A human lymphoma organoid model for evaluating and targeting the follicular lymphoma tumor immune microenvironment

SUMMARY

Heterogeneity in the tumor microenvironment (TME) of Follicular Lymphomas (FL) can affect clinical outcomes. Current immunotherapeutic strategies, including antibody- and cell-based therapies, variably overcome pro-tumorigenic mechanisms for sustained disease control. Modeling the intact FL TME, with its native, syngeneic tumor-infiltrating leukocytes, is a major challenge. Here, we describe an organoid culture method for cultivating patient-derived lymphoma organoids (PDLOs), which include cells from the native FL TME. We define the robustness of this method by successfully culturing cryopreserved FL specimens from diverse patients and demonstrate the stability of TME cellular composition, tumor somatic mutations, gene expression profiles, and B/T cell receptor dynamics over 3 weeks. PDLOs treated with CD19:CD3 and CD20:CD3 therapeutic bispecific antibodies showed B cell killing and T cell activation. This stable system offers a robust platform for advancing precision medicine efforts in FL through patient-specific modeling, high-throughput screening, TME signature identification, and treatment response evaluation.

eTOC Blurb

Using primary follicular lymphoma tumor biopsies, Kastenschmidt et al develop a patient-derived lymphoma organoid model, demonstrating in vitro microenvironment stability over 3 weeks without exogenous cytokines. Treated with bispecific immunotherapies, organoids recapitulated T-cell mediated lymphoma killing, allowing investigation of patient-specific microenvironment determinants of response.

Graphical Abstract

INTRODUCTION

Follicular lymphoma (FL), the most common indolent lymphoma, is clinically and genetically heterogeneous. Although FL is considered incurable in most patients with current therapies, progression-free survival on average exceeds 10 years¹. Nevertheless, a subset of FL patients that experience early relapses have substantially higher morbidity and mortality². Immune evasion is a critical aspect of FL persistence and progression³ and the waxing and waning nature of untreated disease suggests dynamic involvement of anti-tumor immunological mechanisms^4,5. Multiple interactions between malignant FL clones and other cell subsets in the tumor immune microenvironment (TME) have been identified as either involved in lymphoma maintenance^6,7 or correlated with clinical outcome⁸. The ability to detect and manipulate such dynamic interactions has become increasingly important with the emergence of novel lymphoma immunotherapies, including bispecific T cell-engaging antibodies (bispecific antibodies)^9–12 and cellular therapies including chimeric antigen receptor bearing T cells (CAR-T)^13,14. The immune effector functions of these diverse therapies require an intact TME and are therefore difficult to model in traditional ex vivo B cell monocultures.

Organoid systems derived from non-lymphoid organs including lung, intestine, and brain show great potential for more accurately representing the TME of human tissues¹⁵. Due to the complexity and diversity of immune cell developmental requirements, it has been challenging to generate organoids that recapitulate immune TMEs using conventional organoid techniques. Prior studies over the last decade have made key contributions and improvements to in vitro and in vivo bioengineering techniques to culture immune cells with their corresponding TME for lymphoma^16–21 but essential limitations remain an issue, including dependence on cell lines as sources of lymphoma or supportive cells, unclear potential for long-term culture longevity, analysis of exclusively murine or xenograft lymphoid tissues, and reliance on exogenous cytokines. Previous attempts to model immune cells and the TME are thus limited in their capacity to recapitulate human physiology and have yet to aid in improving therapeutic translation from pre-clinical models. Further, the inherent variability of both human adaptive immune responses and FL disease on a per-patient basis each add yet another layer of complexity when designing one-size-fits-all therapies.

A robust in vitro, high-throughput model that enables stable co-culture of primary patient-derived malignant lymphoid cells within their native TME could accelerate our understanding of FL biology, including personalized response assessment. We recently demonstrated that lymphoid tissue organoids prepared from non-malignant primary human tonsils, spleens, and lymph nodes can be used to evaluate key germinal center features, including immunoglobulin class-switch recombination and somatic hypermutation, as well as to characterize individual adaptive immune responses to in vitro vaccination against diverse viral pathogens^22,23. In this study, we adapted the platform for the creation of lymphoid tissue organoids derived from primary human follicular lymphoma biopsies and used corresponding patient-derived lymphoma organoids (PDLOs) to dissect the mechanistic underpinnings of novel immunotherapies including CD3:CD19 and CD3:CD20 bispecific antibody treatment.

RESULTS

Patient-derived lymphoma organoids as a platform for FL TME propagation.

To develop and validate the PDLO model system for FL, we studied primary biopsies from 12 FL patients at representative oncological milestones that capture the natural progression of the disease (Methods S1). Specifically, we included excisional surgical biopsies of primarily nodal tissues obtained at initial FL diagnosis (n=3), at disease progression following initial expectant observation (i.e., watchful waiting; n=3), at disease recurrence after prior systemic therapy (n=3), or at aggressive histologic transformation to diffuse large B cell lymphoma (transformed FL/tFL; n=3). Consistent with typical initial disease presentations, the 12 enrolled patients had predominantly advanced disease at initial staging (84%) and a median patient age of 60 years (range 44–84) at tissue biopsy. Specimens were processed under sterile conditions, then gently mechanically dissociated into single cell suspensions, cryopreserved, and PDLOs later generated on a high throughput 96-well platform (STAR Methods). Longitudinal organoid stability and TME characteristics were then evaluated using a multi-modal strategy that integrated flow cytometry immunophenotyping, bulk RNA sequencing, and DNA CAPP-seq²⁴. The combined analysis allowed us to define donor-specific cellular composition, transcriptomic changes, somatic mutational signatures, clonal dynamics, and changes in B and T cell immune receptor repertoires (Figure 1A).

