B-Cell Lymphoma Patient-Derived Xenograft Models Enable Drug Discovery and Are a Platform for Personalized Therapy

Lymphoma is the most common hematologic malignancy, and B-cell lymphoma accounts for 85% of all lymphomas (1). The majority of B-cell lymphoma patients respond to initial therapy; however, most eventually relapse due to the development of therapeutic resistance (2, 3). Thus, an improved understanding of the biology of relapsed/refractory B-cell lymphoma is critically needed to develop alternative treatment strategies for these patients (4–6).

The evaluation of novel drug targets using established B-cell lymphoma cell lines is limited by the inexact correlation between responsiveness observed in the cell line versus the patient sample (7). Similarly, “xenograft” models wherein human cancer cell lines are transplanted into immunocompromised mice do not represent the full spectrum of cancers because (i) the cell lines are not derived from patients, and (ii) the model lacks the tumor microenvironment in vivo. In contrast, patient-derived xenografts (PDX) possess both of these refinements. Unlike the cell line–derived tumor models, PDX mouse models contain heterogeneous tumor cell populations (8) similar to the patient tumor cell population, including possible cancer stem cells (9). Recent studies have indicated that PDX models can also recapitulate the treatment responses of the parental tumor and can be used to predict the choice of therapeutic target and regimen (10–13). Therefore, PDX models provide a valid experimental platform to assess the biology and progression of B-cell lymphoma and its response/resistance to novel therapeutic agents.

We previously established the first mantle cell lymphoma (MCL) PDX model with cells isolated from a patient then transplanted into a human fetal bone chip implanted in the mice to investigate MCL biology and drug responses (14). In this PDX model, the primary MCL tumor metastasized to the lymph nodes, spleen, bone marrow, and gastrointestinal tract of the host mice, mimicking MCL clinical features. Bone marrow involvement has been reported in diffuse large B-cell lymphoma (DLBCL; ref. 15), follicular lymphoma (FL; ref. 16), marginal zone lymphoma (MZL; ref. 17), and Burkitt lymphoma (BL; ref. 18), with a significantly poor prognosis for patients with this involvement (19, 20). Thus, we developed various B-cell lymphoma PDX models and recapitulated the pathological and clinical characteristics, molecular profiles, disease progression, and response to therapeutic agents in these B-cell lymphoma PDXs. Our results indicate that PDX mouse models are an indispensable tool towards personalized treatment for B-cell lymphoma.

Peripheral blood, apheresis, biopsy tissues isolated from spleen and lymph nodes, bone marrow aspirates, ascites, or pleural effusion were obtained from B-cell lymphoma patients who provided informed consent. The sample collection protocol was approved by the Institutional Review Board at The University of Texas MD Anderson Cancer Center. All procedures were conducted in accordance with the Declaration of Helsinki. Mononuclear cells were separated by Ficoll-Hypaque density centrifugation, and tumor cells were isolated using anti-CD19 antibody-coated magnetic microbeads (Miltenyi Biotec) and maintained in RPMI1640 medium (Life Technologies) supplemented with 10% heat-inactivated FBS, penicillin (10,000 U/mL, Sigma), streptomycin (10 mg/mL, Sigma), and l-glutamine (29.2 mg/mL, Life Technologies). These isolated tumor cells were used for molecular profiling, in vitro experiments, and inoculation into mice. The drugs or agents used for the in vitro or in vivo drug assays are listed in Supplementary Table S1.

