Research ArticleCancer

CD40 Pathway Activation Status Predicts Response to CD40 Therapy in Diffuse Large B Cell Lymphoma

See allHide authors and affiliations

Science Translational Medicine  16 Mar 2011:
Vol. 3, Issue 74, pp. 74ra22
DOI: 10.1126/scitranslmed.3001620

Abstract

The primary function of B cells, critical components of the adaptive immune response, is to produce antibodies against foreign antigens, as well as to perform isotype class switching, which changes the heavy chain of an antibody so that it can interact with different repertoires of effector cells. CD40 is a member of the tumor necrosis factor superfamily of cell surface receptors that transmits survival signals to B cells. In contrast, in B cell cancers, stimulation of CD40 signaling results in a heterogeneous response in which cells can sometimes undergo cell death in response to treatment, depending on the system studied. We found an association between sensitivity to CD40 stimulation and mutation of the tumor suppressor p53 in a panel of non-Hodgkin’s lymphoma cell lines. Consistent with p53’s tumor suppressor role, we found that higher levels of intrinsic DNA damage and increased proliferation rates, as well as higher levels of BCL6, a transcriptional repressor proto-oncogene, were associated with sensitivity to CD40 stimulation. In addition, CD40 treatment–resistant cell lines were sensitized to CD40 stimulation after the introduction of DNA-damaging agents. Using gene expression analysis, we also showed that resistant cell lines exhibited a preexisting activated CD40 pathway and that an mRNA expression signature comprising CD40 target genes predicted sensitivity and resistance to CD40-activating agents in cell lines and mouse xenograft models. Finally, the gene signature predicted tumor shrinkage and progression-free survival in patients with diffuse large B cell lymphoma treated with dacetuzumab, a monoclonal antibody with partial CD40 agonist activity. These data show that CD40 pathway activation status may be useful in predicting the antitumor activity of CD40-stimulating therapeutic drugs.

Introduction

CD40 is a member of the tumor necrosis factor receptor (TNFR) superfamily that is expressed on the surface of B cells and has a critical role, via interaction with the CD40 ligand (CD40L), in T cell–dependent immune responses. Mutations in CD40L result in X-linked hyper-IgM (immunoglobulin M) syndrome (1, 2), an inherited disease in which males are susceptible to infections as a result of defects in CD40L and consequent low IgA, IgE, and IgG. Mice deficient in CD40 have defects in affinity maturation of antibodies, germinal center formation, and memory B cell development (3). The role of CD40 in B cell cancers has been enigmatic. Stimulation of the CD40 pathway promotes apoptosis of certain B cell lines in vitro and in xenograft tumor models (4, 5). In contrast, CD40 stimulation can cause proliferation of non-Hodgkin’s lymphoma (NHL) cells (69) and promote lymphomagenesis (10) and the transformation of primary B cells (11). The molecular mechanisms that account for these contrasting phenotypes are unknown.

BCL6 is a transcriptional repressor proto-oncogene that is expressed in germinal center cells and controls a plethora of genes involved in cell cycle arrest, apoptosis, differentiation, cell activation, and the DNA damage response (12). Translocations and mutations of the 5′ noncoding region that result in deregulated BCL6 expression have been identified in diffuse large B cell lymphoma (DLBCL) and follicular lymphoma (1316). Transgenic mice engineered to overexpress BCL6 in B cells develop lymphomas that resemble DLBCL (17). The oncogenic effects of BCL6 can be partially attributed to repression of the DNA damage response machinery. A key DNA damage sensor, ATR (ataxia-telangiectasia–mutated and Rad3-related), is transcriptionally repressed by BCL6 (18). The repression of ATR permits the survival of the germinal center B cells, centroblasts, in response to rapid proliferation and activation-induced cytosine deaminase (AID)–induced DNA damage during the germinal center reaction. This ensures that B cells with the highest affinity for antigen are selected and form memory B or plasma cells. Stimulation of CD40 by CD40L can reverse this response by down-regulating BCL6, promoting cell death of heavily predamaged centroblasts (19).

To define the biological pathways and molecular factors that determine response to stimulation of CD40, we used a panel of NHL cell lines that are primarily of DLBCL origin and harbor many of the genetic defects observed in human lymphoma—somatic mutations, amplifications, deletions, and translocations (20)—and patient samples.

Results

Baseline molecular characteristics that associate with a reduction in cell viability caused by CD40 stimulation

We first assessed the ability of dacetuzumab (SGN-40), a CD40-stimulatory IgG1 monoclonal antibody that is under clinical development for B cell cancers (21), or an isotype control antibody to reduce cell viability in a panel of NHL cell lines. After determining the specific reduction in cell viability by SGN-40 compared to isotype control after 4 days, we classified the cell lines as sensitive or resistant, with sensitive defined as a reduction in viability of at least 25% (IC25) (Fig. 1A). Cell lines that did not achieve a reduction of 25% at the maximum concentration were classified as resistant (IC25 = 25 μg/ml). After stimulation of the CD40 pathway with SGN-40, 17 of 30 cell lines (57%) were classified as sensitive, and the remainder (13 of 30, 43%) as resistant. Similar cellular sensitivity phenotypes were observed using soluble CD40L to stimulate CD40 (fig. S1). Furthermore, at least half of the resistant cell lines showed increased viability with the addition of SGN-40 (fig. S2).