Figure 1. Patient-derived lymphoma organoids are a robust and stable in vitro system of the follicular lymphoma tumor microenvironment.

(A) Biopsies from lymphoma patients were isolated, dissociated, and cryopreserved. Organoids were generated and grown for up to 21 days in culture. Comprehensive immune profiling and longitudinal analyses of PDLOs were performed on days 0, 7, 14, and 21. Assays included flow cytometry, DNA mutational profiling (CAPP-seq), and RNA gene expression with B/T cell receptor clonal dynamics. (B) Representative bright field image of organoids generated from a healthy tonsil tissue donor and PDLO on day 5 of culture. (C) Flow cytometry quantification of total live cells in PDLOs over time. (D) Quantification and (E) representative gating of FL cells in each PDLO over time. Lymphoma cells were defined as live, CD3- cells with clonally restricted BCR light chains as defined clinically (Table S1). (F-I) Quantification and characterization of (F) total T cells and subsets including (G) CD8+, (H) CD4+, and (I) Tfh (CD4+CXCR5+PD-1+ cells).

Patient-derived organoids are viable and highly stable in culture.

First, we validated that the PDLO system adequately sustained both FL cells and the TME. Using bright field microscopy, we found that PDLOs formed aggregate clusters similar to those we previously observed in healthy lymphoid organoids from human tonsils²² (Figure 1B). Overall, PDLOs remained viable in culture until the end of the experiment on day 21 (Figure 1C). We next assessed the initial representation and longitudinal stability of malignant lymphoma cells in the PDLOs, which were defined as CD3-negative cells with restricted B cell receptor (BCR) light chain isotypes that matched the patients’ clinical tumor immunophenotype (Figure 1 D–E, Supplemental Figure 1A–B, Supplemental Table 1). We observed substantial variability in the cellular composition of each PDLO, as might be expected given clinical variation in tumor and TME composition. Nevertheless, the proportion of malignant cells was largely maintained in each PDLO over time and was well-matched with the malignant B cell burden in each primary tumor, as quantified on day 0 (median FL cells of all donors over time: 35.6 +/− 4.45, coefficient of variation (CV) 70.0%) (Figure 1D and Supplemental Figure 1B–C). We next evaluated the non-B cell populations direct ex vivo and longitudinally in PDLOs. Dendritic cells (classical and plasmacytoid), NK cells and monocytes/macrophages using flow cytometry were found in very small quantities (<1% of live cells, data not shown) on day 0 and therefore excluded from further analyses. Most non-malignant cells in the FL TME were T cell populations. While total T cell proportions varied between tumors, their representation within individual PDLOs was stable in composition and frequency over time (Figure 1F). Similarly, we found that both CD4 and CD8 T cell compartments were well-maintained during PDLO culture (Figure 1G–H). When focusing on T follicular helper (Tfh) cells, previously described as a critical subset promoting immune evasion in the FL TME⁶, we verified that CXCR5+PD1+ Tfh populations remained stable in PDLO cultures (Figure 1I).

To more deeply characterize the cell phenotypes and transcriptional stability of PDLOs over time, we performed longitudinal bulk transcriptomic profiling and cell subset deconvolution with CIBERSORTx²⁴. Cell populations assessed using CIBERSORTx were well-correlated to flow cytometry measurements (Supplemental Figure 1D).

Tumor B cells in PDLOs display clonal and mutational stability.

To evaluate PDLOs for mutational stability and potential selection during culture, we used CAPP-Seq to assess missense somatic mutations pre-culture (day 0) and at weekly time points (days 7, 14, and 21). As expected from the mutational landscape of FL²⁵, we identified typical recurrent mutations in our cohort, including in KMT2D (50%), CREBBP (50%), BCL2 (67%), EZH2 (42%), TNFRSF14 (33%), STAT6 (25%), and TP53 (25%) (Supplemental Table 2). By comparing serial PDLOs to coding mutations detected at day 0, we determined that organoid mutational profiles remained stable across three weeks in culture (median fold change (FC) 0.94 +/− 0.028, CV 15.8%, Figure 2A). As an orthogonal measure of clonal stability, we evaluated the dominant B cell receptor (BCR) heavy- and light-chain rearrangements from RNA-seq data and compared them to the dominant BCR rearrangements identified at baseline. While the proportion of BCR-restricted B cells was variable between donors (consistent with differing levels of tumor burden from each patient’s biopsy), the proportion of tumor BCR clones was stable in most donors (median FC 0.91 +/− 0.17, CV 25.3%) at subsequent time points (Figure 2B, Supplemental Table 3). We next evaluated serial RNA gene expression data from days 7, 14, and 21 in culture. Unsupervised clustering of gene expression profiles revealed co-clustering of samples from individual patients, consistent with a stable and patient-specific TME composition (Figure 2C). We considered the association of mutational profile with stability in culture and found that while CREBBP mutations were significantly associated with stability (p=0.014), cases with TP53 mutations were significantly less stable (p=0.008, Figure 2D). In summary, our PDLO technique successfully supported the long-term culture of primary lymphoma cells with their native TME. Flow cytometry and molecular analyses demonstrated that tumor and non-tumor immune cells from the TME are phenotypically, mutationally, and transcriptionally stable over a three-week culture period.