Six- to 8-week-old male CB-17 SCID mice (Harlan) or NSG (Nod SCID Gamma) mice (The Jackson Laboratory) were housed and monitored in our animal research facility. All experimental procedures and protocols were approved by the Institutional Animal Care and Use Committee of The University of Texas MD Anderson Cancer Center. Fresh human fetal bones of 17 to 19 gestational weeks (Advanced Bioscience Resources) were subcutaneously implanted into SCID or NSG mice (SCID/NSG-hu). Approximately 4 to 6 weeks following implantation, 5 × 10⁶ freshly isolated lymphoma cells were directly injected into human fetal bone implants within SCID/NSG-hu hosts after the mice were anesthetized with 5% isoflurane vaporizer. Mouse serum was collected, and the levels of circulating human β₂M in mouse serum [human β₂-microglobulin (β₂M) ELISA Kit (Abnova Corporation)] were used to monitor tumor engraftment and burden. Once tumor growth was detected in the first generation (G), the mice were sacrificed, and tumor masses were isolated. The tumor cells were tested for human CD20 expression and then inoculated in NSG mice as the second generation (G2). Isolated G2 tumor cells were utilized for high-throughput drug screening. Meanwhile, the remaining tumor mass was cut into 3 mm³ sections that were then passaged into 10 to 20 NSG mice for next-generation (G3) tumor growth. From G3, the tumor mass was equally cut into 3 mm³ pieces and passaged into 20 to 50 NSG mice (3–5 mice/group, dependent on the experimental drug treatment) for in vivo treatment. During and after treatment with vehicle control or the indicated drugs, mouse serum was collected, and tumor burden was evaluated by measuring either human β₂M levels or tumor volume to determine therapeutic efficacy in the PDX models. Freshly isolated cells from the tumor mass of each generation and treatment group were labeled with FITC/PE-conjugated antihuman CD5, CD10, CD20, or CD45 mAb (BD Biosciences) to validate the population of human lymphoma cells by flow cytometry.

Excised tissues from patients and PDX mice were fixed in 10% formalin solution, processed by standard methods, embedded in paraffin, sectioned at 5 μm, and stained with H&E. For IHC evaluation, the tissues were stained with antibodies against human CD20, KI-67, Cyclin D1 or PAX5 (Dako), counterstained with Harris hematoxylin and examined by standard light microscopy. Samples were analyzed using an Olympus BX51TF microscope equipped with UPlan FL 40×/0.75 and 20×/0.50 objective lenses (Olympus).

Tumor cells isolated from original patient samples and/or from different generations of PDX mice were utilized for high-throughput in vitro screening of potential drugs and their combinations using the CellTiter-Glo luminescent cell viability assay (Promega). Cells were seeded in triplicate in a 96-well plate with 2 × 10⁵ cells per well and were treated with different combinations of potential therapeutic agents (Supplementary Table S1) for 48 hours. In the last 30 minutes, 50 μL of CellTiter 96 Aqueous One Solution Reagent were added to the culture wells and incubated at 37°C in 5% CO₂. Light absorbance of formazan was measured at 495 nm on a universal microplate reader equipped with KC4 software (Biotek Instruments).

After fasting for 8 to 12 hours, the PDX mice were imaged by microPET/CT under general anesthesia with 5% isoflurane vaporizer. The ¹⁸F-fluorodeoxyglucose (¹⁸F-FDG) biodistribution was evaluated with the Albira trimodal PET/SPECT/CT system [PET, SPECT (single photon emission CT), and CT).

PDX tumors (3 mm³) were incubated with 1% SDS (with beta-mercaptoethanol) and diluted in five 2-fold serial dilutions in lysis buffer containing 1% SDS. Serial diluted lysates were arrayed on nitrocellulose-coated slides (Grace Bio-lab) with the Aushon 2470 Arrayer (Aushon BioSystems). A total of 5,808 array spots were arranged on each slide, including the spots corresponding to positive and negative controls prepared from mixed cell lysates or dilution buffer, respectively. Each slide was probed with a validated primary antibody plus a biotin-conjugated secondary antibody. Only antibodies with a Pearson correlation coefficient between reverse-phase protein array assay (RPPA) and Western blotting of greater than 0.7 were used. Antibodies with a single or dominant band on Western blotting were assessed by direct comparison to RPPA using cell lines with differential protein expression or modulated with ligands/inhibitors or siRNA for phospho- or structural proteins, respectively. The antibodies used to probe the slides are listed in Supplementary Table S2. The signal obtained was amplified using a Dako Cytomation–catalyzed system (Dako) and visualized by DAB colorimetric reaction. The slides were scanned, analyzed, and quantified using customized Microvigene software (VigeneTech Inc.) to generate spot intensity. Each dilution curve was fitted with a logistic model (“Supercurve Fitting” developed by the Department of Bioinformatics and Computational Biology in MD Anderson Cancer Center). The fitted curve was plotted with the signal intensities, both observed and fitted, on the y-axis and the log2-concentration of proteins on the x-axis for diagnostic purposes. The protein concentrations of each slide set were normalized by median polish, which was corrected across samples by the linear expression values using the median expression levels of all antibody experiments to calculate a loading correction factor for each sample.