Fig. 1

Identifying the biological parameters that associate with response to CD40 stimulation in NHL cell lines. (A) Sensitivity of NHL cell lines after stimulation of the CD40 pathway. Thirty NHL cell lines were treated with SGN-40, and viability was assessed after 96 hours. Cell lines were classified as sensitive or resistant on the basis of IC25 values. Mean and SD values are from three independent experiments. (B) p53 mutations are enriched in cell lines sensitive to SGN-40. Statistical significance of mutations and sensitivity to SGN-40 was assessed by the Wilcoxon test. (C) Sensitive cell lines show more intrinsic DNA damage. Data from flow cytometry are expressed as mean fluorescence intensity (MFI). Sensitive versus resistant P value was calculated by the Wilcoxon test. (D) Chemotherapeutic agents cause CD40 stimulation–resistant cell lines to become sensitive. A3/Kawakami cells were treated as indicated for 96 hours, and percent cell viability was determined. Means and SDs are from four independent experiments. (E) Sensitive cell lines have a shorter doubling time than resistant cell lines. A parametric two-sided Student’s t test was used to assess significance. (F) BCL6 protein levels are lower in resistant cell lines. Data represent quantitation of Western blots by densitometry. Lower horizontal bars indicate sensitivity to SGN-40. Statistical significance was assessed by the Wilcoxon test.

To identify factors associated with the response to SGN-40, we sequenced the cell lines for mutations in 21 key oncogenes and tumor suppressors as described (22) and calculated the statistical association with sensitivity (Fig. 1B, fig. S3, and table S1). The only gene tested that displayed a significant association with SGN-40 sensitivity was the tumor suppressor p53. Cell lines susceptible to CD40 stimulation predominantly harbored mutations in p53 (12 of 17), whereas cell lines that were resistant to CD40 stimulation predominantly had a wild-type p53 genotype (10 of 13) (P = 0.0203).

Because mutation of p53 is a relatively common genetic lesion in DLBCL (23) and somatic mutation and/or loss of p53 can result in excessive DNA damage and genomic instability (24, 25), we next determined whether intrinsic DNA damage was associated with response to CD40 stimulation. Cell lines were assessed by flow cytometry for the DNA damage marker pH2AX in the absence of SGN-40 treatment, and we found that cell lines sensitive to CD40 stimulation on average displayed more intrinsic DNA damage than did the resistant cell lines (Fig. 1C, P < 0.0001). To test whether intrinsic damage was indeed important for response to CD40 stimulation, we treated the SGN-40–resistant cell line A3/Kawakami, which has low levels of pH2AX [mean fluorescence intensity (MFI) = 1.264], with the DNA-damaging agents gemcitabine or cisplatin and SGN-40 in a dose matrix and assessed cell viability (Fig. 1D). Addition of either DNA-damaging agent sensitized cells to SGN-40. We saw the same pattern in four additional SGN-40–resistant cell lines treated with SGN-40 and gemcitabine or cisplatin (figs. S4 and S5). Together, these data suggest that the presence of DNA damage is an important determinant of the antitumor response to CD40 simulation.

Given that DNA damage can also result from hyperproliferation in centroblasts as a result of BCL6 repression of p21 transcription via Miz-1 (26), we tested whether there was an association of the response to CD40 stimulation and the doubling time of the NHL cell lines (Fig. 1E, P = 0.04). Indeed, sensitive cell lines had a shorter mean doubling time than did resistant cells lines. In addition, BCL6 protein levels by Western blotting were higher in the sensitive cell lines, with seven of nine having more BCL6; only two of the nine resistant cell lines had expression levels of BCL6 similar to those of sensitive cell lines (Fig. 1F, P = 0.05).

Sensitivity to CD40 stimulation in vitro is associated with an active CD40 pathway

The lower levels of BCL6 protein in resistant cell lines than in sensitive cell lines suggested that there might be fundamental differences in the CD40 signaling pathway between the cells with the two response phenotypes because CD40 stimulation can down-regulate BCL6 (19). We explored the possibility that CD40 pathway activation status might play a role in determining response. Gene signatures of pathway activation can predict specific aberrations in pathways and sensitivity to specific targeted therapies such as inhibitors of RAS or MEK (mitogen-activated or extracellular signal–regulated protein kinase kinase) (27, 28). Furthermore, activation of the CD40 pathway results in a downstream signaling cascade and transcriptional programs (29, 30). To generate a comprehensive CD40 gene-pathway activation signature (gPAS), we treated seven sensitive cell lines across the sensitivity spectrum to capture heterogeneity [four highly sensitive (<0.1 μg/ml) and three of intermediate sensitivity (<0.7 μg/ml)] with SGN-40 or an isotype control antibody for 24 hours and isolated RNA for gene expression microarray analysis (Fig. 2A). After filtering out genes for low intensity values or low variation across samples, we selected the top 50 up- and down-regulated genes (figs. S6 and S7) to generate a CD40 gPAS score. All of the NHL cell lines tested were then assigned a CD40 gPAS score before SGN-40 treatment to determine whether there was a relationship between CD40 sensitivity or resistance and baseline CD40 pathway activation. This analysis identified an association between resistance to SGN-40 and a higher baseline CD40 pathway activation, whereas cell lines that were sensitive to SGN-40 had a baseline gene signature representative of an inactive CD40 pathway (Fig. 2A, P = 0.023). We confirmed these results in an independent set of 12 cell lines (Fig. 2B, P = 0.018). We obtained similar results by assessing a group of genes previously reported to be regulated by CD40L in Ramos cells cocultured with fibroblasts overexpressing CD40L (fig. S8, CD40 Basso) (30). Because CD40 can also activate downstream nuclear factor κB (NFκB) signaling, we assessed whether the CD40 gPAS is associated with (i) an active NFκB pathway as assessed by pIKBa Western blotting or NFκB gene signature or (ii) mutations of key signaling components of the NFκB pathway such as CARD11 and A20 (20, 31). No statistically significant correlations were observed with any of the covariates between CD40 gPAS or sensitivity to SGN-40 stimulation (figs. S9 and S10 and table S1), suggesting that CD40 gPAS is specific to CD40 signaling and not a general B cell activation or NFκB signature.

