

The TCF4 (E2-2) transcription factor is an oncogene in ABC-like DLBCL

Gains of 18q have been previously attributed to the BCL2 oncogene (15, 17). However, our analysis of this large cohort provided the resolution to identify two significant peaks of DNA copy gain on chromosome 18: 18q21.2 (16 genes, Q = 4.8 × 10−14) and 18q22.1 (70 genes, Q = 1.1 × 10−7; table S3). We further integrated gene expression profiling data from 249 tumors to identify the likely targets of these lesions by testing for an increase in gene expression within the most significant peaks of DNA copy gain. This highlighted TCF4 and BCL2 as likely targets of the 18q21.2 and 18q22.1 gains, respectively (Fig. 2A and table S6). Most 18q CNAs incorporated both of these genes (Fig. 2, A and B), with only 7.3 or 1.0% of ABC-like DLBCLs having excisional CNAs targeting TCF4 or BCL2 alone, respectively (Fig. 2B). We observed that 18q21.2 DNA copy number gains were associated with significantly higher TCF4 transcript abundance when comparing within cell of origin subtypes (GCB-like, P = 0.0031; ABC-like, P < 0.0001). However, the expression of TCF4 in ABC-like DLBCL tumors was higher than that in GCB-like tumors, such that ABC-like tumors with diploid 18q21.2 had TCF4 expression equivalent to GCB-like tumors with 18q21.2 gain (Fig. 2C). In line with this, ABC-like DLBCL cell lines had higher expression of TCF4 protein than GCB-like DLBCL cell lines irrespective of DNA copy number, and this expression was increased in association with DNA copy gain (Fig. 2, D and E). Therefore, TCF4 is more highly expressed in ABC-like DLBCL compared to GCB-like DLBCL and is further increased by DNA copy gain. However, this observation should be prospectively validated by immunohistochemistry in patient tumors when appropriate antibodies become available.

Fig. 2 The TCF4 gene is a key target of 18q DNA copy number gains.

(A) A schematic of 18q DNA copy number gains is shown, with each line representing a single tumor and deeper shades of red indicating higher DNA copy number. The GISTIC Q value is shown at the top of the diagram and the two significant (Q value of <0.1) peaks are highlighted with arrows. (B) The frequency of tumors with DNA copy number gains that include both the TCF4 and BCL2 genes (purple), the TCF4 gene and not the BCL2 gene (pink), or the BCL2 gene and not the TCF4 gene (yellow) is shown for all tumors (left) and for the ABC-like only (right). (C) TCF4 expression from microarrays is shown for GCB-like DLBCL with diploid 18q (green, n = 74), GCB-like DLBCL with 18q copy gain (beige, n = 22), ABC-like DLBCL tumors with diploid for 18q (orange, n = 59), and ABC-like DLBCL tumors with 18q DNA copy number gains (red, n = 52). P values are from Mann-Whitney test. P > 0.05 is denoted as ns, **P < 0.01, and ***P < 0.001. (D) Protein expression of TCF4 and BCL2 in ABC-like DLBCL cell lines, ordered according to increasing DNA copy number of the TCF4 locus. The GCB-like cell line, OCI-Ly1, is shown for reference. (E) Protein expression of TCF4 and BCL2 in GCB-like DLBCL cell lines. The ABC-like cell lines, U2932 and SUDHL2, are shown for reference. (F) The frequency of TCF4 DNA copy gains, TCF3 mutation, and ID3 mutation in a cohort of 108 BL tumors. A Fisher’s exact test was used to compare the frequency of TCF3 and ID3 mutations (black bars) in tumors with TCF4 gain (red bars) compared to those without TCF4 gain, P = 0.0191.

The TCF4 gene encodes an E2 family transcription factor, E2-2. Mutations of another E2 transcription factor, TCF3, and its negative regulators ID2 and ID3 are frequent in Burkitt lymphoma (BL) and promote immunoglobulin signaling (18, 19). We interrogated the mutation status of these genes and TCF4 copy gains in our cohort of 140 DLBCLs and a previous cohort of 108 BLs that were sequenced and analyzed with the same approach (20). We did not observe recurrent mutations of TCF4 or ID2 in this BL cohort, and mutations of TCF3 and ID3 were infrequent in DLBCL. However, gains of TCF4 were present at the same frequency as TCF3 mutations in BL (18%). Furthermore, TCF4 gains were significantly mutually exclusive from TCF3 and ID3 mutations (Fisher, P = 0.019; Fig. 2F), suggesting that TCF4 gains may serve a similar function as TCF3/ID3 mutations in promoting immunoglobulin signaling. We speculate that the preference for mutating TCF3 and ID3 in BL is due to their higher expression in BL compared to DLBCL, which may be related to the germinal center B cell–specific expression of TCF3 (fig. S4). In contrast, TCF4 was variably expressed in both BL and DLBCL and did not decline in expression in normal B cells following germinal center exit (fig. S4). These data show that the TCF4 gene is highly expressed in ABC-like DLBCL, with expression further promoted by frequent 18q21.2 DNA copy gains, and implicates TCF4 in immunoglobulin signaling.