OncoPlus is a 1,212 hybrid capture gene panel developed at the University of Chicago. The panel contains genes associated with cancer in both the somatic and inherited contexts, and utilizes a custom SeqCap EZ capture (Roche Applied Science) with 2 × 100 bp sequencing performed on an Illumina Hiseq-2500. Data were processed in customized computational pipelines using a combination of publicly available and customized software. The serial generations of PDX samples with the original patient samples were sequenced using OncoPlus. Approximately 1,500 variants were detected within the exonic territories of the 1,212 genes in each sample. These variants were then filtered on the basis of their 1000G frequencies (21) to remove common germ line SNPs, and their coding effects.

Multiple linear regression was applied to investigate drug treatment effects on the primary patient tumors and PDXs in vitro. Linear mixed-effects regression model was used to assess the change of human β₂M or CD5⁺CD20⁺ cells by treatment. Survival analysis was conducted by log-rank test. P values of <0.05 were considered statistically significant.

We collected clinical samples from 16 patients with several types of B-cell lymphomas (Table 1), including MCL (n = 8), DLBCL (n = 3), FL (n = 2), BL (n = 1), and MZL (n = 2). In total, 25% of the patients had indolent lymphomas (4/16). Furthermore, of the 16 patients, 9 (56%) were newly diagnosed and untreated clinically, and 7 (44%) patients were relapsed after prior treatment with 1–3 chemotherapy or targeted therapy treatments. Of the 8 MCL patients, 4 (25%) patients were exposed to ibrutinib treatment after prior chemotherapies. In addition, 10 patients displayed bone marrow involvement as detected by bone marrow aspiration, except for PT6, PT9, and PT13 who displayed no evidence of bone marrow infiltration. No bone marrow aspirations were performed on PT4, PT12, and PT16. The clinical pathology reports showed TP53 deletion in PT5, Bcl-2 deficiency in PT2, and Bcl-2 and Bcl-6 rearrangement in PT16. These genetic changes may reflect refractoriness or may be potential molecular targets in both patients and PDXs.

Fresh lymphoma cells isolated from patient samples were injected into SCID/NSG-hu mice. A schema of the establishment and the application of the primary B-cell lymphoma PDX model are shown in Supplementary Fig. S1A. Human β₂M levels, an indicator of lymphoma tumor growth, increased progressively in the peripheral blood of PDX mice, reflecting the successful engraftment and subsequent growth of the patient tumors. The levels of circulating human β₂M after 3 weeks of patient tumor cell engraftment in the 16 B-cell lymphoma PDX mouse models showed successful tumor engraftment compared with the human β₂M levels in NSG or NSG-hu mice (Supplementary Fig. S1B, P < 0.05). The tumors cells from PT5 were inoculated into SCID-hu mice and were not passaged to next generations. Of the remaining 15 PDXs, 10 of 15 PDXs were serially passaged for more than six generations, indicating a 67% passage success rate.

We investigated whether the PDXs of each generation possessed similar clinical, pathological, and molecular features as the original patient tumor. The histologic examination of nine generations of subcutaneous tumor masses of PT1 MCL-PDXs and PT2 BL-PDXs indicated consistent tumor features in each generation (Supplementary Fig. S2A). Once the tumors grew in the subcutaneous implant sites, the tumors frequently metastasized to the mouse spleen, mimicking disease progression in humans (Supplementary Fig. S2B and S2C). Specifically, PET/CT scans of PT1 and PT5 diagnosed splenomegaly in the clinic (Supplementary Fig. S2D, left), which was also observed in the respective patient PDXs (PT1-G3 NSG mice and PT5-G1 SCID-hu mice; Supplementary Fig. S2D, middle). Furthermore, H&E staining showed the characteristic histology of human lymphoma cells in the spleens of PT1-G3 and PT5-G1 PDX mice (Supplementary Fig. S2D, right).