Fig. 2

Relationship between CD40 pathway activation and resistance to CD40 stimulation. (A) Procedures for generating CD40 gene-pathway activation signature (gPAS) and correlation with resistance to CD40 stimulation. (Upper right panel) Top 50 up-regulated and top 50 down-regulated gene expression in all cell lines, with lower panel designating sensitivity phenotype. (Lower right panel) CD40 gPAS score in all cell lines. Red and green in heat maps represent high and low, respectively, expression of gene or signature score and represent number of SDs away from the mean value. Black rectangles indicate sensitivity to CD40 stimulation or classified as CD40 gPAS–positive. Statistical significance was assessed by Fisher’s exact test. (B) CD40 gPAS status correlates with resistance in an independent set of 12 NHL cell lines. Black rectangles indicate sensitivity to CD40 stimulation or classified as CD40 gPAS–positive. Statistical significance was assessed by Fisher’s exact test. (C) The GCB subset of DLBCL cell lines is sensitive to stimulation of the CD40 pathway. Subtype classification was determined by gene expression. Red and black colors represent sensitive and resistant, respectively, to CD40 stimulation. Statistical significance was assessed between GCB and ABC by Fisher’s exact test.

Upon engagement of the CD40 receptor, a signal transduction cascade is triggered that includes the activation of the phosphatidylinositol 3-kinase (PI3K) pathway and phosphorylation of AKT, and activation of the MKK (mitogen-activated protein kinase kinase) cascade and phosphorylation of c-Jun N-terminal kinase (JNK), p38, and extracellular signal–regulated kinase (ERK) (32, 33). Because the CD40 gPAS suggested that the CD40 pathway is switched on in resistant cell lines, we determined whether baseline phosphorylation of any of these key signaling nodes differed between sensitive and resistant cell lines using quantitative Western blotting normalized by total AKT, JNK, ERK, or p38 (fig. S9). Constitutive phospho-p38 levels were significantly higher in the sensitive cell lines (P = 0.003) but not phospho-AKT or phospho-JNK. In contrast, phospho-ERK was detected at higher levels in the resistant cell lines (P = 0.023), which is consistent with results from a mouse model of constitutively active CD40 signaling (10).

There are three major molecular subtypes of DLBCL as determined by gene expression profiling: germinal center B cell–like (GCB), activated B cell–like (ABC), and those that are unclassified, each of which represents different stages of B cell differentiation and displays distinct clinical behaviors (34, 35). The DLBCL cell lines were categorized according to the subtype that their gene expression profiling most closely resembled at baseline (table S1), and sensitivity to CD40 stimulation was assessed between each subtype (Fig. 2C). Cell lines from each of the three subtypes were susceptible to CD40 stimulation; however, we noted a significant enrichment of CD40 activity in the GCB-like subtype, with 12 of 15 GCB-like cell lines showing sensitivity (GCB versus ABC, P = 0.006). In contrast, the unclassified and ABC groups displayed sensitivity to CD40 stimulation in only one of five and one of seven cell lines, respectively.

A 15-gene signature derived from in vitro sensitivity to CD40 stimulation

Our data to this point suggested that CD40 pathway activation status, BCL6 biology, and the GCB/ABC classification were all important parameters defining the response to CD40 stimulation. We set out to generate a molecular signature that could capture this information and then tested this signature in a clinical trial setting. Before the initiation of therapy, lymphoma patients routinely undergo a diagnostic tissue biopsy, which is preserved by formalin fixation and paraffin embedding (FFPE). RNA extracted from FFPE tissue is often degraded and of poor quality as a result of extensive cross-links from formalin fixation (3638). Because quantitative reverse transcription–polymerase chain reaction (qRT-PCR) assays can be used in fixed tissue (39, 40), we developed a method to select genes that can predict sensitivity to SGN-40 (Fig. 3A). A list of genes that were significantly correlated or anticorrelated with the IC25 sensitivity values was generated and then selected if they were directly regulated by CD40 stimulation (BTG2, CD22, CD40, CD44, CD79b, CTSC, LMO2, PUS7, UAP1, and VNN2), part of the CD40 pathway network (EPDR1, IGF1R, and RGS13), or a component of the GCB/ABC classifier (BCL6, CD44, and LMO2) (fig. S11). The stepwise procedure initially selected 14 genes, and BCL6 was added at the last step of the procedure because of our experimental observations (Fig. 1F). This group of 15 genes was then used to classify the training set of 30 cell lines with the K-nearest neighbor (KNN) algorithm. The signature predicted susceptibility to SGN-40 treatment in 23 of 30 (77% by 10-fold cross-validation) cell lines in this training set (Fig. 3B), and samples with a cross-validated prediction of sensitivity had a significantly lower mean IC25 (P = 0.01, Wilcoxon test).

Fig. 3

CD40 gene signature selection, training, and testing on NHL cell lines. (A) Schematic of gene feature selection of gene signature. (B) Training set of NHL cell lines for gene signature and predicted classification. Heat map represents sensitivity to CD40 stimulation and predicted classification with KNN. Red and black represents sensitive and resistant, respectively, to CD40 stimulation. (C) The 15-gene signature can predict sensitivity to CD40 stimulation in an independent set of NHL cell lines. Heat map representation as in (B).

To test the ability of the gene signature to predict sensitivity in an independent cohort of cell lines, we determined the susceptibility of 12 additional NHL cell lines to CD40 stimulation in cell viability assays. Gene expression microarrays were performed, and sensitivity to CD40 stimulation was predicted with the gene signature (Fig. 3C). Of the 12 cell lines tested, 8 were sensitive and 4 were resistant to CD40 stimulation. The signature predicted that 9 of the 12 cell lines would be sensitive, and 3 of the 12 cell lines would be resistant. Overall, the signature was able to predict susceptibility to SGN-40 treatment in 9 of 12 (75%) in this independent set of cell lines.

We next studied four NHL xenograft mouse models and assessed their susceptibility to inhibition of growth in vivo after administration of SGN-40 (Fig. 4, A to D). Gene expression profiling of the implanted tumor before SGN-40 treatment was performed, and CD40 surface expression on the tumor cells was confirmed with flow cytometry (Fig. 4E). The gene signature predicted that all four xenograft models would be sensitive to SGN-40 (Fig. 4F). Intraperitoneal administration of SGN-40 resulted in a statistically significant reduction in the tumor volume of the implanted U-698-M, Ramos-RA1, and Farage-X1 cell lines (Fig. 4, A, B, and D, respectively). In contrast, the SU-DHL4-Luc model (Fig. 4C) failed to show a statistically significant reduction in tumor volume relative to the mock-treated xenografts.

