Patient-derived xenografts of low-grade B-cell lymphomas demonstrate roles of the tumor microenvironment

The clinical behavior of mature B-cell lymphomas reflects the properties of both the tumor environment (TE) and neoplastic cells.^1,2  For example, observational studies of human specimens have shown relationships between features of the nonneoplastic immune cells and prognosis.^3,4  To dissect the effects of the TE from those intrinsic to neoplastic cells, we used a xenograft system for B-cell non-Hodgkin lymphomas (NHL), focusing on follicular lymphoma (FL) and marginal zone lymphoma (MZL) because the TE is particularly well studied and clearly relevant in these diseases.¹  We reasoned that if the TE was largely self-organizing and the properties of the neoplastic cells were largely cell-autonomous, then the xenograft would retain many of the native properties seen in the patient. On the contrary, if the TE is dependent on the systemic environment of the host to maintain its tumor-associated functions and the properties of the neoplastic cells are responsive to environmental cues, then properties of the neoplastic cells might be distinctive in the xenograft setting. Therefore, a xenograft model could allow us to test a basic question in tumor biology: How dependent are the properties of neoplastic cells on environmental cues?

A robust system for xenografting NHL specimens is critical to our studies. Although genetic models of low-grade NHL exist, these cannot reproduce the interpatient variation in the TE that originally allowed Dave and colleagues to demonstrate the pivotal role of the TE in prognosis of FL.³  Therefore, we sought an approach to xenografting NHL specimens such that patient-specific components of the TE could be systematically studied. Host NOD.Cg-Prkdc^scidIl2rg^tm1Wjl/SzJ (NSG) mice lack B, T, and NK cells and have proven robust for creating xenografts of leukemia and lymphoma.⁵  Here we use the omental tumor xenograft (OTX) model, which supports rapid and preferential engraftment of both tumor cells and tumor-associated nonmalignant cells (the TE) into an anatomically well-defined site, the greater omentum, allowing for quantitative measurements of the grafts. Using nondisrupted pieces of human non-small cell lung, breast, and ovarian carcinomas, we have previously demonstrated long-term engraftment in the murine omentum of tumor-derived memory T cells, B cells, plasma cells, dendritic cells, and monocytes.^6,7  An adipose-rich, highly vascularized structure, the omentum is a key organ for lymphocyte trafficking to and from the peritoneal cavity, potentially accounting for the preferential engraftment of human immune cells at this site.⁸  Here, we apply the OTX system to single-cell suspensions of NHL specimens, allowing separate implantation of the neoplastic cells and the tumor-associated nonneoplastic components. The data reveal distinctive growth and differentiation potentials of individual patients’ lymphomas, allowing the recognition of aspects of these neoplasms that are modulated by their environment.

Tumor collection and banking was performed as previously described and approved by the various institutional research subjects’ review boards,⁹  in accordance with the Declaration of Helsinki. Histology and immunohistochemical staining used standard techniques. A magnetic bead system (LS columns MidiMacs; BD Biosciences, San Jose, CA) was used to positively select CD4 (BW135/80 Fab) and CD8 (M-T466) cells and to negatively select B cells (using a cocktail of biotin-conjugated antibodies against CD2, CD3, CD4, CD14, CD15, CD16, CD56, CD61, CD235a, FcεRI, CD34, and Anti-Biotin MicroBeads [“B-CLL” kit]; Miltenyi Biotec Inc., Auburn, CA). Viability was performed by 7-aminoactinomycin D exclusion. For qualitative studies, DNA was extracted from sections of formalin-fixed paraffin-embedded tissue (range, 4-25 microns, depending on the size of the graft) or ∼5 × 10⁶ cryopreserved cells (tumor inputs). For quantitative studies, DNA was extracted from the fresh mesenteric organs obtained at necropsy.