Next, to investigate the immunologic and pathologic characteristics between the patient tumors and their respective PDX tumors, immunophenotype and IHC data were extracted from our patient clinical database and were compared with the PDX immunophenotype and IHC PDX findings. The representative data of five models (PT2-BL-G12, PT7-MZL-G6, PT8-MCL-G2, PT10-MCL-G3, and PT12-DLBCL-G3) displayed the same immunophenotypes (Fig. 1A) and anti-human CD20-positive staining (Fig. 1B) as their original patient tumors. In addition, anti-human PAX5-positive staining of PT2-BL and PT10-MCL also matched the original patient tumor staining results (Fig. 1C). Meanwhile, the percentage of human cyclin D1- and Ki-67–expressing cells was consistent between the two MCL PDXs (PT8-MCL and PT10-MCL) and their respective patient samples (Fig. 1D).

Using the OncoPlus Universal Cancer Mutation Analysis Panel, we identified the same mutations in 1,212 cancer-associated genes between the patient primary tumors and PT8-MCL and PT10-MCL PDX-generated tumors across multiple generations. As a result, each sample had approximately 55 variants, which included somatic variants as well as some rare SNPs that cannot be distinguished as a matched normal was not sequenced. The somatic variants included pathogenic drivers as well as passenger mutations of uncertain significance. A list of all variants detected by UCM-OncoPlus in the primary tissue is shown in Supplementary Table S3. Only one additional mutation was identified in PT8-MCL-G3. Specifically, PT8-MCL-G3 gained a Thr1627Met mutation in DNAH5 at 20% allele frequency. The DNAH5 mutation has not been reported to be associated with B-cell lymphoma. On the basis of this analysis, these two representative PDX mouse models appear to maintain original patient mutations of these specific 1,212 cancer-associated genes without the loss or accumulation of additional mutations (Fig. 2).

In the phase II ibrutinib single-agent clinical trial, we observed ibrutinib-induced “compartmental shift” (lymphocytosis) of tumor cells from the primary tumor site into the peripheral blood in 34% of MCL patients treated with ibrutinib, an oral covalent inhibitor of Bruton tyrosine kinase (BTK; ref. 22). Here, we imitated this phenomenon in PT5-MCL-bearing PDX mice. The patient PT5 tumor cells were inoculated into the human fetal bone chip of SCID-hu mice. Three-week posttumor inoculation, these mice were administered ibrutinib (25 mg/kg oral gavage daily). A transient increase of human CD5⁺CD20⁺ cells in the mouse peripheral blood was detected by flow cytometry on day 10 of treatment in the ibrutinib-treated group but not in the control group, representing an ibrutinib-induced shift of human MCL cells from the implanted bone chip (primary tumor site) to the mouse peripheral blood (Fig. 3A and B). Specifically, approximately 50% cells were human CD20-positive in the peripheral blood on day 10 of ibrutinib treatment, but these cells were not observed in the peripheral blood of control animals (P < 0.0001 between vehicle control and ibrutinib groups at day 10). These findings suggest that we were effectively able to recapitulate the biology of the human disease using the PDX mouse model.

In addition, once the ibrutinib-induced transient increase of human CD5⁺CD20⁺ cells in the mouse peripheral blood was detected, the MCL-bearing SCID-hu mice were treated with ibrutinib plus rituximab to determine whether this combination increased survival and reduced tumor burden compared with single agent therapy. This combination was utilized because we hypothesized that targeting CD20 with rituximab while simultaneously targeting BTK with ibrutinib would produce greater anticancer effects in vivo. The MCL-bearing G1 mice were divided into the following four groups: vehicle control, ibrutinib treatment alone, rituximab treatment alone, and ibrutinib and rituximab combination treatment. Rituximab was intravenously administered at 10 mg/kg every 3 days for a total of 7 doses. Tumor burden was monitored by assessing human β₂M levels in the mouse serum before combination treatment (day 0) and after treatment (day 30). The rituximab and ibrutinib combination reduced the human β₂M levels to almost undetectable levels compared with either single agent group or the vehicle control group (Fig. 3C, P < 0.01, ibrutinib plus rituximab vs. vehicle control or rituximab; P < 0.05, ibrutinib plus rituximab vs. ibrutinib). Furthermore, ibrutinib markedly increased the overall survival of the MCL-bearing mice compared with vehicle control (n = 5) or rituximab alone (n = 5). Importantly, all mice treated with ibrutinib plus rituximab (n = 5) survived at least 90 days after beginning combination therapy (Fig. 3D, P = 0.0027 for ibrutinib + rituximab vs. rituximab alone; P = 0.0026 for ibrutinib plus rituximab vs. vehicle control and P = 0.134 for ibrutinib + rituximab vs. IBN alone). These results demonstrate the marked effects of this combination in promoting survival in MCL-bearing PDX mice. On the basis of these preclinical data, ibrutinib plus rituximab was investigated and found to be an effective regimen in a clinical trial with relapsed or refractory MCL patients (23).