TCF4 regulates IGHM and MYC expression in ABC-like DLBCL

To identify potential target genes of TCF4, we performed differential gene expression analysis of primary DLBCL tumors with TCF4 DNA copy gain (n = 51) compared to those without (n = 59). We limited this analysis to ABC-like tumors to eliminate the confounding effect of genes that differ in expression between cell of origin subtypes. A total of 355 genes (472 probe sets) and 87 genes (107 probe sets) had significantly higher or lower expression in tumors with TCF4 gain, respectively (Q < 0.1, fold change of ≥1.2; Fig. 3A and table S7). We performed chromatin immunoprecipitation sequencing (ChIP-seq) on the ABC-like DLBCL cell lines SUDHL2 and TMD8 with tetracycline-inducible Myc-DDK–tagged TCF4 to define whether these genes were direct transcriptional targets of TCF4 (Fig. 3A). The TCF4 protein expression induced in SUDHL2 and TMD8 cells was titrated to be comparable to that in the U2932 cell line with TCF4 copy gain (fig. S5). Intersecting significant (P < 0.01) peaks from both cell lines, we identified TCF4 binding proximal to 180 of 355 genes with increased expression and 46 of 87 genes with decreased expression in tumors with TCF4 copy gain (Fig. 3, A and B, and table S8). These peaks showed a significant enrichment of motifs containing E-box consensus sequences (CANNTG; Q value of <0.1; fig. S6), and many of the same regions were also bound by TCF4 in plasmacytoid dendritic cell neoplasm (fig. S6) (21), providing evidence that we detected on-target binding. The immunoglobulin heavy-chain locus was among the most highly TCF4-loaded regions in the genome (Fig. 3B), in line with the higher expression of IGHM in ABC-like DLBCL tumors with TCF4 copy gain (Fig. 3C). These included peaks immediately upstream and downstream of the IGHM and IGHD genes, respectively, in regions with corresponding acetylated histone 3 at lysine 27 (H3K27Ac) in normal CD20+ B cells, indicating that they are bona fide enhancers (Fig. 3D). We validated TCF4 binding to these two immunoglobulin loci, as well as a downstream enhancer of MYC, using ChIP-qPCR (quantitative polymerase chain reaction). This confirmed TCF4 binding at all three loci that was significantly greater than that of a control IgG antibody (P < 0.001) in three cell lines with tetracycline-inducible expression of TCF4 (SUDHL2, TMD8, and HBL1) and two cell lines with high baseline expression of TCF4 (U2932 and RIVA) (P < 0.001; Fig. 3E). Immunoglobulin- and MYC-encoding genes are therefore likely targets of the TCF4 transcription factor in ABC-like DLBCL.

Fig. 3 TCF4 regulates IgM expression in ABC-like DLBCL.

(A) Differential gene expression analysis of 110 primary ABC-like DLBCL tumors with or without TCF4 DNA copy number gain. ChIP-seq signal for TCF4 from SUDHL2 and TMD8 cell lines shows TCF4 binding at intragenic or distant enhancer elements. (B) Significant TCF4 ChIP-seq peaks from SUDHL2 and TMD8 cells (P < 0.01) are shown, ordered from strongest (top) to weakest (bottom) signal ratio compared to the input control. (C) A violin plot shows expression of IGHM in ABC-like DLBCL tumors with either a 18q21 gain or diploid 18q21. (D) Two of the TCF4 peaks at the immunoglobulin heavy-chain locus are shown for TCF4 ChIP (blue) compared to the equivalent input control (gray). Yellow shading indicates the statistically significant peak (P < 0.01). For reference, Encyclopedia of DNA elements (ENCODE) data for H3K27 acetylation (H3K27Ac) ChIP-seq in CD20+ B cells are shown. (E) The binding of TCF4 to the two immunoglobulin heavy-chain loci and the MYC enhancer locus by ChIP-qPCR in two cell lines with high TCF4 DNA copy number and protein expression (U2932 and RIVA) and three cell lines with tetracycline-inducible TCF4 expression (SUDHL2, TMD8, and HBL1). IP, immunoprecipitation. The signal was significantly above that of the isotype control IgG antibody for all cell lines and loci. Student’s t test (P < 0.001). Each bar represents the mean ± SEM of three independent experiments.

Tetracycline-inducible expression of TCF4 in ABC-like DLBCL cell lines led to a marked increase in IGHM at the transcript (Fig. 4A) and protein levels (Fig. 4B). Confocal microscopy showed that the TCF4-induced IgM protein accumulated on the cell surface and in intracellular compartments (fig. S7). In comparison, BCL2 expression was not induced by TCF4 overexpression, and TCF4-induced MYC expression was restricted to two of the three cell lines (SUDHL2 and HBL1; fig. S8). To assess the consequence of increased IgM expression on BCR signaling, we performed BCR cross-linking with anti-IgM Fab (aIgM), with or without tetracycline-inducible TCF4 expression, and assessed the activation of the proximal BCR signaling kinases Bruton’s tyrosine kinase (BTK) and B-cell linker (BLNK). In the absence of BCR cross-linking, TCF4-induced IgM expression had no detectable effect on BTK or BLNK phosphorylation. However, the increase of BTK and BLNK phosphorylation was significantly potentiated by TCF4 overexpression after BCR cross-linking (P < 0.05; Fig. 4, C and D). These data show that TCF4-induced IgM expression can promote BCR signaling in ABC-like DLBCL in the presence of BCR stimulation.

Fig. 4 Induced expression of TCF4 in ABC-like DLBCL cell lines drives MYC and IgM expression and potentiates BCR signaling.