Figure 2. PDLOs show mutational and BCR clonal stability over time.

(A) Sequential DNA sequencing demonstrates stability of coding mutations on days 0, 7, 14, and 21 in PDLOs. (B) The repertoire occupancy of the dominant BCR heavy and light chain clone for each donor is defined using RNA-seq. The dominant BCR clones represent the FL cells in each PDLO. (C) Unsupervised hierarchical clustering of gene expression profile similarity (Euclidian distance) demonstrates patient-specific co-clustering of sequential samples over 21 days in culture. (D) Comparison of per-gene fold change from day 0 to final timepoint demonstrates that while CREBBP mutations are associated with PDLO stability, cases bearing TP53 mutations are significantly less stable in culture. AF = Allelic Fraction.

PDLOs recapitulate bispecific immune engager therapeutic responses.

Having demonstrated the stability of PDLOs for both the malignant tumor B cells as well as their TME counterparts, we next assessed whether the platform could be used to replicate expected perturbations after T cell-dependent immunotherapy responses¹². On day 4 after organoid culture preparation, we added a CD19:CD3 bispecific antibody to each PDLO and assessed therapeutic responses compared to untreated PDLO controls. Where sufficient biopsy material was available, we compared the bispecific antibody treatment to the corresponding unconjugated anti-CD19 and anti-CD3 monoclonal antibodies as controls. We found that the unlinked antibodies had minimal effect within the PDLOs (Supplemental Figure 2A–B). Using a similar multimodal approach as in the initial PDLO characterization, we analyzed the effect of bispecific immunotherapy treatment in each PDLO (Figure 3A, Supplemental Figure 1A for gating strategy). By day 11 (7 days after bispecific antibody treatment), the CD3:CD19 bispecific antibody had induced a significant decrease (median FC 0.169 +/− 0.056, CV 67.40%, p=0.03) in lymphoma cells in each PDLO (Figure 3B and Supplemental Figure 2B).

Figure 3. CD3:CD19 bispecific antibodies induce FL cell killing in PDLOs.

(A) PDLOs were treated on day 4 with bispecific antibodies. A multimodal analysis approach was used to characterize the cellular response following treatment. (B) Evaluation of treatment response using flow cytometry to quantify total viable cells, viable lymphoma cells, and the fold change of lymphoma cells compared to untreated autologous PDLOs. (C) T cell proportions and (D) activation of CD4+ and CD8+ T cells in PDLOs following CD3:CD19 treatment. (E) Secreted cytokines in PDLO supernatants on day 7, measured using Luminex. (F) KEGG pathway gene sets enriched in CD4+ T cells from treated versus untreated PDLOs. Dot size indicates the count of enriched genes and color designates adjusted P-value. (G) The proportion of the BCR repertoire occupied by the largest BCR clone in treated and untreated PDLOs. * p<0.05 using paired Mann-Whitney U tests to compare groups. P values shown are for comparisons against the untreated control unless otherwise indicated by lines. ns= not significant; n=6.

As the bispecific antibody effector mechanism is T cell dependent, we assessed T cell population frequencies and activation phenotypes following CD3:CD19 antibody treatment. Further, as total live cells decreased following CD3:CD19, but not to the same magnitude as FL cells, we hypothesized that T cells were maintained or expanded following therapy. We found that total CD4+ and CD8+ T cell proportions were not altered by the treatment (Figure 3C). However, both CD8 and CD4 T cell populations were significantly activated (as measured by CD38 induction) after CD3:CD19 bispecific antibody treatment (Figure 3D). Correspondingly, secreted levels of Th1 cytokines (i.e., interferon gamma [IFNγ] and tumor necrosis factor alpha [TNF-α]) in culture supernatants at day 7 (3 days post treatment) were significantly higher in CD3:CD19 treated PDLOs compared to untreated (Figure 3E). Interestingly, the anti-inflammatory cytokine IL-10 was also increased following CD3:CD19 treatment (Figure 3E). In agreement with the flow cytometry data, CIBERSORTx analysis of bulk RNA sequencing data from PDLOs following bispecific therapy showed a decrease in B cells, associated with bispecific-induced killing (Supplemental Figure 3C). We also observed increases in the frequency of various T cell populations, including CD8 T cells, T follicular helper (Tfh) cells, and regulatory T cells as a proportion of total live cells (Supplemental Figure 3C).

We next assessed transcriptomic changes in individual PDLOs following CD3:CD19 treatment to define gene signatures and clonal dynamics associated with the bispecific antibody response. Gene set enrichment analysis of treated versus untreated PDLOs revealed upregulation of multiple pathways associated with T cell-mediated immune responses, including the MAPK pathway, PD-L1 pathway, and Th17 cell differentiation (Figure 3F). Considering BCR clonality as a specific marker of lymphoma B cells, the frequency of dominant monoclonal BCR clonotypes was significantly reduced following CD3:CD19 bispecific antibody treatment, demonstrating targeted bispecific-induced killing of lymphoma cells (Figure 3G). Together, these data demonstrate CD3:CD19 bispecific antibody treatment efficacy can be measured and characterized on a per-patient basis using PDLOs.