We next investigated whether the primary patient tumors and PDXs displayed similar responses to drug treatment in vitro. As shown in Fig. 4, freshly isolated PT2-BL cells from the original patient sample (Fig. 4A), from the G1 SCID-hu mouse (Fig. 4B), and from the G2 NSG mouse (Fig. 4C) were treated with a panel of drugs: ibrutinib, BGB-3111, carfilzomib, ABT-199, Cal-101 (idelalisib), and KPT-330 at different doses. All of the tumor cells, from both the patient and PDXs, showed the same drug response pattern. Of note, PT2-BL primary tumor cells were mostly resistant to the BCL-2 inhibitor ABT-199 (Fig. 4A), which may be explained by the Bcl-2 deficiency identified in the PT2-BL clinical pathology report (data not shown). Importantly, the PDX tumors from both PT2-BL-G1 and PT2-BL-G2 were also resistant to ABT-199 (Fig. 4B and C); even after multiple passages to G6 and G7, the PT2-BL PDX model still maintained resistance to ABT-199 (Supplementary Fig. S3), indicating that the PDXs most likely retained this Bcl-2 deficiency during multiple generations of tumor passage. These results demonstrated that the PDX model reliably displayed a drug resistance pattern that accurately reflected the disease biology of the patients.

PT8 was a patient with relapsed/refractory MCL who had clinical primary resistance to ibrutinib. We validated this resistance to ibrutinib by testing in vitro growth inhibition of the primary patient tumor cells (Fig. 4D) and PT8-MCL cells isolated from PDX tumors of G2 (Fig. 4E) and G3 (Fig. 4F). In addition, the in vitro growth inhibition data showed that these MCL cells from the PT8 primary patient sample and PDX tumors were sensitive to carfilzomib or ABT-199 (Fig. 4D–F). Furthermore, carfilzomib completely inhibited tumor growth in PT8-MCL-G3 PDX mice, indicating that carfilzomib inhibited the tumor growth of a primary ibrutinib-resistant tumor in vivo using a PDX mouse model (Supplementary Fig. S4).

PT12 was a patient with newly diagnosed ABC-type DLBCL. The tumor cells from the original patient, PDX-G2, and PDX-G3 of PT12 showed the same drug response pattern (Fig. 4G–I). Taken together, these results demonstrated that PDX models can be used to examine the effects of small-molecule–targeted agents in vitro with eventual validation in vivo.

PT1-MCL was confirmed ibrutinib-sensitive by in vitro testing of the primary tumor cells, and the PDX was established and passaged to subsequent generations. Beginning in G3, the PT1-MCL PDX mice were exposed to ibrutinib by daily oral gavage to confer ibrutinib resistance as shown in Fig. 5A. This daily ibrutinib administration induced the development of an acquired ibrutinib-resistant tumor in PDX-G4. The cell viability of isolated PDX tumor cells treated with ibrutinib was not significantly different between G1 and G2 as well as between G5 and G6 (Fig. 5B, P > 0.05). However, the cell viability of the G5 and G6 tumor cells was significantly higher than of the G1 and G2 tumor cells after ibrutinib treatment (P < 0.01), indicating that the G5 and G6 tumor cells were resistant to ibrutinib. We also validated ibrutinib resistance in vivo by treating the PDX-G5 mice with vehicle control or ibrutinib (25 mg/kg, oral gavage daily). The tumor masses and PET-CT scans demonstrated no difference in tumor burden between the ibrutinib-treated mice and vehicle control mice (Fig. 5C, P > 0.05), indicating that acquired drug resistance was established in PDX-G5.