(A) Tetracycline-induced expression of TCF4 in ABC-like DLBCL cell lines with low TCF4 copy number resulted in an increase in IGHM transcripts compared to control cells. (B) The effect of tetracycline-induced expression of TCF4 on IgM protein in ABC-like DLBCL cell lines with low TCF4 copy number. The quantification of triplicate experiments is shown in fig. S8. (C) A representative Western blot shows the phosphorylation of downstream kinases from IgM, BTK, and BLNK, with or without tetracycline-induced TCF4 and/or BCR stimulation with an αIgM cross-linking antibody. (D) The quantification of Western blots from triplicate experiments combining TCF4 induction and BCR stimulation, as shown in (C). Each bar represents the mean ± SEM of three independent experiments. Groups were compared by Student’s t test. P > 0.05 is denoted as ns, *P < 0.05, and **P < 0.01.

TCF4 is a dependency in ABC-like DLBCLs with 18q gain

To further validate functional dependencies upon TCF4 in ABC-like DLBCL, we generated two TCF4 dominantnegative (dn) constructs, each encoding a mutant protein capable of sequestering endogenous wild-type TCF4 into DNA-binding incompetent heterodimers. The TCF4ΔBR mutant has a deletion of the basic region (amino acids 567 to 581) before the helix-loop-helix domain that eliminates the DNA binding sequence. The TCF4R582P mutation within the helix-loop-helix domain previously identified in Pitt-Hopkins syndrome (22) and completely abrogates the binding of TCF4 to DNA (Fig. 5A). We expressed these TCF4dn proteins in DLBCL cell lines using a tetracycline-inducible expression construct. In a competition assay with parental cells, the expression of either TCF4ΔBR or TCF4R582P caused a time-dependent relative depletion of ABC-like DLBCL cell lines with TCF4 copy gain (U2932, RIVA, OCI-Ly10, and TMD8) but did not have a marked effect on the relative cell growth of ABC-like cell lines with diploid TCF4 (HBL1 and SUDHL2) or GCB-like cell lines (SUDHL4, KARPAS-422, and OCI-Ly1; Fig. 5B). We performed functional analysis of the effect of TCF4dn at the 48-hour time point before their inducing toxicity. Using ChIP-qPCR, we confirmed that both constructs significantly inhibited binding of TCF4 to the IGHM and MYC enhancer elements in the RIVA and U2932 cell lines (P < 0.01; Fig. 5C). This, in turn, led to a significant reduction in the expression of both IgM and MYC (P < 0.05; Fig. 5, D and E), without a reduction in BRD4 expression (Fig. 5D). This trend was also observed using short-hairpin RNA (shRNA) knockdown of TCF4 (fig. S9). Therefore, IgM and MYC expression is dependent on TCF4 function in ABC-like DLBCLs with TCF4 gain, and elimination of TCF4 activity is associated with a competitive disadvantage to these cells.

Fig. 5 Functional dependency upon TCF4 in ABC-like DLBCL.

(A) A schematic of the two TCF4 dominant-negative (TCF4dn) constructs. The TCF4ΔBR construct has an in-frame deletion of the basic region before the helix-loop-helix domain. The TCF4R582P construct has a single amino acid change within the helix-loop-helix domain. (B) A cell competition assay was performed by mixing equal fractions of green fluorescent protein–positive (GFP+) cells having either the TCF4ΔBR or TCF4R582P tetracycline-inducible dominant-negative construct with the parental cell line. Cells were exposed to doxycycline and the GFP+ fraction measured every 2 to 3 days for 10 days. One-way analysis of variance (ANOVA) comparing ABC-like cell lines with TCF4 DNA copy gain either to ABC-like cell lines with diploid TCF4 (**P < 0.01 and ***P < 0.001) or to GCB-like cell lines (###P < 0.001). Each point represents the mean ± SEM of three independent experiments. (C) TCF4 ChIP-qPCR was performed for the two immunoglobulin enhancers and the MYC enhancer in the presence or absence of dominant-negative constructs in the U2932 and RIVA cell lines with high TCF4 DNA copy number gain and protein expression. Each bar represents the mean ± SEM of three independent experiments, with statistical significance assessed by Student’s t test compared to empty vector (EV) control. **P < 0.01. (D) A representative Western blot shows the expression of the DDK-tagged TCF4dn and the expression of TCF4 target genes IgM and MYC. (E) Quantification of triplicate experiments from (D) shows the attenuation of IgM and MYC expression for both of the unique TCF4dn constructs and in both the U2932 and RIVA ABC-like DLBCL cell lines. Groups were compared by Student’s t test. *P < 0.05 and **P < 0.01.

TCF4 is regulated by BRD4 and can be targeted by BET proteolysis-targeting chimera

The TCF4 gene was one of the most highly BRD4-loaded genes in DLBCL, including in ABC-like DLBCL cell lines with TCF4 copy gain (fig. S10). Furthermore, knockdown of BRD4 in U2932 and RIVA markedly reduced TCF4 expression (fig. S10). We therefore evaluated small-molecule BET inhibitors and a BET protein degrader, ARV771, as potential avenues for reducing TCF4 expression in ABC-like DLBCL cell lines with a high TCF4 copy number. The small-molecule BET inhibitors JQ1 and OTX015 resulted in an up-regulation of BRD4 that was not observed with ARV771, likely because of its role as a substoichiometric BRD4 degrader (Fig. 6A and fig. S11). The increased potency of ARV771 was associated with a greater efficacy in reducing protein expression of the BRD4 target genes MYC and TCF4 (Fig. 6A). In line with previous observations using small-molecule inhibitors (23, 24), ARV771 also induced apoptosis of these cell lines (fig. S12).