PDLOs reveal correlates of effective bispecific therapeutic response.

Given the diversity of PDLOs compared to B cell monocultures, a potential advantage of their use is the ability to evaluate correlates of response for individual patients. To assess another clinically relevant immunotherapy, we treated the same patients’ FL PDLOs with a CD20:CD3 bispecific antibody using the same workflow as in Figure 3A. While B cell killing was observed with the CD20:CD3 bispecific treatment, the magnitude of induced killing was reduced compared to analogous experiments with the same dose of the CD3:CD19 bispecific antibody over the same time course (Figure 4A). We further compared the fractions of lymphoma cells that were killed following treatment with CD3:19 and CD3:20 (Figure 4B) and found responses were both patient- and therapy-specific. We next sought to identify potential mechanisms explaining these divergent responses to similar bispecific antibody treatments. As with the CD3:CD19 antibody treatment, the CD3:CD20 bispecific antibody did not induce changes in CD8+ or CD4+ T cell proportions but did increase the activation of both CD8+ and CD4+ T cells (Figure 4C–4D). We found that overall, the CD3:CD19 bispecific antibody treatment was better able to induce CD4+ T cell activation compared to a CD3:CD20 bispecific antibody (Figure 4D). Indeed, the efficiency of tumor killing was better correlated with CD4+ T cell activation than with CD8+ T cell activation when all PDLOs were analyzed together, independent of bispecific treatment type (Figure 4E).

Figure 4. Cellular features and correlates of distinct bispecific therapy responses.

(A) Fold change of live lymphoma cells in PDLOs following CD3:CD19 or CD3:CD20 bispecific therapy 7 days post-treatment. (B) Comparison of the remaining fraction of lymphoma cells (fold change of live lymphoma cells) of CD3:CD19 and CD3:CD20 treated PDLOs. (C-D) The proportion and activation of (C) CD8+ and (D) CD4+ T cells following treatment with different bispecifics. Activated T cells were identified by CD38 expression using flow cytometry. (E) Correlation analysis of CD8+ and CD4+ activation and FL killing 7 days post-treatment. (F) Proportion of activated Tfh and non-Tfh CD4+ T cells 7 days after bispecific treatment. (G) Linear regression analysis of Tfh or non-Tfh activation and FL killing following bispecific treatment. (H) Gene expression of activation and exhaustion markers in PDLOs following treatment compared to control. Expression levels are shown as RNA transcripts per million (TPM). (I) Secreted cytokines in PDLO supernatants on day 7 measured using Luminex. (J) Correlation of day 0 cell frequencies of Tfh, non-Tfh, and CD8+ T cells compared to FL killing following treatments with CD3:CD19 and CD3:CD20 therapy. * p<0.05 using paired Mann-Whitney U tests to compare groups. P values shown are for comparisons against the untreated control unless otherwise indicated by lines. Spearman’s rank correlation coefficient was calculated for correlation analyses. n=6.

We next more deeply interrogated the CD4+ T cell compartment to understand their differing responses to each bispecific therapy. As with the total CD4+ T cell proportions, both Tfh and non-Tfh CD4+ T cell proportions were largely unchanged following treatment with either bispecific therapy (Supplemental Figure 3A–B). However, we did observe differences in Tfh versus non-Tfh activation states depending on the bispecific antibody used. CD3:CD19 induced significantly more Tfh and non-Tfh T cell activation compared to CD3:CD20 treatment (Figure 4F and Supplemental Figure 3C). We further identified a significant positive correlation between Tfh activation and tumor killing, whereas little association was observed with non-Tfh T cell activation (Figure 4G). Transcriptomic analyses further revealed that CD3:CD19 treatment substantially induced the expression of key immunological checkpoints, including TIGIT, LAG3, CTLA4, and PD1 (Figure 4H). The cytokines IFNγ, TNF-α, and IL-10 were also more elevated in CD3:CD19- treated culture supernatants compared to CD3:CD20 (Figure 4I). These results seem consistent with continuous T cell activation potentially inducing an exhaustion phenotype, as previously observed during continuous bispecific exposure²⁶.

The ability to characterize pre- and post-treatment phenotypes on an individual basis using the PDLO system allowed us to probe for features of the original tumor that may predict bispecific treatment response. We performed a linear regression analysis of cell frequencies direct ex vivo (day 0) and tumor killing (fold change of live lymphoma cells) following treatment with CD3:CD19 and CD3:CD20 (Figure 4J). We found that the proportion of remaining lymphoma cells was inversely correlated with day 0 Tfh proportions in the tumor following treatment with CD3:CD20, whereas little association was observed with CD3:CD19. Tumor killing was more strongly associated with the proportion of non-Tfh at baseline following CD3:CD20 treatment compared to CD3:CD19. Baseline CD8+ T cell proportions were also positively correlated with tumor killing, although a stronger association was found with CD3:CD20 than with CD3:CD19 treatment.