Fig. 4

Prediction of response to CD40 stimulation in xenograft mouse models. (A to D) Efficacy assessment of CD40 stimulation on xenograft tumor models: (A) U-698-M, (B) Ramos-RA1, (C) SU-DHL-4-luc, and (D) Farage-X1. Two-sided Student’s t test was used to assess statistical significance between treatment arms. IP, intraperitoneal. (E) CD40 expressed on the surface of xenograft tumor cells. Data are expressed as percent maximum signal by flow cytometry. The blue peak corresponds to an isotype-matched control, and the red peak corresponds to CD40. (F) Actual CD40 stimulation response by xenograft tumors as determined by experiment and gene signature.

We next developed qRT-PCR assays for each of the 15 genes in the signature for testing of archival-fixed clinical tissue. Specific probes for each gene were designed and tested on cell lines that were formalin-fixed and paraffin-embedded to mimic the material available from clinical trials, and performance was assessed (fig. S12). The qRT-PCR gene signature gave identical sensitivity predictions to those determined by microarrays and was subsequently taken forward to test on SGN-40 clinical trial samples.

Prediction of response to CD40 stimulation in patients with DLBCL by CD40 pathway activation status

In recent Phase 1 (41) and Phase 2 SGN-40 monotherapy clinical trials, 21 of 57 (37%) patients with DLBCL that had relapsed after multiple therapies displayed tumor shrinkage in response to SGN-40 treatment. RNA was extracted from fixed tumor tissue harvested before SGN-40 treatment from 39 patients and profiled with the qRT-PCR gene signature (see Materials and Methods). The clinical characteristics of these patients are described in fig. S13. Each patient’s maximal percentage tumor shrinkage, as determined by the sum of the product of diameters (SPDs), was calculated, and a threshold of >10% tumor shrinkage was required. Patient samples were classified as signature-positive or signature-negative on the basis of the gene signature prediction. Twenty-one of 24 signature-negative patients (88%) displayed no measurable tumor shrinkage in response to SGN-40. Ten of 15 signature-positive patients (67%) displayed significant tumor shrinkage in response to SGN-40 (Fig. 5A). The overall accuracy of the signature in classifying patients correctly was 80% (P = 0.004). Consistent with the observed tumor shrinkage, the progression-free survival (PFS) of the signature-positive patients (predicted to respond) was significantly longer than that of the signature-negative patients, with a median PFS of 169 and 40 days, respectively (P = 0.001) (Fig. 5B). From these data, we conclude that a 15-gene qRT-PCR signature was able to predict outcomes after CD40 pathway stimulation with SGN-40.

Fig. 5

qRT-PCR 15-gene signature and CD40 gPAS predict sensitivity to SGN-40 in clinical trials. (A) The 15-gene signature predicts tumor shrinkage in DLBCL patients treated with SGN-40. Accuracy of signature is depicted in the lower right inset. (B) A Kaplan-Meier plot showing that the 15-gene signature–positive patients have a greater progression-free survival than signature-negative patients. (C) Schematic of fixed tissue, genome-wide expression profiling methodology. (D) CD40 gPAS status predicts sensitivity to SGN-40 in clinical trial samples. Heat map represents CD40 gPAS and GCB signature score by intensity of color. Red to green color represents signature score as being high to low. A high signature score for GCB or CD40 gPAS represents GCB subtype or CD40 pathway activation, respectively. The permutation tests (randomizing sample labels 10,000 rounds) were used to estimate the two-sided P values, which were calculated as the proportion of the sampled permutation where the area under the curves (AUCs) were greater than observed AUC (AUCobs) or less than 1 − AUCob. Cell line values were calculated from the data sensitivity data represented in Fig. 2.

Finally, we determined whether a gene signature consisting only of CD40 target genes (CD40 gPAS) could predict response to CD40 agonist therapy. We developed a method to enable genome-wide gene expression profiling from fixed tissue and applied this to the same clinical Phase 1 and 2 clinical trial samples (Fig. 5C) (see Materials and Methods). The CD40 gPAS signature was predominantly present in patients who did not show tumor shrinkage upon treatment with SGN-40 (Fig. 5D). Similar but less significant observations were found with a previously published CD40L activation signature (fig. S14 and table S2). There was no statistical association with response to SGN-40 with the GCB/ABC signatures despite our preclinical observations (Fig. 5D). Overall, these data suggest that CD40 pathway activation is a characteristic that can predict the responsiveness of tumors to CD40 stimulation.

Discussion

The role of CD40 in B cell cancers can differ in various cellular contexts. Stimulation of the CD40 pathway can cause tumor cell death (4, 5) or can promote proliferation and survival of NHL cells (6, 8, 9, 42), lymphomagenesis (10), and the transformation of primary B cells (1). Our observations show that preexisting CD40 pathway activation is a robust predictor of the cellular response to CD40 stimulation. Sensitive but not resistant cell lines generally did not show signs of CD40 pathway activation but were killed by SGN-40, perhaps as a result of increased intrinsic DNA damage and reactivation of the DNA damage response machinery upon CD40-triggered BCL6 down-regulation. Resistant cell lines with an active CD40 pathway may have already acquired the antiapoptotic signal that can result from chronic activation of CD40 (10), possibly explaining why stimulation of the CD40 pathway in these cells may promote rather than inhibit cell viability. The coexistence of BCL6 and p53 mutations in sensitive cells is intriguing because it has been shown that BCL6 normally represses p53 expression and dampens the DNA damage response (43). The observed p53 mutations in BCL6-expressing cell lines, in addition to BCL6-mediated repression of ATR and perhaps wild-type p53, likely contribute to lymphomagenesis in certain DLBCLs.