To identify the underlying mechanisms associated with acquired ibrutinib resistance, we first compared the protein expression profiles of the PDX samples before and after ibrutinib exposure by RPPA analysis. The RPPA data revealed that the downstream BTK signaling molecule PLC-γ2, the canonical NFκB protein p65, and Src (pY416) were downregulated in the ibrutinib-resistant PDX cells. Instead, a component of the PI3K catalytic subunit p110 (p110alpha) and members of the BCL-2 antiapoptotic family Bcl-xL and Mcl-1 were upregulated after consistent exposure to ibrutinib (Fig. 5D). These results suggest that alternative signaling pathways may underlie ibrutinib resistance. Next, we aimed to inhibit the PI3K pathway while continuing to inhibit BTK. To this end, we targeted the PI3K pathway with idelalisib (Cal-101), the FDA-approved agent for the treatment of CLL, SLL, and FL, which targets all PI3K p110 isoforms at varying IC₅₀ values, in combination with ibrutinib. The in vitro data showed that the freshly isolated MCL cells from the G5 tumor mass of ibrutinib-exposed PDX mice were resistant to ibrutinib but the combination of ibrutinib with Cal-101 or carfilzomib overcame drug resistance after 48 hours of incubation with 10 μmol/L ibrutinib plus 10 μmol/L Cal-101 or 10 nmol/L carfilzomib (Fig. 5E; Control vs. IBN, carfilzomib, or Cal-101, P > 0.05; Control vs. IBN+Cal-101 or IBN+carfilzomib, P < 0.01). Next, the combination of ibrutinib with Cal-101 or carfilzomib effectively inhibited tumor growth of this ibrutinib-resistant tumor in vivo (Fig. 5F and G, ibrutinib vs. ibrutinib + Cal-101 or ibrutinib + carfilzomib, P < 0.01). These results indicated that even though the tumor was resistant to single-agent ibrutinib, targeting the PI3K pathway simultaneously with Cal-101 or targeting the proteasome downstream of PI3K signaling significantly inhibited tumor growth; therefore, simultaneous inhibition of the BCR signaling pathway and the PI3K signaling pathway or the proteasome may be an effective method to treat ibrutinib resistance.

We determined whether an established drug-resistant PDX model could precisely inform the therapeutic choices for an individual patient in the clinic. PT15 was an 88-year-old female with relapsed MCL (Table 1). Her treatment history included six cycles of R-CHOP, two cycles of rituximab plus bendamustine, rituximab, and bortezomib for 10 cycles, local radiation, resumption of rituximab and bendamustine for an additional six cycles, and high-dose chemotherapy followed by autologous stem cell transplantation. The patient relapsed after short remissions from all of these therapies. However, the patient responded to ibrutinib single-agent treatment and was in complete remission for 3 years (green arrow in Fig. 6A). Once the patient relapsed from ibrutinib single-agent therapy (red arrow in Fig. 6A), the patient's PDX model was created using these now ibrutinib-resistant lymphoma cells.

After 79 days from the collection of PT15 MCL tumor cells, we isolated PT15-MCL-G2 cells and treated with a panel of drugs. We found that the G2 cells were most sensitive to bortezomib (BTZ, velcade) compared with other agents (Fig. 6B, P ≤ 0.002). Bortezomib is FDA-approved for relapsed/refractory MCL; therefore, we were able to treat PT15 with a bortezomib-regimen guided by the PDX results (bortezomib, rituximab, and dexamethasone). The bortezomib-based regimen dramatically reduced the levels of tumor cells in the peripheral blood (lymphocytosis) of the patient (purple arrow in Fig. 6A). Taken together, our data indicated that the PDX mouse model identified an efficacious therapy for a relapsed/refractory MCL patient, strongly suggesting that the PDX model is a valid experimental platform that can guide clinical decision-making with respect to therapeutic agents.