DISCUSSION

Substantial immunological perturbations in the tumor microenvironment (TME) underlie clinical heterogeneity and variability in FL outcomes. While several immunotherapeutic strategies are now cornerstones of FL management, these antibody- and cell-based therapies variably achieve durable disease control. Clinical and genomic heterogeneity remain important challenges in FL management, with inconsistent response rates in relapsed disease and few strategies to guide personalized treatment selection for individual patients.

The development of a robust ex vivo platform to evaluate patient-specific treatment responses is likely to have utility for understanding FL responses to candidate immunotherapies. Thus far, preclinical modeling of the intact FL TME as a cohesive endogenous unit has remained elusive, in part because many previous tumor culture strategies, including organoid-based methods, required artificial reconstitution to recapitulate the TME. Recent studies have advanced the field of B cell lymphoma organoid or co-culture systems^19,20,27, with each model having unique advantages and disadvantages. Organoids developed from cell lines or murine tissue offer advantages of scale and replicability, but may not recapitulate the full complexity of the lymphoma TME or inter-patient heterogeneity^16,17. Several recently developed systems have demonstrated utility for short-interval assessment of immune-directed therapies. Araujo-Ayala et al. integrated mantle cell lymphoma tumor biopsies with allogeneic myeloid cells to generate a 3D culture system and used this system to assess BTK and checkpoint inhibitor therapy¹⁹. Although the addition of a myeloid compartment is beneficial for modeling therapeutic response, the allogeneic source of these cells is problematic for long-term culture due to the potential for mixed leukocyte reactions. Similarly, a DLBCL tumor-derived system prepared by Shah et al. validated new potential therapeutic targets, but the use of mouse feeder cells inhibits the ability to test T cell-mediated therapies²⁷. As with prior systems^17,19,21, the utilization of exogenous cytokines or allogeneic cells offers precise control of culture conditions but may limit accurate representation of the endogenous environment and could affect detection of patient-specific correlates of response. Finally, Roider et al. nicely compared bispecific T cell engagers in an in vitro lymphoma culture system, but their study largely focused on short-term treatment responses and did not deeply characterize the long-term stability of their model²⁰.

In this study, we describe the development and validation of a novel organoid method for in vitro culture of fully autologous PDLOs and overcome prior limitations by including the intact, native FL TME without the requirement for supportive exogenous cytokines or cells. We show the robustness of this system for successful propagation of cryopreserved FL primary tumor specimens across variable disease milestones, including aggressive histological transformation. We demonstrate the fidelity and stability of individual FL avatars over 3 weeks using multimodal longitudinal characterization of TME cellular composition, tumor somatic mutations, gene expression profiles, and B/T cell receptor dynamics. This deep characterization of primary patient tumors (and their corresponding PDLOs) to our knowledge represents the most comprehensive analysis of FL models to date. While preceding studies have contributed new insights to the field, long-term culture stability is critical for characterizing TME interactions in FL due to its indolent nature.

Separately, using CD3:CD20 and CD3:CD19 bispecific antibodies to engage native T cell effectors in the FL TME, we show specific killing of B cells and cognate T cell activation in PDLOs. Our study identified a potential dual role for Tfh cells before and after T cell-engaging bispecific antibody treatment. We observed that a high starting proportion of Tfh in the TME was negatively correlated with tumor killing, consistent with prior data showing that Tfh can support lymphoma cell survival through the induction of immunosuppressive cytokine release and regulatory T cell activation⁶. Supporting this finding, we report elevated IL-10 secretion following bispecific therapy. The immunosuppressive capacity of Tfh may influence the efficacy of certain T cell-engaging therapies, as suggested by work from Bunse et al. demonstrating the therapeutic potential of CXCR5-targeted CAR T cells to eliminate both lymphoma and Tfh cells²⁸. However, in other settings, such as during an infection response, aberrant Tfh cells can kill germinal center B cells, ultimately leading to a stunted GC response²⁹. We observed that Tfh activation after therapy was correlated with improved tumor killing independent of the bispecific antibody tumor target, and hypothesize that properly activated Tfh might contribute to bispecific-mediated lymphoma killing. The overall relationship between baseline Tfh, Tfh activation, and tumor killing may be complex, with both immune activating and suppressive pathways involved in response to treatment.

Collectively, these results demonstrate the promise of a FL PDLO system as a patient-specific modeling platform that enables high throughput screening and identification of TME signatures associated with individual outcomes. Given its stability, this system provides a robust platform for evaluating treatment responses and advancing precision medicine in FL. We evaluated stability over three weeks, but PDLO viability may extend well beyond this timeframe. With sufficient clinical biopsy material, many organoids can be generated from a single lymphoma patient’s tumor. One can then use high throughput screening strategies to test a multitude of approved and emerging lymphoma therapeutic compounds and assess whether organoids can recapitulate patient-specific clinical responses. This approach is complementary to testing in cell lines or allogeneic human systems. Further, this model system allows the interrogation of many TME features important for our mechanistic understanding of treatment responses and can potentially be used to inform future targets for therapeutic intervention^30,31. Our future studies will include utilizing the PDLO system to further characterize the pre-treatment dynamics of the TME, including the role for Tfh and other emergent effector signatures in relation to tumor killing. Although this study included only FL patients, additional integration of different lymphoma types as well as cell populations from autologous blood or bone marrow will be possible in the future.