CD40 signaling drives multiple downstream signal transduction cascades; constitutive phospho-ERK was detected primarily in resistant cell lines that showed evidence of an active CD40 pathway activation. Phospho-ERK was detected in a mouse model of constitutively active CD40 signaling in B cells and was required for prolonged CD40-dependent cell survival (10). BCL6 down-regulation can also be mediated by the MEK/ERK pathway through direct phosphorylation of BCL6 by ERK and resultant ubiquitin-mediated degradation (44), which could also explain the lower levels of BCL6 in resistant cell lines.

In current clinical practice, tumor biopsy tissue is routinely fixed in formalin and then embedded in paraffin blocks. Genome-wide mRNA expression studies are technically challenging in formalin-fixed tissue and yet provide useful information. Researchers have had variable success with fixed tissue and technology platforms such as microarrays and complementary DNA (cDNA)–mediated annealing, selection, extension, and ligation (DASL) (45, 46). By training the Gene 1.0ST microarray platform on frozen and fixed tissue from matched cases to identify probes that correlate with one another, we hoped to harness the genes on the array that are most representative of the tumor biology. We detected signatures that predicted response to CD40 pathway stimulation in the clinical samples that were similar to those in cell lines. This approach may be applicable to malignancies other than lymphoma.

In sum, our findings suggest that antitumor responses to CD40 stimulation in NHL depend on the preexisting status of the CD40 pathway. NHL cells were more sensitive to SGN-40–mediated cell death when the CD40 pathway was inactive and were less sensitive in tumors in which the pathway was constitutively activated. The hypothesis that antitumor responses of agonist CD40 therapeutics can be predicted by CD40 pathway signatures will need to be further validated in upcoming clinical trials.

Materials and Methods

Antibodies and cell lines

SGN-40 was obtained from Seattle Genetics. Antibodies to BCL6, actin, and tubulin were all obtained from Santa Cruz Biotechnology. All other antibodies for Western blotting were obtained from Cell Signaling. The NHL cell lines DOHH2, Granta-519, Karpas-1106P, MHH-PREB-1, NU-DUL-1, NU-DHL-1, Ramos, RC-K8, REC-1, SC-1, SU-DHL-1, SU-DHL-4, SU-DHL-5, SU-DHL-6, SU-DHL-8, SU-DHL-10, SU-DHL-16, U-698-M, WSU-DLCL2, and WSU-NHL were obtained from Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH (DSMZ). The DB, Farage, HT, MC116, Pfeiffer, and Toledo cell lines were obtained from the American Type Culture Collection (ATCC). The A4/Fukada, SCC-3, and TK cell lines were obtained from Japan Health Sciences Foundation. The OCI-LY3, OCI-LY10, OCI-LY7, and U2932 cells were provided by L. Staudt (National Institutes of Health/National Cancer Institute) and R. Gascoyne (British Columbia Cancer Agency). All cell lines were maintained in RPMI 1640 supplemented with 10% fetal bovine serum (Sigma) and 2 mM l-glutamine except for OCI-LY3, OCI-LY10, and OCI-LY7, which were maintained in Iscove’s modified Dulbecco’s medium supplemented with 2-mercaptoethanol, penicillin/streptomycin, and 20% heparinized human plasma.

Cell viability assays

The in vitro efficacy of SGN-40 was measured by CellTiter-Glo Luminescent Cell Viability Assay (Promega Corp.). Cells were plated in quadruplicate at 1 × 103 to 5 × 103 per well in 384-well plates in RPMI containing 2% fetal bovine serum overnight before treatment with SGN-40, or the isotype control anti-gD, with cross-linker. The antibodies were added to experimental wells at a final concentration of 25, 9, 3, 1, 0.33, 0.11, 0.037, 0.012, 0.004, 0.0014, 0.00046, or 0.0015 μg/ml, with “non–drug-treated” control wells receiving medium alone. Cell viability was assessed 96 hours after drug treatment. The concentration of SGN-40 resulting in 25 and 50% inhibition of cell viability was calculated from a four-variable curve analysis and was determined from a minimum of three independent experiments.

Microarray data generation, somatic mutations, and data analysis

Gene expression analysis of NHL cell lines and tumor samples was carried out on the Affymetrix HGU133Plus_2.0 and Gene 1.0ST platform (Affymetrix), respectively. Microarray data sets are available from Gene Expression Omnibus (GEO) with accession numbers GSE15329, GSE18376, and GSE18377. The NHL cell line data on HGU133Plus_2.0 were normalized with gcRMA, and the tumor data on Gene 1.0ST were normalized with VSN. Somatic mutations across the NHL cell lines have been previously described (22). All data analyses were conducted with R (47).

CD40 gPAS

After filtering out the probe sets with the low intensity values or low variation across samples, paired t test was used to identify top 50 up- and down-regulated probe sets, defined as gUp50 and gDn50, between the seven pairs of the SGN-40–sensitive cell lines (treated with SGN-40 or isotype control): SU-DHL-16, SU-DHL-10, SU-DHL-8, SU-DHL-5, HT, Karpas-1106P, and RL. CD40 gPAS scores were calculated for each NHL cell line by subtracting the mean of the z score–transformed intensity values of the 50 down-regulated probe sets from the one of the 50 up-regulated probe sets. For the clinical samples, the gene symbols were used to map the gUp50 and gDn50 from HGU133Plus_2.0 to Gene 1.0ST, and then the mapped Gene 1.0ST transcripts were further filtered by only keeping the above-defined high-quality transcripts. Finally, the same method was applied to calculate the gPAS scores. For the area under the curve (AUC) test, permutation test was applied to test the performance of the gPAS scores in comparison with GCB/ABC signature by shuffling sample labels 10,000 times.