OTX model: 5 × 10⁶ to 10 × 10⁶ viable cells (suspended in 500 μL RPMI-1640) were injected into the peritoneum (intraperitoneally; i.p.) of young adult male and female NSG mice. In preliminary experiments, engraftment was assessed at several intervals (range, 2-60 days); 14 days was selected as the standard time for assessment of engraftment. Engraftment was assessed by harvesting the intra-abdominal organs; the omentum (often with associated variable amounts of pancreas, spleen, bowel, and liver) and the spleen were separately prepared for histology or other quantitative studies. Histologic blocks containing omentum were serially sectioned to exhaustion (median, 240 sections/block); selected sections were hematoxylin and eosin stained or used for immunohistochemistry.

Various molecular approaches were used to confirm that each xenograft contained the neoplastic B cells specific to the tumor that had been implanted (Table 1). Clonality assays based on the IGHV and IGKV used commercially available systems (IdentiClone Assay; InVivoScribe Technologies, Inc., San Diego, CA). The clonal immunoglobulin heavy chain (IGH) sequence was determined using IGVH family primers, pooled into 3 sets, coupled with a single downstream primer beyond the IGHJ genes (see supplemental Methods). IG sequences were analyzed using the International ImMunoGeneTics information system (http://www.imgt.org/). The t(14;18) translocation junction was identified by polymerase chain reaction (PCR), as previously described.¹⁰ 

Amplicon-based libraries were produced using Q5 polymerase (NEB, Ipswich, MA) and 8N bar-coded primers containing the Illumina S16 overhang adapter sequences followed by germline specific sequences (supplemental Methods). Family-specific IGH primers were designed for the IGHV FR2 and IGHJ FR4 regions. The multiplex IGH PCRs used 3 different primer pools, each with ∼150 μg DNA per reaction. Two rounds of PCR generated amplicons with bar codes on each end; bar-coded primers were removed (ExoI digestion, 20 U/50 μL; 37°C; 1 hour; NEB) followed by heat inactivation (20 min; 80°C). Bar-coded products were amplified (primers to the Illumina adapters; 5 cycles) and size selected (SPRIselect beads; Beckman Coulter, Inc., Brea, CA), keeping fragments 200-1000 nt. Indexed products larger than 250 nt were selected from a final amplification (15 cycles; Illumina adapter sequences; SPRIselect). See supplemental Table 3 for thermocycler settings. Sequencing was performed by the University of Rochester Genomic Research Center (Illumina MiSeq, V3 chemistry, 300 nt paired-end reads).

Illumina 2 × 301 paired-end read data were processed to remove adapter sequences and low-quality bases and merged into single end-to-end reads using AdapterRemoval v2.¹¹  Read data with identical bar codes were processed to generate a consensus read, followed by bar-code removal. Read ends were compared with primer sequences to assign reads to a primer pair, provided there was a primer sequence match greater than 22/25 nt. Reads generated with identical primer pairs were clustered and counted, using Needleman-Wunsch scores¹²  as a similarity metric to produce a smaller collection of clonal sequences. The presumed IGH sequences were batch processed, using IMGT High-VQUEST (https://www.imgt.org/HighV-QUEST/login.action) to determine whether the sequence represents a productive rearrangement and to obtain its CDR3 amino acid sequence. These amino acid sequences were then clustered and counted, based on Needleman-Wunsch scores, to identify clonally related sequences and determine their frequencies in the sample.

Studies of more than 1000 mice using cell suspensions from more than 50 different primary lymphomas, both fresh and in viably frozen formats, suggested the OTX approach using frozen cells was effective. To objectively assess the efficacy of the OTX model for tumor engraftment in an unbiased manner, we selected 20 NHL specimens from a pool of more than 1000 viably frozen, single-cell suspensions of nonchronic lymphocytic leukemia NHL tumors (Table 1). These cases were selected as the 20 B-cell NHL specimens in our tumor bank with the largest number of aliquots; no other selection criteria were used. The 20 cryopreserved B-cell lymphoma suspensions were each injected i.p. into 3 mice for the survey study, and omental-associated organs of each mouse were subjected to exhaustive histologic review (see “Methods”). 17 specimens showed histologic evidence of engraftment in the omentum at 14 days in all 3 of the mice used for each specimen. Three of the 20 samples (tumor O, P, and K with viabilities of 5%-10% of the thawed cell suspension) failed to engraft in any mice; no other parameter affecting engraftment, including the number of aliquots available, was identified. Engraftment was defined as morphology- and immunohistochemical-based detection of human lymphocytes (note that the NSG strain has no endogenous lymphocytes, a feature that greatly simplifies scoring of engraftment).