Fig. 6 BET proteolysis-targeting chimeras inhibit TCF4 expression.

(A) Protein expression after the treatment of ABC-like DLBCL cell lines with high TCF4 DNA copy number using small-molecule BET inhibitors JQ1 and OTX015. (B) Changes in gene expression induced by the treatment of U2932 and RIVA cell lines with 50 nM of ARV771 for 24 hours. (C) Gene set enrichment analyses are shown for U2932 (green) and RIVA (blue) for the set of genes that were more highly expressed in primary ABC-like tumors with TCF4 DNA copy number gain compared to those tumors without DNA copy number gain, as shown in Fig. 3A. (D) The TCF4 target genes IgM and MYC are reduced by ARV771 treatment but can be partially rescued by enforced expression of DDK-tagged tetracycline-inducible TCF4 (t.o.TCF4). (E and F) Quantification of triplicate experiments for (E) U2932 and (F) RIVA. Each bar represents the mean ± SEM of three independent experiments compared using Student’s t test. *P < 0.05 and **P < 0.01. (G) The effect of rescuing TCF4 expression on the apoptosis induced by ARV771 treatment, as shown by the percentage of annexin-V+ TOPRO-3+ cells with or without TCF4 rescue by doxycycline-induced expression. Bars represent means ± SEM. Student’s t test, *P < 0.05.

Reductions of TCF4 with ARV771 treatment were accompanied by reduced expression of TCF4 target genes, including reductions of IgM at the transcript (Q < 0.1) and protein levels (P < 0.01; Fig. 6, A and B, and table S9). However, a change in expression of the genes that were increased in association with TCF4 DNA copy number gain in primary tumors was not significant (U2932, Q = 0.148; RIVA, Q = 0.201; Fig. 6C). Reductions in TCF4 and IgM expression were not observed after ARV771 treatment in GCB-like DLBCL cell lines (fig. S13). The MYC gene is a direct target of BRD4 (25), and immunoglobulin genes are also regulated by Oct coactivator from B cells (OCA-B), which is reduced by BET inhibition (26). We therefore aimed to determine whether the reduction of MYC and IgM expression and the induction of apoptosis in by ARV-771 in the cell lines with high DNA copy number of TCF4 were directly related to reduction of TCF4 expression. We investigated this with a rescue experiment in which TCF4 expression was maintained in RIVA and U2932 cells during ARV771 treatment using the tetracycline-inducible expression construct. The rescue of TCF4 expression significantly reduced the percentage of apoptotic cells after ARV771 treatment in both cell lines (P < 0.05) but did not eliminate the cytotoxicity of the compound (Fig. 6G). In addition, enforced TCF4 expression during ARV771 treatment significantly rescued the expression of IgM and MYC (P < 0.05; Fig. 6, D to F). Together, these data show that the reduction of TCF4 is one of the mechanisms by which BET inhibition reduces IgM and MYC expression and induces apoptosis in ABC-like DLBCL cells with TCF4 DNA copy number gain.

The promising in vitro activity of ARV771 led us to test whether this compound would be efficacious in vivo. In xenografts of U2932 (Fig. 7, A to E) and RIVA (Fig. 7, F to J) cell lines with high TCF4 expression, we observed that ARV771 was able to reduce tumor growth. At the end of treatment, tumors were significantly smaller in ARV771-treated mice (P < 0.05; Fig. 7, B and G) and showed reduced BRD4, TCF4, IgM, and MYC expression (Fig. 7, C and H), demonstrating that the molecule was able to reach the tumor site at a sufficient concentration to have a functional effect. There were no signs of toxicity in these mice, and treatment with ARV-771 was associated with a significant prolongation of their survival (P < 0.05; Fig. 7, E and J). Our data provide a clear functional rationale for BET inhibition in ABC-like DLBCL and show that ARV771 is effective at eliminating TCF4 and its target genes and treating ABC-like cell lines DLBCL in vitro and in vivo.

Fig. 7 In vivo activity of ARV771 in ABC-like DLBCL.

(A to E) Murine xenografts of the U2932 cell line were allowed to become established and then treated with ARV771 (30 mg/kg) daily, 5 days/week for 3 weeks. (A and B) At the end of treatment, the luminescence was quantified for ARV771-treated and vehicle control–treated mice. (C) A representative tumor shows on-target reduction of BRD4, TCF4, IgM, and MYC expression. (D) Treatment significantly inhibited tumor growth (Mann-Whitney test, ***P < 0.001, **P < 0.01, and *P < 0.05) and (E) led to significantly prolonged survival in ARV771-treated mice despite the short duration of treatment (E, shaded blue). Mantel-Cox log-rank test, P = 0.033. (F to J) Murine xenografts of the RIVA cell line were allowed to become established and then treated with ARV771 (30 mg/kg) daily, 5 days/week for 2 weeks. (F and G) At the end of treatment, the luminescence was quantified in ARV771-treated and vehicle control–treated mice. (H) A representative tumor shows on-target reduction of BRD4, TCF4, IgM, and MYC expression. (I) Treatment significantly inhibited tumor growth (Mann-Whitney test, *P < 0.05) and (J) led to significantly prolonged survival in ARV771-treated mice despite the short duration of treatment (K, shaded blue). Mantel-Cox log-rank test, P = 0.028.