GCB DLBCL, ABC signatures

GCB/ABC classification was determined with Wright’s procedure (48). The lymphochip data were downloaded from the Lymphoma/Leukemia Molecular Profiling Project gateway (49). Among the original 27 GCB/ABC signature genes on the lymphochip, 23 can be mapped onto HGU133Plus_2.0 chip. The linear predictive scores of the mapped genes in our data set were calculated according to the Wright’s procedure, including rescaling our data set based on the mean and SD of individual genes in the lymphochip data set.

FFPE gene expression

We filtered probes based on a matching pair of 18 frozen and 18 fixed tissue DLBCL specimens based on a statistical significance threshold to generate a transcript set that would harbor the most reliable signal in fixed tissue. Matched frozen and fixed DLBCL cases were commercially procured from Cytomyx. Eighteen pairs of matched fresh-frozen and fixed tissue samples were profiled on the Gene 1.0ST platform. Spearman ranking correlation test was used to select a subset of high-quality transcripts with the following cutoff: ρ > 0.5 and P < 0.05.

Phospho-H2AX analysis

Fluorescence-activated cell sorting (FACS) was used to quantify H2AX phosphorylation in NHL cell lines before SGN-40 treatment. The H2AX Phosphorylation Assay kit (Upstate) was used following the manufacturer’s recommendations. Briefly, cells were fixed, permeabilized, and stained with the anti–phospho-histone H2AX-FITC (fluorescein isothiocyanate) conjugate, which recognizes H2AX phosphorylated at Ser139, or with normal mouse IgG conjugate as control, and then subjected to analysis on a FACSCalibur (Becton-Dickinson). Relative H2AX phosphorylation is expressed as the ratio between MFI of the stained sample and that of its control IgG.

Fluorescence in situ hybridization analysis

Commercially available LSI BCL6 Dual Color, Break Apart Rearrangement probe (Vysis/Abbott Laboratories) was used for the fluorescence in situ hybridization (FISH) experiments to detect the status of BCL6 in the cell lines used for the study. Cell lines were prepared for cytogenetic analysis as previously described (50). FISH on cytogenetic preparations was performed as described previously (51). Interphase nuclei and metaphases from normal human lymphocytes were used to set the cutoff limits for a positive result. The cutoff was set at 2.5% (average ± 3SD).

Microarray probe set selection using 30 NHL cell lines

The stepwise selection procedure depicted in Fig. 3A was used to select seven pairs of microarray gene probe sets (genes) for use in classifying cell lines and clinical samples. BCL6 was added to the 14 genes, yielding a total of 15 genes plus 5 normalizing genes. The total of 20 genes was prespecified as a practical number for the qRT-PCR assay. Fifteen genes is more than necessary for optimal cross-validated prediction of IC25 in the cell lines themselves; less than 5 genes are necessary for optimal prediction in this small sample. We allowed for more genes than necessary to increase the chance that a subset of the genes in the assay would perform well in clinical samples of FFPE tissue, which were obtained subsequent to development of the assay. The choice of the subset was made using the clinical samples as described in the next section.

The gene selection procedure was applied to a filtered subset of the 54,675 probe sets on the HGU133Plus2 arrays. Duplicate microarrays were available for a subset of the cell lines, and these were used to select the 9825 genes in the upper tertile of correlation between log2 expression in the duplicates and either the upper 90th percentile of mean expression or the upper 90th percentile of variance in expression. The retained gene probe sets had either high mean or high variance across the samples, and their measurements were replicable across duplicate arrays. This set was then reduced to the 637 meeting a 10% false discovery rate cutoff for association between log2 expression and IC25 rank. All 14 genes were chosen from these 637. First, we identified the 20 candidate genes with the highest correlation between log2 expression and the IC25 ranks. From the 20 candidates, a single probe was selected qualitatively, according to best evidence for CD40 regulation, CD40 pathway involvement, and/or membership in the GCB/ABC classifier. Qualitative selection improves the interpretability of the assay, and performance is similar to simply selecting the highest ranked among the 20. Next, a pair gene was chosen similarly from the 20 genes most anticorrelated with the first gene. The expression of the pair gene is subtracted from the expression of the main gene. This directly increases the dynamic range of the assay and can be interpreted as constructing a log scale ratio of up- to down-regulated genes (that is, by computing the signed average of log2 expression of the two genes, where the signs are for the correlation with IC25). We then selected the next main gene, this time from the 20 most associated with IC25 after adjustment for the signed average of the first two genes, using a linear model. The second pair gene was chosen using the same anticorrelation criterion. The remaining pairs are added to the signed average in stepwise fashion, with the selection of each main gene adjusted for the pairs chosen previously. The resulting seven main genes are not all highly correlated with one another, providing a more efficient representation of potentially multiple predictive pathways or subpathways. The use of the signed average in place of the linear model weights reduces the tendency to overfit the model for IC25. This is used as a candidate probe selection procedure, rather than a complete classifier. Final classification of the cell lines can be done either with a threshold for the signed average or with KNN with the 15 genes (that is, including seven pairs plus BCL6). Similar accuracies were obtained with KNN, with K between 4 and 13 (fig. S15). KNN performed better when using the genes with clinical samples on the qRT-PCR platform, as described in the next section.

Xenograft experiments

SU-DHL-4, a human DLBCL, was obtained from DSMZ and engineered to stably express luciferase gene (SU-DHL-4-luc cells). Farage-X1 was an in vivo–selected cell line derived from the original Farage xenograft tumors. Ramos-RA1 and WSU-DLCL2 were obtained from ATCC and DSMZ, respectively. All cells were cultured in RPMI 1640 media plus 1% l-glutamine with 10% fetal bovine serum plus hygromycin (600 mg/ml) and resuspended in Hank’s balanced salt solution (HBSS) before inoculations into animals. Female Fox Chase CB17/Icr SCID (severe combined immunodeficient) mice were obtained from Charles River Laboratories. All animal studies were performed in accordance with Genentech Institutional Animal Care and Use Committee (IACUC)–approved guidelines. Each mouse was given a subcutaneous injection of 20 million cells in the right flank area. When mean tumor volume reached 100 to 250 mm3, animals were randomly divided into groups of eight mice each and dosed with buffer or SGN-40 (10 mg/kg). All treatments were administered by intraperitoneal injections. Tumors were measured twice weekly until the end of the study. CD40 expression on tumors was confirmed by flow cytometry. Tumors were recovered from untreated mice, minced, and put through a 30-μm cell strainer (BD Biosciences) to achieve a single-cell suspension. The resulting single-cell suspension was stained with CD4-PE (phycoerythrin) antibody (Immunotech) and 7-aminoactinomycin D (7AAD; BD Biosciences) for flow cytometry (FACS) analysis. The MFI of CD40 was calculated from 7AAD-negative (viable cells) population.