Recent studies have shown a remarkable correlation between drug activity in PDXs and clinical outcomes when patients with advanced cancers were treated with selected regimens based on their PDX treatment responses (10, 24). These findings suggest that PDXs are a robust model to assess responses to novel drugs and can be used to predict clinical efficacies of treatment regimens. Furthermore, patient-derived primary cancer cell cultures (PDPC) from a biopsy sample were shown to retain tumor heterogeneity and were used to identify an effective therapy for a patient with respiratory papillomatosis (25). In addition, PDPCs have also been used to identify effective drug combinations to overcome resistance to targeted therapy in lung cancer (26). These studies provide the rationale that PDX-based adaptive therapy could be utilized to select a beneficial patient regimen.

In this study, we established 16 different B-cell lymphoma PDX models. The overall passage success rate was 67% in all of the PDX models. MCL-PDXs and DLBCL-PDXs had a 75% (6/8) and 67% (2/3) success rate of passaging across multiple generations, respectively. These success rates are higher than other recently published B-cell lymphoma PDX models (8, 27). Specifically, Townsend and colleagues created a large, publicly available repository of leukemia and lymphoma patient-derived PDXs. The engraftment success rates by tail-vein injection were higher in acute lymphoblastic leukemias but lower in lymphomas. In addition, Townsend and colleagues implanted lymphoma tissue under the renal capsule with a 30.2% success rate and experienced difficulty in developing low-grade lymphoma models (27). Of note, we also successfully set up a PT2-BL-PDX model that was passaged across multiple generations, and the clinical pathology report showed a Bcl-2 deficiency in the patient's original tumor. Correspondingly, PT2-BL tumor cells from the original patient sample displayed resistance to ABT-199. Furthermore, the PT2-BL PDX model reliably displayed resistance to ABT-199 even after 7 generations, suggesting that that the PDXs most likely maintained the Bcl-2 deficiency across multiple generations.

To further elucidate the accurate reflection of genetic mutations in B-cell lymphoma PDXs compared with the original patient tumor, we investigated the genetic similarities of the PDX tumors by sequencing two sets of PDX models in comparison with the patient primary tumors. Only an additional Thr1627Met mutation in DNAH5 was observed at 20% allele frequency in G3 of one PDX model compared with the patient tumor. This DNAH5 mutation has not been previously reported to be associated with B-cell lymphoma. These results suggest that these two PDXs maintained the original patient genetic profiles without the loss or accumulation of additional mutations.

Most B-cell lymphoma patients relapse after initial therapy, and secondary therapies are urgently needed to cause remission. Ibrutinib, a first-in-class, once-daily, oral BTK inhibitor, was approved by the FDA in 2013 to treat relapsed/refractory MCL. In our prior multiple-center phase II clinical trial, the overall response rate in relapsed/refractory MCL patients was 68%, with a median progression free survival (PFS) of 13.9 months, surpassing the effectiveness of other therapies (22). However, despite the dramatic responses to ibrutinib, resistance to ibrutinib inevitably develops. Moreover, the patients who initially show lengthy, durable responses to ibrutinib often acquire resistance and relapse at a median of 17 months. Once patients relapse after ibrutinib treatment, the 1-year survival rate is only 22% (28). Using the PDX model, we first established primary ibrutinib-resistant PDXs using tumor cells collected from PT8 and PT10, who were patients with relapsed/refractory MCL who had clinical primary resistance to ibrutinib. To identify regimens that can be potentially utilized to overcome primary ibrutinib resistance, we performed cell viability assays in vitro that showed that the tumor cells from PT8 PDX-G2 and PT8 PDX-G3, as well as the original primary tumor cells, were resistant to ibrutinib but sensitive to the proteasome inhibitor carfilzomib. Next, the in vivo data in PT8 PDX-G3 validated ibrutinib resistance and carfilzomib activity, demonstrating that the PDX model can be employed to identify regimens to treat therapeutic resistance.