DISCUSSION

The ABC-like subtype is one of two major molecular subtypes of DLBCL that are recognized by the World Health Organization classification (27). These tumors are driven by chronic active BCR signaling emanating from autoreactive IgM localized to the cell surface and intracellular lysosomes (8, 2830). Mutations in CD79B and MYD88 deregulate this signaling through the reduction of LYN-mediated negative feedback and the activation of interleukin-1 receptor–associated kinase signaling, respectively (8, 10). However, recent murine studies have shown that MYD88 mutation alone drove a phenotype reminiscent of peripheral tolerance, and this was only relieved by the combination of MYD88 and CD79B mutations together or by increased expression of surface IgM (31). The ABC-like phenotype is therefore the result of cumulative epistatic genetic alterations, rather than a single dominant driver mutation. In further support of this notion, a recent genomic study defined coassociated sets of genetic alterations that cosegregate with unique genetic subsets of ABC-like DLBCL (32). The “cluster 5” subset of ABC-like DLBCL in this study included frequent MYD88 and CD79B mutations, but the most frequent genetic alteration in this subtype was DNA copy number gain of 18q (32).

We identified the TCF4 (also called E2-2) gene as a key target of 18q DNA copy number gains in DLBCL. The TCF4 gene is closely related to TCF3 (also called E2A), with both genes encoding basic helix-loop-helix transcription factors that form dimers and recognize the E-box consensus sequence CANNTG (33). Murine conditional knockout studies showed that Tcf3 and Tcf4 are critical regulators of germinal center B cell and plasma cell development, in part because of their role in activating immunoglobulin heavy- and light-chain enhancer elements (34, 35). The ID2 and ID3 proteins bind to and inhibit the activity of TCF3 and TCF4 by forming heterodimers that are incapable of binding DNA (33). The TCF3 and ID3 genes are recurrently mutated in another form of B cell lymphoma, BL, with the mutations residing in the interface between TCF3 and ID3, thereby preventing their interaction (18, 19). We observed that TCF4 DNA copy number gains are also frequent in BL and that they mutually exclude TCF3 and ID3 mutations, providing further evidence for the importance of TCF3/TCF4 deregulation in this disease. In contrast to BL, TCF3 and ID3 mutations were rare in ABC-like DLBCL, but TCF4 DNA copy number gains were present at more than twice the frequency. The higher mutation rate for ID3 and TCF3 in BL may be due to a higher expression of these genes in BL and an associated greater reliance on TCF3 activity for survival. However, as TCF4 was also variably expressed in these tumors and ID3 mutations may presumably also reduce the interaction between ID3 and TCF4, the propensity for ID3 to mutate at a much higher frequency in BL compared TCF3 may be associated with the ability of these mutations to deregulate both TCF3 and TCF4 activity. On the basis of their different expression patterns during B cell development and phenotypes from single-gene conditional knockout studies (34, 35), we speculate that TCF3 and TCF4 may have some nonredundant roles, with TCF3 potentially being more important in germinal center B cells and TCF4 playing a more prominent role in post-germinal center B cells. This may explain the propensity for TCF4 deregulation in ABC-like DLBCL, which aligns with activated B cells that are primed for germinal center exit.

We observed a marked up-regulation of IgM transcript expression in primary tumors with TCF4 DNA copy number gain. We also identified binding sites for TCF4 in the immunoglobulin heavy-chain locus and showed that experimental modulation of TCF4 abundance or activity was sufficient to control IgM expression in ABC-like DLBCL cell lines. Together, this provides strong evidence for a functional role of TCF4 in promoting IgM expression in ABC-like DLBCL. This is particularly important in this disease because >90% of ABC-like DLBCL cases express IgM and the disease etiology centers on pathogenic signaling downstream of this receptor (7, 8). TCF4 was more highly expressed in ABC-like DLBCL compared to GCB-like DLBCL, even in cases without DNA copy number gain of the locus. This suggests that this axis may be active in all ABC-like DLBCLs and further enhanced in the ~40% that harbor 18q DNA copy number gains. This is akin to the role of EZH2 in GCB-like DLBCL, which promotes the survival and proliferation of all germinal center B cells but has enhanced activity in the context of activating somatic mutations (36, 37). We therefore hypothesize that TCF4 may participate in a critical functional axis of immunoglobulin regulation in all ABC-like DLBCL.