Clinical samples and qRT-PCR signature

All clinical studies were conducted in compliance with the Declaration of Helsinki, the International Conference on Harmonization Good Clinical Practices, and the applicable U.S. Food and Drug Administration regulations. Institutional review boards (IRBs) approved the study for each site, and patients provided written informed consent before any study procedure.

FFPE archival tumor tissue from patients with DLBCL from the Phase 1 and 2 clinical trials described below was obtained from the clinical investigation sites with appropriate IRB approval and patient consent. Sections (4 to 6 μm) derived from the tumor tissue were mounted on glass slides, and one slide for each case was subjected to hematoxylin and eosin (H&E) staining with standard pathology laboratory protocol. H&E slides were used as a guide to macrodissect the remaining tumor-containing region for RNA extraction with the Ambion RecoverAll Total Nucleic Acid Isolation Kit for FFPE Tissues (Applied Biosystems). Total RNA (450 ng) per sample was reverse-transcribed in a total reaction volume of 20 μl with High Capacity Reverse Transcription cDNA Synthesis kit (Applied Biosystems). Manufacturer’s recommendations were followed with the exception of a shortened 60-min RT reaction at 37°C. Total RNA (5 ng) equivalent cDNA was mixed with 2× Universal Master Mix (no UNG) (Applied Biosystems). All amplifications were performed in triplicate in 384-well plates with a two-step (95°C for 15 s, 60°C for 1 min) PCR amplification procedure. Reactions were carried out to 40 cycles on a validated ABI 7900 real-time PCR system. Sequences of the primers and probes used are shown in fig. S16.

Using DLBCL patient samples from Phase 1 (11 samples) and Phase 2 (28 samples) clinical trials, we developed a classifier based on qRT-PCR for tumor size reduction of at least 10%, herein defined as sensitive, with weighted KNN, with weights for the 15 markers (UAP1, BTG2, CD40, VNN2, RGS13, CD22, LMO2, IFITM1, CTSC, CD44, PUS7, BCL6, EPDR1, IGF1R, and CD79B) determined with penalized regression (GLMNET). Model parameters were determined by cross-validation. A class was assigned for a new sample using the known classes of the K-nearest reference samples, where K is an integer between 1 and the total number of training samples. The nearest reference samples (nearest neighbors) are those with the smallest weighted average of the absolute differences (WAAD) between each of the 15 probe measurements for UAP1, BTG2, CD40, VNN2, RGS13, CD22, LMO2, IFITM1, CTSC, CD44, PUS7, BCL6, EPDR1, IGF1R, and CD79B, where the differences are between the probe measurements of new sample to be classified and those from each reference sample. We controlled the potential for overfitting by using a penalized regression approach and by performing 100 cross-validation replications to select the final classification model (that is, the model among the replicates with median overall accuracy). Reported accuracy estimates are medians across the CV (cross validation) replicates, and robust P values were computed via permutation tests over the whole procedure. Similar accuracies were obtained with KNN, with probe weights fixed at 1 and K between 4 and 13 (fig. S17), whereas training the probe weights was found to improve cross-validated overall accuracy by three to five percentage points.

Clinical studies

SG040-0004 study design

This was an open-label, multidose, Phase 2 study in patients with relapsed DLBCL conducted at 10 clinical sites in the United States. The study was conducted in compliance with the Declaration of Helsinki, the International Conference on Harmonization Good Clinical Practices, and the applicable U.S. Food and Drug Administration regulations. IRBs approved the study for each site, and patients provided written informed consent before any study procedure.

All patients were treated with SGN-40 using the intrapatient dose-escalation schedule (fig. S18). Patients received six doses of SGN-40 in cycle 1. Patients with continuing clinical benefit could receive additional treatment cycles consisting of four doses of SGN-40. Clinical benefit was defined as stable disease, partial remission, or complete remission based on imaging studies and absence of significant unresolved toxicity at the time that the next cycle was scheduled to begin. Restaging evaluations were scheduled for 1 week after the last dose of SGN-40 in cycles 1, 2, 4, and 6. Patients could receive a maximum of 12 treatment cycles, continuing until disease progression or 2 cycles beyond complete remission.

SG040-0004 study population

Eligible patients were ≥18 years old with histologic diagnosis of diffuse large B cell lymphoma (DLBCL) and at least one measurable site of disease (≥2 cm). Patients were required to have progressed or relapsed since the most recent therapy and have received at least one systemic therapy (consisting of combination chemotherapy plus rituximab). In addition, patients must also have received standard salvage therapy consisting of combination chemotherapy with or without rituximab, or have been ineligible for intensive salvage therapy. If applicable, patients must have completed autologous stem cell transplant (aSCT) at least 12 weeks before registration and be at least 4 weeks since last chemotherapy or monoclonal antibody. Patients were required to have adequate performance status, no significant comorbidities or central nervous system (CNS) lymphoma, and adequate organ function.

SG040-0002 study design

This was an open-label, dose-escalation, Phase 1 study in patients with refractory or recurrent B cell NHL at five U.S. centers. The study was conducted in compliance with the Declaration of Helsinki, the International Conference on Harmonization Good Clinical Practices, and the applicable U.S. Food and Drug Administration regulations. IRBs approved the study for each site, and patients provided written informed consent before any study procedure.