LIMITATIONS OF STUDY

Although the described PDLO system enables a new way to integrate therapeutic responses to treat FL, some limitations of this platform should be considered. In this study, we were limited to analysis of research-grade bispecific compounds rather than to clinical-grade biologics. Although we observed a strong therapeutic response in PDLOs, the performance of clinical-grade molecules in vivo may differ from what we observed here. This may be partly due to the focused nature of the system, which is lymphoid cell dominant and does not incorporate circulating cells that could contribute additional modulatory roles during the in vivo response. Furthermore, bispecific therapies have only recently been approved for use in FL⁹, and at present time it has not been possible to benchmark findings from bispecific treatment in PDLOs to patient-specific clinical responses. Since the PDLOs produced in this study were predominately composed of normal and malignant B cells and a variety of CD4 T cells, future studies will be needed to augment the representation of other key lymphoma TME constituents in PDLOs, including stromal cells, ideally from autologous sources.

STAR METHODS

RESOURCE AVAILABILITY

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Lisa Wagar (lwagar@hs.uci.edu).

Materials availability

Reagents generated in this study will be made available on request, but we may require a payment and/or a completed Materials Transfer Agreement if there is potential for commercial application.

Data and code availability

• RNA-seq and targeted DNA sequencing data have been deposited at dbGAP and are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

• This paper does not report original code.

• Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

KEY RESOURCES TABLE.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Human TruStain FcX	Biolegend	BioLegend Cat# 422302, RRID:AB_2818986
Anti-Human Kappa light chain (MHK-49) FITC	Biolegend	BioLegend Cat# 316506, RRID:AB_493611
Anti-Human Lambda light chain (MHL-38) APC Fire 750	Biolegend	BioLegend Cat# 316625, RRID:AB_2687265
Anti-Human CD8 (RPA-T8) PerCP-Cy5.5	Biolegend	BioLegend Cat# 301032, RRID:AB_AB_893422
Anti-Human CD10 (HI10a) APC	Biolegend	BioLegend Cat# 312210, RRID:AB_314921
Anti-Human PD1 (EH12.2H7) AF700	Biolegend	BioLegend Cat# 329952, RRID:AB_2566364
Anti-Human CD19 (HIB19) BV650	Biolegend	BioLegend Cat# 302238, RRID:AB_2562097
Anti-Human CD20 (2H7) BV605	Biolegend	BioLegend Cat# 302334, RRID:AB_2563398
Anti-Human CD4 (OKT4) BV785	Biolegend	BioLegend Cat# 317442, RRID:AB_2563242
Anti-Human CXCR5 (J252D4) PE	Biolegend	BioLegend Cat# 356904, RRID:AB_2561813
Anti-Human CD38 (HIT2) PE-Dazzle594	Biolegend	BioLegend Cat# 303538, RRID:AB_2564105
Anti-Human CD3 (HIT3A) PE-Cy5	Biolegend	BioLegend Cat# 300310, RRID:AB_314046
Anti-Human CD27 (O323) PE-Cy7	Biolegend	BioLegend Cat# 302838, RRID:AB_2561919
Anti-Human CD5 (L17F12) Pacific blue	Biolegend	BioLegend Cat# 364024, RRID:AB_2566250
Anti-Human CD3-CD19 bispecific antibody	Invivogen	bimab-hcd19cd3
Anti-Human CD3-β-galactosidase bispecific antibody	Invivogen	bimab-bgalhcd3
Anti-Human CD19-β-galactosidase bispecific antibody	Invivogen	bimab-hcd19bgal
Anti-Human CD3-CD20 bispecific antibody	BPS Bioscience	100836
Anti-Human CD20 antibody	BPS Bioscience	71209
Biological samples
Lymphoma patient biopsies	Stanford University	Methods S1
Fetal Bovine Serum, heat inactivated	R&D Systems	S11550
Bovine Serum Albumin (BSA)	Fisher Scientific	BP9700100
Chemicals, peptides, and recombinant proteins
Normocin	InvivoGen	ant-nr-1
Penicillin-Streptomycin	Gibco	15140122
Antibiotic-Antimycotic	Gibco	15240096
DMSO	Sigma	D4540-100ML
RPMI1640 with glutamax	Gibco	61870036
Nonessential amino acids	Gibco	11140050
Sodium pyruvate	Gibco	11360-070
Insulin, selenium, transferrin supplement	Gibco	41400045
PBS	Gibco	10010-023
Sodium azide	Thermo Fisher	71448-16
EDTA 0.5M, pH 8.0	Invitrogen	15-575-020
Critical commercial assays
Cyto-Fast Fix/Perm buffer	Biolegend	426803
Zombie Aqua™ Fixable Viability Kit	Biolegend	423101
Custom Premix Human Cyto Panel A 36 Plex luminex assay	Millipore Sigma	HCYTA-60K-36C
AllPrep DNA/RNA Micro Kit	Qiagen	80284
RNA Clean & Concentrator-5	Zymo	R1013
SMARTer Stranded Total RNA-Seq Kit v3	Takara	634485
Deposited data
Raw and analyzed data - dbGaP Accession ID	This paper	phs003410.v1.p1
Software and algorithms
FlowJo	BD Bioscience	Version 10.8.1
xPONENT software	Luminex Corp	Version 4.3
AfterQC	Chen et al32	https://github.com/OpenGene/AfterQC
STAR	Dobin et al33	https://github.com/alexdobin/STAR
Umi_tools	Smith et al34	https://github.com/CGATOxford/UMI-tools
RNASeQC	Graubert et al35	https://github.com/CGATOxford/UMI-tools
Salmon	Patro et al36	https://combine-lab.github.io/salmon/about/
DESeq2	Love et al37	https://bioconductor.org
Fgsea	Korotkevich et al38	https://github.com/ctlab/fgsea
GOSeq	Young et al39	https://bioconductor.org/packages/goseq/
CIBERSORTx	Newman et al24	https://cibersortx.stanford.edu/
ClusterProfiler	Wu et al40	https://github.com/YuLab-SMU/clusterProfiler
BWA	Li et al 41	https://bio-bwa.sourceforge.net/
CAPP-Seq	Newman et al42
gnomAD Database	Karczewski et al43	https://gnomad.broadinstitute.org/
MixCR	Bolotin et al44	https://mixcr.com/
Other
Corning Costar 96-Well, Ultra-Low Binding, Flat-Bottom Microplate	Corning	3474