For the 17 specimens that engrafted, at least 1 clone-specific molecular feature of the original tumor was identified in the corresponding graft (Table 1). Identity of clone-specific IGHV or IGKV was established through PCR-based clonality assays. Additional approaches to corroborate clonal identity of the graft with the original tumor included identification of the BCL2:IGH translocation breakpoint and the presence of BCL1 overexpression (as an indicator of the BCL1:IGH translocation in a single case of mantle cell lymphoma).

Although the time course and characteristics of the grafts depended substantially on the specific source specimen, there were several generalizable features. Of the 19 specimens that were viable at thaw, 17 engrafted in all 3 mice tested for each specimen. The omental xenografts were microscopic or just visible at 2 weeks and retained cytologic features of the original neoplasm (Figure 1). All 17 xenografts contained both B and T cells in variable proportions (Figure 1, columns 3 and 4). In every case, specimens that implanted in the omentum also involved the splenic white pulp at 14 days (eg, see Figure 1, specimen C, columns 2, 3, and 4). In some tumors (eg, Figure 1, specimen G Lung, column 3), B cells were abundant at other distant sites by 1 month. Engraftment did not appear to reflect Epstein-Barr virus-mediated immortalization: 16/17 xenografts were negative for the 2 early Epstein-Barr virus transcripts, EBER 1 and EBER 2. All xenografts of tumor F were floridly positive for Epstein-Barr virus; only rare positive cells were present in the original tumor specimen (data not shown).

Extensive immunohistochemical studies showed no definitive evidence of human macrophages (as marked by CD68 or CD163) or dendritic cells (marked by CD21 or CD23) at any time in the xenografts, despite their relative abundance in many of the original specimens; it is unclear whether this reflects a failure of these cells to survive processing (eg, disaggregation, filtering, cryopreservation, or thaw) or a failure of these cell types to engraft at the sites examined. In all engrafted specimens, the ratio of T cells to B cells was higher in the xenograft compared with original tumor (data not shown). The preponderance of T cells were CD4, rather than CD8, with only sparse CD56-, CD57-, FOXP1-, or PD1-positive cells detected in a subset of specimens (data not shown).

Engraftment was entirely dependent on nonneoplastic cells (Figure 2). Aliquots of tumors J (FL) and W (MZL) were depleted of non-B-hematopoietic cells, using a cocktail of antibodies (CD2, CD14, CD16, CD36, CD43, CD235a). Exhaustive sectioning of the omentums and thorough necropsy of the animals injected with samples consisting only of B cells showed no evidence of engraftment. In contrast, mock-depleted aliquots engrafted similarly to the nonmanipulated specimens. The ability to engraft was reconstituted by adding back the nonneoplastic components (Figure 2D).

To define the components of the TE required for engraftment, we performed various flow-based cell sorting studies to selectively deplete or augment the T follicular helper and T regulatory compartments. In extensive studies of specimens J and W, no discernable effect on engraftment was seen (data not shown). Because depletion of the defined subtypes of T cells did not discernably alter engraftment, we depleted either all CD4- or CD8-positive cells (Figure 3). Flow cytometry before i.p. injection indicated that all CD4-positive and CD8-positive cells in these specimens were CD3⁺. For 4 of 5 specimens studied, engraftment of tumor-derived B cells was entirely dependent on the presence of CD4-positive cells (Figure 3 shows representative data for tumor W). Specimen E showed marked attenuation of B-cell engraftment on CD4⁺ cell depletion (data not shown). Depletion of CD8-positive cells had no discernable effect on the engraftment of any of the 4 specimens tested (J, W, M, L).