Proteins in the BET family, including BRD4, are attractive therapeutic targets in cancer because of their role in the transcriptional activation of oncogenes such as MYC (25, 38). In DLBCL, BRD4 targets include key transcription factors such as BCL6, PAX5, and IRF4 (26). Furthermore, previous studies have shown that BET inhibition is preferentially cytotoxic to ABC-like compared to GCB-like DLBCL cell lines (23, 24) and reduces nuclear factor κB signaling (23). However, the mechanism underlying these observations has not been defined. We have highlighted TCF4 as another prominent target of BRD4 in DLBCL, as has been previously described in plasmacytoid dendritic cell neoplasms (21). Because of the difficulty in directly drugging transcription factors, BET inhibition represents a logical avenue for reducing TCF4 expression in ABC-like DLBCL. However, small-molecule inhibitors have also been shown to result in the up-regulation of BRD4 expression (39). We therefore evaluated a BET protein proteolysis-targeting chimera (PROTAC), ARV771, which degrades BET proteins in a substoichiometric fashion by combining a BET-targeting warhead from OTX015 with a moiety that recruits the von Hippel-Lindau (VHL) ubiquitin ligase (39). The inhibition of TCF4 expression by BET PROTAC led to the coordinate down-regulation of TCF4 target genes that were highly expressed in primary tumors with TCF4 DNA copy gain, such as IgM and MYC, which could be rescued by enforced expression of TCF4. This was an unexpected observation for the MYC gene, which is also a direct BRD4 target (25). Thus, although the effects of BET inhibition are multifaceted, our data show that a portion of the broad transcriptional changes and toxicity mediated by BET inhibition in ABC-like DLBCLs with 18q DNA copy number gains can be attributed to the downstream effects of reduced TCF4 expression.

Together, our data highlight TCF4 DNA copy gains as a functional rationale for BET inhibition in ABC-like DLBCL and show that the BET PROTAC ARV771 is able to inhibit TCF4 expression and activity. However, overexpression of BCL2 has been described as a resistance mechanism for BET inhibitors (40), and we observed that most of the 18q DNA copy number gains in DLBCL encompass both TCF4 and BCL2. We therefore posit that the promising activity of BET inhibitors in ABC-like DLBCL may be enhanced by combination with a BCL2 inhibitor such as venetoclax. In support of this, BET inhibitors have been shown to act synergistically with venetoclax in myeloid leukemia (40) and in another form of B cell lymphoma, mantle cell lymphoma (41). Combination of BET and BCL2 inhibition therefore represents an attractive therapeutic avenue for future investigation in ABC-like DLBCL.

There are some limitations to this study that should be addressed in future research. We showed TCF4 expression at the transcript level in primary tumor samples but were unable to perform immunohistochemical staining for TCF4 on these tumors because of the poor performance of available monoclonal antibodies. Prospective validation of TCF4 protein expression and its association with clinical outcome should therefore be performed in a large series of primary DLBCL tumor samples. In addition, we showed a clear functional rationale for targeting TCF4 using BET inhibitors and showed that this was efficacious in in vitro and using cell line xenografts. However, as xenografts do not predict clinical outcomes, the association between TCF4 DNA copy number status or cell of origin subtype and response to BET inhibitors would be best evaluated as a correlate in future clinical trials. Because of the potential toxicities associated with BET inhibition, it would also be desirable to identify a more direct avenue for specifically targeting TCF4.

MATERIALS AND METHODS

Study design

The objective of this work was to define the genetic etiology of the ABC-like subtype of DLBCL. We used integrative genomic analysis of primary DLBCL tumors from 1000 patients, including 140 fresh-frozen DLBCL tumors obtained from the University of Nebraska Medical Center (UNMC) lymphoma tissue bank (Institutional Review Board 161-95-EP). In vitro analysis of TCF4 function using tetracycline-inducible expression of TCF4 or TCF4dn constructs in ABC-like DLBCL cell lines. All animal studies were performed under a protocol approved by the Institutional Animal Care and Use Committee (IACUC) at MD Anderson Cancer Center, an Association for Assessment and Accreditation of Laboratory Animal Care–accredited institution. BET inhibitors were tested in ABC-like DLBCL cell lines in vitro and in vivo using murine xenografts as a therapeutic avenue to directly target TCF4 expression.

DNA copy number data acquisition and processing

Publicly available data for SNP microarrays and array comparative genome hybridization platforms with more than 200,000 markers were downloaded from the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/; table S1) (15, 17, 4248). Data for all arrays were represented as log2 copy number change and segmented using the circular binary segmentation tool on GenePattern (49). We identified peaks of significant DNA copy number loss and gain using GISTIC2.0 (14). The thresholds used for DNA copy number gain and loss were 0.2 copies over a region encompassing 100 markers. Peaks with a residual Q value of <0.1 were considered significant.

Cell of origin subtyping and integrative analysis of gene expression profiling data

Raw CEL files for matched Affymetrix U133 Plus 2.0 gene expression microarray data were obtained for 249 previously published DLBCLs [GSE11318 (17) and GSE34171 (15)] and an additional 98 tumors from UNMC. Data were normalized using Robust Multi-array Average in the ExpressionFileCreator module of GenePattern (49). The GSE11318 and GSE34171 datasets were batch-corrected and combined using ComBat (50). The additional UNMC gene expression dataset (GSE10846) could not be batch-corrected and was analyzed separately only for cell of origin subtype. Cell of origin subtype was determined using a previously described Bayesian classifier (51) and validated using available outcome data (fig. S1 and table S1). For 108 cases, the previously reported Nanostring-derived cell of origin classification was used (44).

The targets of DNA CNAs were determined using integrative analysis, as previously described (52). Briefly, differential gene expression analysis was performed between tumors (n = 249) with or without each lesion, limited to the set of genes within the GISTIC-defined peak (table S6). To identify the signature associated with 18q21 DNA copy number gain, we performed differential gene expression analysis within ABC-like DLBCL cases (n = 110). Genes with a fold change of ≥1.2 in the direction of the CNA and a false discovery rate (FDR) Q value of <0.1 were considered statistically significant. Associations between CNAs and cell of origin subtype were determined using a Fisher exact test.