EXPERIMENTAL MODEL AND SUBJECT DETAILS

Informed consent and sample collection

Patients undergoing treatment and care at Stanford University were consented for an observational study of blood and tissue-based biomarkers approved by the local IRB (NCT00398177). All patients were provided with written informed consent per protocol. Tumor biopsy samples were obtained as indicated by standard of clinical care at diagnosis or relapse from individuals with diagnosed or suspected lymphoma. Treatment decisions were made at the discretion of the treating physician in accordance with institutional standards. Patients included in this study were enrolled between 2012 and 2020. A summary of patient age, sex, disease type, and clinical variables are provided in Methods S1.

Tissue processing and generation of immune organoids

Surgically excised lymph nodes or splenic tissue containing viable lymphoma cells were collected as fresh tissue in saline or PBS at diagnosis or relapse. Samples were mechanically dissociated in DPBS over ice utilizing a sterile pestle and stainless-steel cell dissociation sieve into single-cell suspensions, resuspended in freezing media (10% DMSO, 50% FBS, 40% RPMI), and stored in liquid nitrogen until preparation as immune organoids.

To generate organoids, cryopreserved cells were thawed and excess DNA was removed with benzonase treatment (1:10,000 in media while thawing cells; Sigma). Cells were washed with RPMI+10% FBS and plated at a final density of 7.5 ×10⁶/ml (200 μl final volume; 1.5×10⁶ live cells/well) in flat 96-well ultra-low attachment plates (Corning). Organoid media was composed of RPMI1640 with glutamax, 10% FBS, 1x nonessential amino acids, 1x sodium pyruvate, 1x penicillin–streptomycin, 1x Normocin (InvivoGen), 1x Insulin-Transferrin-Selenium (ITS -G) supplement (Gibco). Cultures were incubated at 37°C, 5% CO₂ with humidity and media was replenished every other day by exchanging 30% of the volume with fresh organoid media. Healthy tonsil organoids were generated and maintained as previously described²².

Flow cytometry

PDLOs were harvested and cells were washed with FACS buffer (PBS + 0.1% BSA, 0.05% sodium azide, and 2 mM EDTA) to remove any residual antibodies or factors generated during culture. Cells were first incubated with anti-kappa and -lambda light chain antibodies, Human TruStain FcX (Biolegend), and zombie viability stain (Biolegend) for 30 minutes on ice while protected from light. Cells were thoroughly washed with FACS buffer and stained with additional antibody cocktails for 30 minutes on ice (Supplemental Table 4). All data were collected using a Quanteon ACEA NovoCyte Quanteon (Agilent) flow cytometer and analyzed using FlowJo (v10.8) software.

Cytokine analysis

IFNγ, TNFα, and IL-10 in culture supernatants were measured using a custom 36-plex human cytokine Luminex assay (Millipore Sigma). Supernatants were diluted 1:1 and processed per manufacturer’s instructions, in duplicates, in a 96-well plate format on the MAGPIX multiplexing system (Luminex Corp). Standard curves were fit using five-parameter logistic regression on the xPONENT software (Luminex Corp).

Bispecific antibody treatment

On day 4 of culture, bispecific antibodies and corresponding control antibodies were added to cultures at a final concentration of 100 ng/ml. For anti-human CD3:CD19 (Invivogen) bispecific antibodies, unconjugated anti-human CD19-beta galectin (Invivogen) and anti-human CD20-beta galectin (Invivogen) were added in combination to serve as a control. PDLO treatment with anti-human CD20 agonist (BPS Bioscience) was used as a control for anti-human CD3:CD20 (BPS Bioscience) treated groups.

RNA/DNA isolation

Organoids were harvested and washed with PBS and 500,000–750,000 cells were isolated. RNA and DNA were isolated using AllPrep DNA/RNA Micro kit (Qiagen) per manufacturer instructions. Eluted RNA was DNase treated using RNA clean and concentrator-5 kits (Zymo Research).