We hypothesized that the composition of the TE has been selected through the process of tumorigenesis to provide supportive cues to which neoplastic B cells (compared with nonneoplastic B cells) are especially sensitive. To test this hypothesis, we exploited the consistent presence of nonneoplastic B cells in the tumors. The nonneoplastic B cells presumably represent cells that transit the tumor but are not intrinsic to the neoplastic process. If the TE preferentially supports the neoplastic B cells, a time-dependent reduction in the relative number of nonneoplastic B cells is predicted. At the 2-week point, all grafts contained substantial numbers of CD20-positive B cells. Contrary to expectations, the fraction of nonneoplastic B cells increased for all specimens tested (Figure 4). For 2 MZL specimens (W and M; each with 1%-2% of B cells at thaw identified as nonclonal; data not shown), the fraction of B cells that were nonclonal showed a modest increase, but the clonal B cells remained greater than 90% of all B cells by 2 weeks. In contrast, xenografts established with 2 FL samples (J and L; each tumor with about 20% of B cells at thaw identified as nonclonal by flow cytometry) showed a substantial relative loss of neoplastic B cells by 2 weeks.

In the course of these studies, we observed a consistent accumulation of plasma cells in all grafts, regardless of the original histology of the tumor. The plasma cell fraction (marked by CD138), initially undetectable by our methods, increased in the xenografts, and the mature B-cell (CD20 positive) fraction decreased. Although this overall accumulation of plasma cells was consistent across all specimens, the time course of accumulation and evidence of clonal restriction varied with the specific specimen. The evolution of the xenograft composition was studied in detail using tumors J (FL) and W (MZL) (Figures 5 and 6). The T-cell fraction and Ki67 index were increased at day 7 compared with that seen at day 1, and especially compared with the original tumor. In replicate experiments, tumor J (FL) showed an abrupt loss of CD20 expression and gain of CD138 expression between days 7 and 14, whereas tumor W (MZL) showed a similarly consistent increase in cells expressing CD138 between days 14 and 21. The plasma cells that emerged in the xenografts of 4 specimens were light chain restricted, suggesting the plasma cells were directly derived from the neoplastic B cells (Figure 7 contrasts the light chain restriction of specimens J and W). Definitive monoclonality (light chain restriction) of the plasma cells was seen in 2/4 MZLs (specimens M and W), 1/7 FLs (specimen E; data not shown), and the single mantle cell lymphoma (specimen C; data not shown). All other xenografts showed a mixture of κ- and λ-expressing plasma cells, and therefore could be entirely polyclonal or a mixture of clonally derived and polyclonal plasma cells (Table 1). Note that there is strong correlation between the presence of overtly clonal plasma cell proliferations and the lack of nonclonal B cells in the original diagnostic specimen (Wilcoxon rank-sum, P < .001); for example, the 4 specimens (W, M, C, E) with the least polyclonal B cells in the original diagnostic specimen are the 4 with overtly clonal plasma cells. These data suggest that in the environment of the xenograft, both neoplastic and nonneoplastic B cells undergo plasmacytic differentiation, indicating that the differentiation state of neoplastic B cells, similar to nonneoplastic B cells, is responsive to environmental cues, and is not solely determined by cell-autonomous, fixed genetic, or epigenetic factors.