We also established acquired ibrutinib-resistant PDXs by administering ibrutinib daily to G3 and G4 mice established using an ibrutinib-sensitive patient sample. In the acquired ibrutinib-resistant PDXs, PLC-γ2, p65, and Src were down-regulated; however, BCL-2 family members and a PI3K signaling component were up-regulated. Both BTK and PI3K are involved in proximal BCR signaling, and once the BTK-mediated effect became inactive due to desensitization to ibrutinib, the signals mediated by the PI3K pathway may possibly promote growth and survival (29). Finally, our in vivo data showed that the combination of ibrutinib with idelalisib, as well as with its combination with carfilzomib, resulted in halting tumor growth in vivo. The results suggest that PDXs can be used as a translational model to explore alternative therapies and drug combinations in the context of acquired drug resistance. These findings support the results observed in ibrutinib-resistant chronic lymphocytic leukemia (CLL; ref. 30).

We also created an ibrutinib-resistant PDX mouse model using the clinically acquired resistant PT15 tumor cells. The preclinical data obtained in this model guided the therapy of the patient, dramatically reducing the patient's lymphocytosis. Furthermore, we calculated the mean passage time per generation across five generations for 10 B-cell lymphoma PDX mouse models and found that the mean passage time of 3 of 10 models was less than 1 month and the mean time ranged from 30 to 40 days for 4 models (Supplementary Table S4). These data support the use of PDX mouse models to provide personalized therapy to individual patients as previously reported by Hidalgo and colleagues (24). Hidalgo and colleagues reported a pilot study in which treatments for patients with advanced solid tumors were selected based on the activity of novel agents against the corresponding PDX model. They observed a response rate of 88% for treatment(s) deemed effective by the model that were subsequently chosen for the patients (24). In addition, Stebbing and colleagues demonstrated the effectiveness of developing personalized therapies for rare tumors such as sarcomas. A correlation between the PDX results and clinical outcome was observed in 13 of 16 (81%) sarcoma patients, with no patients progressing during the PDX-predicted therapy (31). The previous reports along with our results discussed here strongly indicate that B-cell lymphoma PDXs can be used as a personalized therapy platform.

In this study, PDXs identified novel treatment choices that overcame drug resistance. Furthermore, the correlations between drug resistance and effective treatment responses may ultimately help identify biomarkers that can potentially predict effective treatment outcomes, ultimately personalizing therapy for B-cell lymphoma patients.

S. Kadri is a consultant/advisory board member for GLG consulting. No potential conflicts of interest were disclosed by the other authors.

Conception and design: L. Zhang, K. Nomie, Jack Wang, Jacqueline Wang, L. Li, Q. Yi, M. Wang

Development of methodology: L. Zhang, K. Nomie, H. Zhang, W. Tao, Jacqueline Wang, M. Wang

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): L. Zhang, H. Zhang, L. Pham, S. Li, D. Santos, W. Chen, O. Oriabure, Y. Liu, H. Guo, C. Li, Jack, Jacqueline Wang, S. Thirumurthi, S.Y. Huang, Y. Shao, Y.L. Wang, M. Wang

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): L. Zhang, K. Nomie, H. Zhang, L. Pham, S. Kadri, J. Segal, S. Zhou, S. Sharma, Jacqueline Wang, Y.L. Wang, M. Wang

Writing, review, and/or revision of the manuscript: L. Zhang, K. Nomie, T. Bell, S. Li, S. Zhou, S. Huang, B. Fang, Jacqueline Wang, M. Ahmed, S.Y. Huang, Y.L. Wang, M. Wang

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): L. Zhang, T. Bell, O. Oriabure, S. Huang, Z. Chen, M. Badillo, L. Lam, Y.L. Wang, M. Wang

Study supervision: L. Zhang, K. Nomie, B. Fang, M. Wang

Other (pathology review and picture taken): S. Li

Other (collection of tissue specimens): D. Santos

Other (ran all the next-generation sequencing (NGS) experiments for the samples using the hybrid capture panel and also helped in the NGS data interpretation): S. Sharma

We would like to thank the Kinder Foundation Research Fund and the Garfield Mantle Cell Lymphoma Research Fund for their philanthropy.

This work was supported by The University of Texas MD Anderson Moon Shot Fund (to M. Wang, L. Zhang); Cancer Center Support Grant (CCSG; P30 CA016672, DePinho) and the National Cancer Institute (R21 CA202104; to M. Wang). This work was also supported by the generous donations made to the MD Anderson Cancer Center Mantle Cell Lymphoma Program of Excellence.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.