Six cohorts were evaluated (fig. S19). In the first dosing group, all patients received SGN-40 at a flat dose of 2 mg/kg administered intravenously weekly for 4 weeks followed by intrapatient dose-loading schedule for all subsequent cohorts in this study. A minimum of 4 patients were enrolled in each cohort, and an additional 15 patients were treated at the highest dose schedule to examine further the safety and tolerability of SGN-40.

Patients were treated with diphenhydramine and acetaminophen, 30 to 60 min before and 4 hours after the infusion. Symptoms suggestive of cytokine release syndrome developing within 24 hours of SGN-40 infusion (for example, headache, fever, and muscle aches) were treated with a 24-hour course of prednisone per investigator’s discretion. After cycle 1, patients with radiographic objective responses or stable disease with evidence of clinical benefit were eligible for a second cycle (four weekly doses at the highest cohort-specific dose). After completing treatment, patients were monitored for delayed toxicity, lymphocyte recovery for an additional 6 weeks, and evidence of disease.

SG040-0002 study population

Eligible patients were ≥18 years old with refractory or recurrent B cell NHL and at least one measurable site of disease (≥2 cm). Subtypes of B cell NHL included DLBCL, mantle cell lymphoma (MCL), follicular cell lymphoma (FL), marginal zone lymphoma (MZL), and small lymphocytic lymphoma (SLL) as determined by World Health Organization (WHO) criteria (52). Patients were required to be at least 4 weeks since last chemotherapy, 6 weeks since last rituximab treatment, and 4 months since aSCT. For patients with DLBCL who had not received aSCT as salvage therapy, documentation of refusal or lack of eligibility was required. Patients were required to have adequate performance status, >3-month life expectancy, no significant comorbidities or CNS lymphoma, and adequate organ function.

Supplementary Material

www.sciencetranslationalmedicine.org/cgi/content/full/3/74/74ra22/DC1

Fig. S1. Soluble CD40L and SGN-40 display similar sensitivity profile across NHL cell lines.

Fig. S2. Some of the resistant cell lines to SGN-40 have increased cell viability after treatment with SGN-40 (1 μg/ml), relative to an isotype control antibody.

Fig. S3. Somatic mutation association with SGN-40 activity in NHL cell lines.

Fig. S4. Introduction of DNA damage by gemcitabine sensitizes cells to CD40 stimulation.

Fig. S5. Introduction of DNA damage by cisplatin sensitizes cells to CD40 stimulation in cell lines.

Fig. S6. The top 50 CD40 gPAS up-regulated genes.

Fig. S7. The top 50 CD40 gPAS down-regulated genes.

Fig. S8. A CD40L activation signature correlates with sensitivity to CD40 stimulation in cell lines.

Fig. S9. Increased levels of phospho-ERK are enriched in SGN-40–resistant cell lines.

Fig. S10. NFκB activation is not significantly associated with CD40 gPAS or sensitivity to SGN-40.

Fig. S11. Stepwise algorithm for final 15-gene selection for signature.

Fig. S12. qRT-PCR signature format predicts sensitivity and resistance in preclinical models.

Fig. S13. Baseline patient characteristics from SG040-0002 and SG040-0004 clinical trials that were analyzed in this correlative study.

Fig. S14. A CD40L activation signature is less effective at predicting sensitivity to SGN-40 in clinical trials.

Fig. S15. Cell line model accuracy as a function of K with 100 10-fold CV replications.

Fig. S16. Sequences of primers and probes for 15-gene signature.

Fig. S17. Clinical model accuracy as a function of K with 100 10-fold CV replications.

Fig. S18. Intrapatient dose-escalation schedule for the Phase 2 trial (SG040-0004).

Fig. S19. SGN-40 dose-escalation schedule for cycles 1 and 2 for the Phase 1 trial (SG040-0002).

Table S1. NHL cell line panel data.

Table S2. Patient data.

Footnotes

  • * These authors contributed equally to this work.

  • Citation: B. Burington, P. Yue, X. Shi, R. Advani, J. T. Lau, J. Tan, S. Stinson, J. Stinson, T. Januario, S. de Vos, S. Ansell, A. Forero-Torres, G. Fedorowicz, T. T. C. Yang, K. Elkins, C. Du, S. Mohan, N. Yu, Z. Modrusan, S. Seshagiri, S.-F. Yu, A. Pandita, H. Koeppen, D. French, A. G. Polson, R. Offringa, N. Whiting, A. Ebens, D. Dornan, CD40 Pathway Activation Status Predicts Response to CD40 Therapy in Diffuse Large B Cell Lymphoma. Sci. Transl. Med. 3, 74ra22 (2011).

References and Notes

  1. Acknowledgments: We thank M. Lee, T. Wang, and members of the clinical operations groups at Genentech and Seattle Genetics for their support in making this study a reality. We also thank T. Behrens and J. Settleman for reviewing and editing the manuscript, R. Gentleman and R. Bourgon for statistical analysis methods guidance, and T. Lewis for SGN-40 in vitro experiment advice. We are most grateful to L. Staudt and R. Gascoyne for their generous contribution of cell lines. We apologize to authors whose work could not be cited owing to space limitations. Author contributions: D.D. designed the study and wrote the manuscript. B.B. and P.Y. performed all statistical analyses. X.S., J.T.L., J.T., S. Stinson, J.S., T.J., T.T.C.Y., K.E., C.D., S.M., N.Y., Z.M., S. Seshagiri, S.-F.Y., A.P., H.K., D.F., A.G.P., R.O., and A.E. performed or assisted in preclinical experiments. All other authors were clinical investigators or assisted in dacetuzumab clinical trials. Competing interests: All authors affiliated with Genentech Inc. and Seattle Genetics are employees of their respective affiliation. A patent has been filed describing diagnostic markers that predict patient response to CD40 therapy. Accession numbers: Microarray data sets are available from GEO with accession numbers GSE15329, GSE18376, and GSE18377.
View Abstract

Navigate This Article