We used a robust and quantifiable xenograft system for NHLs to identify the properties of the neoplastic cells that are influenced by the TE. The robust nature of the OTX model allowed us to observe patient-specific variation in the dependence on the TE: in some tumors (J, L), the nonneoplastic B cells are responsive to the proliferation and survival cues present in the xenograft environment, whereas the neoplastic B cells are not; in other tumors (M,W), the neoplastic B cells engrafted effectively. The relative failure of neoplastic cells from tumors J and L to engraft compared with nonneoplastic B cells in the same specimens is all the more surprising, given that these neoplastic cells are positive for the IGH:BCL2 translocation, which presumably would make these cells relatively resistant to apoptosis. Given the robust nature of the model and the internal control comprised of passenger, benign B cells, these negative data for tumors J and L suggest there is some critical environmental factor (whether cellular, humoral, or otherwise) present in the patient that did not survive transfer to the mouse and that the factor or factors is critical to the expansion of neoplastic, but not nonneoplastic, B cells. Furthermore, uniform differentiation of the neoplastic B cells into plasma cells in several tumors (M, W, C, E) indicates the differentiation of the neoplastic cells is responsive to environmental cues, findings that parallel those of a recent similar xenograft study of chronic lymphocytic leukemia.¹³  Therefore, this PDX approach identifies specific patients as having tumors in which the survival and/or the differentiation of the neoplastic cells is dependent on the TE.

We find that the differentiation of neoplastic cells is not cell autonomous, an observation that is consistent with well-described histologic features of some B-cell tumors. MZL, mantle cell lymphomas, and even FLs may be associated in patients with significant numbers of plasma cells that are derived from the neoplastic B cells.^14-16  In the case of MZL, these plasma cell accumulations may be so marked that the neoplasm mimics a plasmacytoma or creates an amyloidoma. More subtle accumulations of clonal plasma cells occur in MZLs arising at mucosal sites: the clonal plasma cells are enriched at the most superficial sites, visually suggesting that the environment is modulating either the differentiation of the tumor cells or that cells with plasmacytic differentiation preferentially traffic to specific locations. In the OTX system, all 17 original cell suspensions lacked significant numbers of plasma cells, whereas the grafts showed a time-dependent accumulation of plasma cells; in 4 tumors (M, W, C, E), the plasma cell population was definitively clonal, and in another 13 tumors, the plasma cell population was substantially or perhaps entirely derived from nonneoplastic B cells. Similarly, chronic lymphoma leukemia specimens uniformly undergo plasma cell differentiation in a T-cell-dependent xenograft system, despite the relative rarity of plasma cell differentiation in patient samples of this tumor.¹³  In summary, our data suggest that the plasma cells associated with some NHLs in patients are a result of a microenvironment-induced differentiation of neoplastic B cells, and more generally, that the microenvironment modulates the differentiation to plasma cells in a wide range of mature B-cell neoplasms.

The reliability of the OTX model creates opportunities to perform “personalized” studies of the role of the tumor microenvironment in tumor biology and therapeutic response. Xenograft models of NHL have been typically limited to higher-grade types of NHL in which the TE effects are limited.¹  Our work also shows that the OTX model is practical: 17/20 unselected primary specimens engrafted in some manner; 13 of which are lower-grade forms (mantle, marginal, and follicular); engraftment was rapid (histologically detectable at 1-2 weeks) and quantifiable; and mice were viable with no deaths from disease burden for at least 1 month, allowing time for experimental interventions. Because primary tumor specimens are typically the limiting resource for the production of xenografts, much attention has been focused on the development of the genetic mouse models of lymphoma^17-21  or xenograft systems that allow for serial transplant.^5,22  Despite the utility of both these approaches, neither captures patient-specific, environmental features. Our approach contrasts particularly with the PDX models that allow serial transplant; it is unclear whether these serial transplant models represent selected tumors and to what extent these serially transplanted tumors are dependent on the host or tumor microenvironment. In contrast, we show that the OTX is robust even when used for unselected low-grade NHL specimens.

Our data are consistent with prior work indicating a key role for CD4⁺ T cells in the survival of normal and malignant cells both in culture and, in the case of chronic lymphocytic leukemia, on transplant.^13,23,24  In contrast to prior studies that have emphasized the efficacy of CD4⁺ T cells to support primary follicular lymphoma specimens in culture,²⁵  we found that in some cases (J, L), tumor-derived CD4⁺ preferentially supported engraftment of nonneoplastic as opposed to neoplastic B cells. In addition, we found no definitive evidence of an effect of removing and supplementing specific T-cell subsets (T follicular helper and T regulatory cells). We did not examine the dependence on macrophages or dendritic cells because the flow cytometry performed at the time of diagnosis did not contain the relevant markers, and subsequent, research grade, flow studies (at thaw) of selected specimens did not detect these populations. Therefore, our data do not exclude that these cell types could support engraftment in the absence of CD4⁺ T cells.