Survival and cell of origin subtype association

Overall survival data were collated from previous studies for 232 patients with DLBCL treated with CHOP combination chemotherapy (15, 17) and 240 patients with DLBCL treated with R-CHOP (15, 53) using a log-rank test. Associations between DNA CNAs or mutations and cell of origin subtype were determined using a Fisher’s exact test, corrected for multiple hypothesis testing using the qvalue R package. The REVEALER tool (16) was implemented in GenePattern using a set of 98 tumors for which gene expression microarray and targeted NGS data were available (table S1). We selected the seed feature as MYD88 mutation because of its association with the ABC-like DLBCL subtype (fig. S2). Additional features included all statistically significant DNA CNAs (n = 24) and recurrently mutated genes (n = 63) that were present in 5% or more of the 98 tumors. A total of 10 iterations of feature selection were performed using the ABC probability score as the phenotype and the ABC-like subtype as the class.

Cloning and tetracycline inducible expression of TCF4

The TCF4 gene was cloned into sleeping beauty vector [pSBtet-GP, Addgene (54)] by replacing luciferase gene with TCF4. Tetracycline-based TCF4-expressing DLBCL cell lines (HBL1, TMD8, and SUDHL2) were generated by cotransfecting transposase-expressing vector [pCMV(CAT)T7-SB100, Addgene (55)] and sleeping beauty vector expressing TCF4 using neon transfection system, according to the manufacturer’s instructions (Neon transfection system, Invitrogen). Transfected cells were selected for stable cell line generation with puromycin (1 μg/ml) for 1 week and maintained in 10% tetracycline-negative fetal bovine serum (Corning) containing RPMI media. The dose of tetracycline required for physiologically relevant protein expression of TCF4 was determined by a dose titration relative to the expression in the U2932 cell line (fig. S3). All experiments were performed with 24 hours of tetracycline treatment at the doses specified in each experiment.

BET inhibitors and treatments

BET inhibitors JQ1 and OTX015 were obtained from Selleck Chemicals. BET-PROTAC (ARV771) was provided by Arvinas Inc. U2932, and RIVA cell lines were treated with indicated concentrations of BET inhibitors (JQ1 and OTX015) or BRD4-PROTAC (ARV771) for 24 hours before immunoblotting. Apoptosis was interrogated by annexin-V/TOPRO-3 staining (Thermo Fisher Scientific) and analyzed using flow cytometry (BD LSRFortessa) and FlowJo software. For gene expression analysis, cell lines (U2932 and RIVA) were untreated or treated with ARV771 (50 ng/ml) for 24 hours. RNA-sequencing (RNA-seq) libraries were constructed using the KAPA RNA HyperPrep Kit with RiboErase and sequenced on a HiSeq 4000 instrument at the MD Anderson Sequencing and Microarray Core Facility. Fastq files were aligned to the GRCh37 assembly using STAR 2.6.0c, preprocessed with RSEM, and quantified with rsem-calculate-expression. DESeq2 (v1.18.1) was used to identify differentially expressed genes using a two-variable (cell line and treatment) analysis with default settings. Gene set enrichment analysis (56) was performed using GenePattern and a list of all genes from RNA-seq ranked by the fold change in expression after ARV771 treatment. The gene set consisted of all genes that showed higher expression in ABC-like DLBCL tumors with TCF4 DNA copy number gain compared to those without (table S9).

In vivo studies

Five million luciferized RIVA or U2932 cells (mixed with Matrigel at a volume ratio of 1:1) were subcutaneously injected in the left flank of male athymic nude mice (nu/nu) (n = 8 per group). Tumor volume was calculated by the ½(length × width2) method. Treatment was initiated when the mean tumor volumes reached ∼150 mm3. Mice were treated with vehicle [10% (1:1 solutol/ethanol) and 90% D5-water, subcutaneously daily, 5 days/week] or ARV-771 (30 mg/kg, subcutaneously, daily, 5 days/week). The RIVA mouse model was treated for 2 weeks. Because of slower tumor growth, the U2932 mouse model was treated for 3 weeks. For bioluminescent imaging, mice were intraperitoneally injected with 100 μl of d-luciferin potassium salt (75 mg/kg; Gold Biotechnology) (reconstituted in 1× phosphate-buffered saline and sterile-filtered through a 0.2-μm filter) incubated for 5 min, anesthetized with isoflurane, and imaged once per week using a Xenogen IVIS-200 imaging system (PerkinElmer) to monitor disease status and treatment efficacy. One mouse from each cohort was euthanized after 3 weeks of treatment for biomarker analysis. Mice bearing tumors greater than 1500 mm3 were removed from study and humanely euthanized (carbon dioxide inhalation and cervical dislocation) according to the IACUC-approved protocol. Veterinarians and veterinary staff assisting in determining when euthanasia was required were blinded to the experimental conditions of the study. The survival of the mice is represented by a Kaplan-Meier plot. For immunoblotting, tumors from respective vehicle and ARV771-treated xenografts were extracted. Tumor lysates were prepared using radioimmunoprecipitation assay lysis buffer and probed for indicated primary antibodies.