RNA library prep and sequencing

An input of 5 or 10 ng of RNA was used to generate libraries for sequencing using the SMARTer Pico V3 kit. Libraries were sequenced using the NovaSeq6000 S4 platform at a target depth of 30 million paired-end reads per sample.

RNA sequence alignment and differential expression analysis

Raw RNA sequencing data underwent demultiplexing, UMI extraction, and was processed with AfterQC³² for end trimming and read quality filtering. Reads were mapped to both the transcriptome and whole genome utilizing STAR³³ and deduplicated by UMI with umi_tools³⁴. Genome-aligned data was used for quality control with RNASeQC³⁵, and transcript-aligned data was quantified using salmon³⁶ with GC and sequencing bias correction. Gene expression data was analyzed in R utilizing the DESeq2 package for differential expression³⁷, fgsea for gene signature enrichment³⁸, and GOSeq for gene ontology and KEGG pathway enrichment³⁹.

RNA deconvolution & cell subset analysis

Cell subsets were quantified from bulk RNA transcript data utilizing CIBERSORTx with LM22 immune cell subset gene signatures, run in absolute mode utilizing B-mode batch correction and 500 permutations for significance²⁴. Cell type-specific gene expression was evaluated using CIBERSORTx in Hi-Res mode, considering the full transcriptome across LM10 major cell subsets with batch correction and an automatic window size of 34. Gene set enrichment analysis of cell subset-specific expression was performed with the ClusterProfiler package⁴⁰.

DNA library preparation and targeted sequencing

Targeted sequencing by the Cancer Personalized Profiling by Deep Sequencing (CAPP-Seq) approach was performed on organoid DNA from the described timepoints²⁴. Isolated DNA was fragmented to ~250–300 bp with the Covaris Focused Ultrasonicator and quantified with the TapeStation 4200 (Agilent) and Qubit dsDNA High Sensitivity Kit (Life Technologies) post-fragmentation. Samples were prepared for sequencing with the Kapa HyperPrep Kit (Roche) utilizing custom adapters for sample and molecular barcoding. Hybridization capture was performed with a sequencing panel including 322 commonly mutated lymphoma genes of interest (Supplemental Table 5). Sequencing was performed on the Illumina HiSeq4000 platform (Illumina, San Diego, CA) using 2×150 paired end reads.

DNA sequence processing and variant calling

Raw sequencing data was first mapped to the hg19 reference genome (NCBI Build 37.1/GRCh37) using Burrows-Wheeler Aligner⁴¹. Stereotypical errors and PCR duplicates were removed by using a nucleotide-level background database and specialized computational workflow including deduplication as previously described⁴².

Mutational analysis and concordance

For D0 organoid samples, SNVs and insertions/deletions were genotyped as previously described²⁴. Candidate coding mutations were filtered for germline polymorphisms using the gnomAD population-based database and paired patient PBMC/germline samples⁴³. Sequencing data from subsequent timepoints was analyzed both in reference to D0 mutations (for stability) and independently (for emergent mutations). Identified genetic aberrations are listed in Supplemental Table 2). To consider the association between mutational profile and PDLO stability, for each case we calculated the relative allele frequency (AF) of all coding variants at D0 versus the final timepoint. These ratios were log10 transformed to create an approximately normal distribution. Each recurrent gene was then tested per-gene against the pool of all other genes via two-sided Wilcoxon test with Bonferroni correction to identify genes significantly associated with stability versus attrition.

Lymphoid cell immune repertoire

Immune repertoire was evaluated from all sequenced samples using two approaches. From bulk RNA sequencing, MixCR version 3.0.13⁴⁴ was utilized in RNA-seq mode to determine clonal fraction, V(D)J usage, and CDR3 amino acid sequence (with partial assemblies allowed). From DNA, rearranged BCR/TCR were enriched by the inclusion of V-, D-, and J-gene family members in the hybridization capture panel. Individual clones were identified with MixCR utilizing the kAligner2 algorithm, and TCR repertoire was additionally assessed using the SABER algorithm⁴³. Clonal rearrangements can be found in Supplemental Table 3.

QUANTIFICATION AND STATISTICAL ANALYSIS

All statistical analyses were performed using R. Paired Mann-Whitney tests were used to identify differences between groups. An * indicates a p value of < 0.05. For correlation analyses, Spearman’s rank correlation tests were performed.

Supplementary Material

1

Table S1. Clinical immunoglobulin light Chain and IHC characteristics of primary patient samples. Related to Figures 1D, 1E.

2

Table S2. Detected PDLO mutations and longitudinal stability. Related to Figures 2A, 2D.

3

Table S3. Detected clonal V(D)J rearrangements and longitudinal stability. Related to Figure 2B.

4

Table S4. Flow cytometry markers and reagents. Related to Figures 1, 3, 4 and STAR Methods Flow Cytometry.

5

Table S5. Genes covered in lymphoma-focused CAPP-Seq hybrid capture selector. Related to Figures 2A, 2D and STAR Methods targeted sequencing and mutational analysis.

Highlights.

• Patient-derived lymphoma organoids recapitulate the tumor lymphoid microenvironment

• Follicular lymphoma organoids are stable in culture for at least three weeks

• Potential applications for immunotherapy testing and personalized medicine