Given the time-dependent changes in the tumor composition for the low-grade lymphomas, we did not systematically attempt to develop long-term xenografts or to serially transplant any lymphomas. At times greater than 30 days, xenografts of 37 lymphoma specimens (largely distinct from those described here) often showed a proliferation of T cells at distant sites suggestive of graft-versus-host disease (data not shown). Analogous to the recent study of chronic lymphocytic leukemia xenografts by Patten et al, some mice appeared to lose the tumor graft.¹³  Our preliminary observations at points beyond a month suggest the engraftment process for low-grade NHLs is inherently unpredictable, as the same tumor specimen could either entirely regress or result in an aggressive process similar to graft-versus-host disease. Regardless, the patient-specific behaviors of the xenografts in the OTX system were highly predictable for times up to 1 month, allowing the OTX model to be used for quantitative assessment of experimental manipulations.

To successfully exploit the OTX model, there are several features potential users should bear in mind. First, although each tumor specimen had a characteristic time course for histologically detectable engraftment (ranging from earliest point examined at 2 days-1 week), all showed visible omental masses by 2 weeks, although some tumors required injection of greater numbers of cells to form a visible mass by 2 weeks (data not shown). In practice, when working with primary specimens, the number of cells needed likely reflects details of the harvest and preparation of individual specimens, which cannot be controlled in a clinical setting. Regardless, the tumor size and composition for the xenografts from each source showed a highly consistent time course across mice up to at least 1 month, allowing quantitative experiments once these parameters are established. Second, a process histologically resembling graft-versus-host disease occurred by 4 weeks in a subset of tumors, occasionally becoming severe by 6 to 8 weeks. This T-cell proliferation, presumably driven by xeno-antigens, likely alters the T-cell composition and activity from that seen in the patient. Furthermore, this T-cell proliferation can be profound and may obscure the neoplastic cells. Third, some components of the TE were not clearly reestablished in the xenografts. For example, no overtly nodular structures were seen and no evidence was detected of follicular dendritic cell networks in any of the FL xenografts, using routine staining for CD21 and CD23. Regardless, the reliability and relative ease of the OTX approach will allow investigators to assess the several parameters needed to optimize the approach.

The ability to reliably xenograft NHL specimens led us to observations regarding what properties are intrinsic to the neoplastic cells, as opposed to those that are modulated by the environment. Given the ability of the xenograft to support nonmalignant B-cell survival in the same specimens, the failure of the FL B cells to survive [despite the presence of t(14;18) and the resultant overexpression of BCL2] indicates that additional environmental factors beyond CD4⁺ T-cells must be responsible for maintaining FL. The propensity of neoplastic B cells to undergo plasmacytic differentiation indicates that the differentiation of the neoplastic cells are substantially the result of environmental cues. Although much recent attention has focused on how the TE may modify tumor-directed immunity, our observations underscore the potential roles of the TE in directly modulating the properties of tumor cells.

This study was supported by the Lymphoma Research Foundation 285230 (W.R.B.) and National Institutes of Health, National Cancer Institute Cancer Core grant CA034196 to the Jackson Laboratory (L.D.S.).

Contribution: W.R.B. designed research, performed research, analyzed data, and wrote the paper; G.N.S., P.J.R., and S.A.S. performed research; J.P.S. analyzed data; J.M.S. designed and performed research and analyzed data; L.D.S. contributed vital reagents and designed research; and R.B.B. and S.H.B. designed research.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: W. Richard Burack, University of Rochester, 601 Elmwood Ave, Box 626, Rochester, NY 14642; e-mail: richard_burack@urmc.rochester.edu.