Statistical analysis

For the genomic studies, we present the significance of tests as a FDR Q value after correction for multiple hypothesis testing using a Benjamini-Hochberg correction. In addition, the means and SDs for each group from differential gene expression profiling analyses are presented in the supplementary tables. Categorical variables were tested using a Fisher’s exact test. Continuous variables were tested for normality using a Shapiro-Wilk test. Normally distributed variables were compared with an unpaired two-tailed Student’s t test, with multiple hypothesis testing correction when appropriate. Variables that did not fit a normal distribution were tested using a Mann-Whitney rank sum test. Tumor size was compared among cohorts by unpaired Student’s t test. Relative GFP percentage in cell competition assays was compared using one-way ANOVA. Mouse survival is shown by a Kaplan-Meier plot. Differences in survival were calculated by a Mantel-Cox log-rank test. P values less than 0.05 and Q values less than 0.1 were considered significant.

SUPPLEMENTARY MATERIALS

stm.sciencemag.org/cgi/content/full/11/497/eaav5599/DC1

Fig. S1. Validation of cell of origin subtyping.

Fig. S2. Comparison of outcome in CHOP-treated patients with DLBCL stratified by cell of origin subtype and 18q21 gain.

Fig. S3. Recurrently mutated genes enriched in the ABC-like DLBCL subtype.

Fig. S4. Expression patterns of TCF3 and TCF4 in normal and malignant B cells.

Fig. S5. Tetracycline induction of TCF4 expression.

Fig. S6. ChIP-seq peaks for TCF4.

Fig. S7. Cellular localization of TCF4-induced IgM.

Fig. S8. Quantification of TCF4-induced MYC and BCL2 protein expression.

Fig. S9. shRNA-mediated knockdown of TCF4.

Fig. S10. BRD4 regulates TCF4 in DLBCL cell lines.

Fig. S11. Western blot quantification of triplicate BET inhibitor experiments.

Fig. S12. ARV771 induces apoptosis in RIVA and U2932 cell lines.

Fig. S13. Effect of ARV-771 on IgM expression in GCB-like DLBCL cell lines.

Data file S1 contains the following supplementary tables:

Table S1. Genomic and clinical data from DLBCL tumors included in this study.

Table S2. NGS statistics.

Table S3. Genes in GISTIC peaks.

Table S4. Cell of origin association of GISTIC peaks.

Table S5. Cell of origin association of recurrently mutated genes.

Table S6. Integrative analysis of DNA CNAs.

Table S7. Differential gene expression analysis of ABC-like tumors with or without TCF4 copy gain.

Table S8. ChIP-seq peaks for TCF4 signature genes.

Table S9. Differentially expressed genes following ARV-771 treatment.

Table S10. Primer sequences.

References (5763)

REFERENCES AND NOTES

Acknowledgments: Arvinas Inc. provided ARV771 for the studies. pSBtet-GP was a gift from E. Kowarz. pCMV(CAT)T7-SB100 was a gift from Z. Izsvak. pHIV-Luc-ZsGreen was a gift from B. Welm. Funding: This research was supported by the Nebraska Department of Health and Human Services (LB506 2016-16 to M.R.G.), the Schweitzer Family Fund (to J.W., R.E.D., and M.R.G.), RO1 CA210250 (to K.B.), R01 CA201380 (to M.R.G.), the Fred & Pamela Buffet Cancer Center Support Grant (P30 CA036727), and the MD Anderson Cancer Center NCI CORE Grant (P30 CA016672). Author contributions: N.J., K.H., S.T., W.F., O.H., and D.K. performed the experiments. N.J., K.H., S.T., K.B., and M.R.G. analyzed data and wrote the manuscript. M.C.J.M., A.B., T.H., Q.D., D.M., C.P., C.L.L., A.J.G., S.R., J.I., J.W., S.S.N., and R.E.D. analyzed or interpreted the data. A.A.A. provided computational resources. E.H., R.K., K.E.S., G.J., R.R., F.G., R.D.G., A.R., J.M.V., M.A.L., and T.G. provided samples or data. M.R.G. conceived and supervised the study. All authors read and approved the manuscript. Competing interests: The authors declare that no competing interests are related to the work in this manuscript. M.R.G. is a consultant Verastem Oncology. S.S.N. is a consultant or advisory board member for Kite/Gilead, Merck, Celgene, Novartis, Unum Therapeutics, Precision Biosciences, Cell Medica, and Incyte. J.M.V. is a consultant for Novartis, AbbVie, Epizyme, Roche, Legend Pharmaceuticals, Kyopharm, Sandoz, Vaniam Group, Janssen/Pharmacyclics, Kite/Gilead, Acerta/AstraZenica, Nordic Nanovector, and Verastem Oncology. R.D.G. is a consultant for Genetech. A.J.G. is a consultant for CiberMed Inc. K.E.S. is a consultant for Celgene and Roche. A.A.A. is a consultant on advisory board member for Celgene, Roche/Genetech, and Gilead Sciences. Data and materials availability: All data associated with this study are present in the paper or the Supplementary Materials. The ChIP-seq and RNA-seq data produced in this study are available in the Gene Expression Omnibus (www.ncbi.nlm.nih.gov/geo/) under accession no. GSE119477. The SNP and gene expression microarray accessions for the previously published data used in this work are listed in table S1. All supplementary tables are in data file S1.
View Abstract

Stay Connected to Science Translational Medicine

Navigate This Article