Abstract

   Cancer models play critical roles in basic cancer research and
   precision medicine. However, current in vitro cancer models are limited
   by their inability to mimic the three-dimensional architecture and
   heterogeneous tumor microenvironments (TME) of in vivo tumors. Here, we
   develop an innovative patient-specific lung cancer assembloid (LCA)
   model by using droplet microfluidic technology based on a
   microinjection strategy. This method enables precise manipulation of
   clinical microsamples and rapid generation of LCAs with good
   intra-batch consistency in size and cell composition by evenly
   encapsulating patient tumor-derived TME cells and lung cancer organoids
   inside microgels. LCAs recapitulate the inter- and intratumoral
   heterogeneity, TME cellular diversity, and genomic and transcriptomic
   landscape of their parental tumors. LCA model could reconstruct the
   functional heterogeneity of cancer-associated fibroblasts and reflect
   the influence of TME on drug responses compared to cancer organoids.
   Notably, LCAs accurately replicate the clinical outcomes of patients,
   suggesting the potential of the LCA model to predict personalized
   treatments. Collectively, our studies provide a valuable method for
   precisely fabricating cancer assembloids and a promising LCA model for
   cancer research and personalized medicine.

   Subject terms: Cancer models, Tissue engineering
     __________________________________________________________________

   Realistic tumour models are critical for the development of clinically
   relevant treatments. Here, the authors develop a lung cancer assembloid
   model which recapitulates key components of the primary tumour, and can
   be used to predict clinical outcome.

Introduction

   Lung cancer is the leading cause of cancer deaths, with ~1.8 million
   deaths worldwide in 2020^[64]1. Despite the increasing availability of
   therapeutic strategies, including targeted therapy and immunotherapy,
   few patients achieve complete remission, and patient responses are
   highly variable^[65]2. It has been appreciated that tumor heterogeneity
   and tumor microenvironments (TMEs) contribute to tumor development and
   poor outcomes of anticancer treatment^[66]3,[67]4. The TME consists of
   extracellular matrix (ECM) and various cellular components, including
   immune cells and stromal cells. Cancer-associated fibroblasts (CAFs)
   are major stromal cells in the TME with the ability to drive cancer
   metastasis and drug resistance and modulate the immune
   microenvironment^[68]5. The TME varies greatly between and within each
   patient, causes great disease diversity and poses a major challenge for
   precision therapy and drug development^[69]6. Hence, reconstructing a
   cancer model with tumor heterogeneity and a personalized TME in vitro
   has become a key issue in cancer research and precision medicine.

   In vitro cancer models have contributed tremendously to cancer research
   and anticancer drug development. However, traditional cancer models,
   including 2D and 3D sphere cultures, lack the heterogeneous cell
   subtypes and molecular features of parental tumors^[70]7,[71]8.
   Patient-derived cancer organoids are currently the “star” cancer model
   that can replicate the pathological morphology and some genetic
   features of parental tumors. However, conventional cancer organoid
   models based on matrigel mainly represent tumor epithelium, endogenous
   stromal and immune cells are gradually lost over time in
   culture^[72]7,[73]9–[74]11. Although some studies reconstituted a part
   of the TME in organoid culture systems by the air-liquid interface
   (ALI) method^[75]12 or coculturing organoids with TME cells such as
   CAFs^[76]13,[77]14 and immune cells^[78]15,[79]16, some other studies
   developed non-Matrigel-based hydrogel 3D cancer models comprised of
   heterogeneous patient-derived tumor cells and stromal
   cells^[80]17,[81]18, the models lacked precise controllability and
   uniformity in addition to labor costs. Some other cancer organoids
   derived from minced tumor fragments could maintain the native tissue
   architecture and TME cell components. However, manual tissue mincing
   results in nonreproducible fragment sizes and nonuniform
   environments^[82]9,[83]19,[84]20. On the other hand, a limited number
   of millimeter-scale tumor fragments derived from small tumor tissues
   are limited in application in high-throughput drug screening.

   Assembloids are 3D structures formed from the fusion and functional
   integration of multiple cell types or organoids, which are the latest
   tools for understanding human development and disease and are now
   considered at the leading edge of stem cell research^[85]21–[86]24.
   Bladder tumor assembloids were created and partially recapitulated the
   in vivo pathophysiological features of urothelial
   carcinoma^[87]23,[88]25. Currently, assembloids are mainly fabricated
   by coculture^[89]26,[90]27 and 3D extrusion printing methods^[91]25.
   The morphology and structure of assembloids fabricated by coculture
   methods are difficult to control and have poor intrabatch
   consistency^[92]28. Although a kidney organoid model^[93]29 with tissue
   morphology could be fabricated by using extrusion-based 3D printing
   method, only 18 organoids with diameter of ~2 mm (0.55 μL in volume for
   each organoid), could be generated per minute. 3D extrusion bioprinting
   is limited in generating micron-size tissue models rapidly with
   requisite size and accuracy^[94]29–[95]32. Rapid preparation of tumor
   assembloids with good intrabatch uniformity remains a great challenge.

   In this study, we report an innovative patient-specific lung cancer
   assembloid (LCA) model generated by microinjection-based droplet
   microfluidic technology that enables precise manipulation of clinical
   microsamples and high-throughput generation of LCAs (Fig. [96]1). LCAs
   are achieved by evenly encapsulating patient tumor-derived TME cells
   and lung cancer organoids (LCOs) inside gelatin methacryloyl
   (GelMA)-Matrigel microgels with good cytocompatibility. This LCA model
   demonstrates good intrabatch consistency in size, cell composition and
   drug response profiling. In addition, these LCAs represent the TME and
   tumor heterogeneity of their parental tumors and replicate the clinical
   responses of patients with lung cancer, highlighting the potential
   utility of our LCA model for basic research and personalized drug
   screening.

Fig. 1. The patient-specific lung cancer assembloid (LCA) model.

   [97]Fig. 1
   [98]Open in a new tab

   a The schematic illustration of the preparation of bioinks loaded with
   lung cancer organoids and TME cells. b Fabrication of LCAs by using a
   droplet microfluidic technology based on a microinjection strategy. c
   The advantages of LCAs which show good intra-batch consistency,
   represent TME heterogeneity of parental tumors and replicate clinical
   drug responses. LCOs, lung cancer organoids; CAFs, cancer associated
   fibroblasts; TME, tumor environment; TILs, tumor infiltrating
   lymphocyte cells.

Results

Establishment of uniform LCAs through a microinjection strategy-based droplet
microfluidic technology

   The rise of cancer assembloids provides a promising tool for cancer
   research. To generate uniform LCAs with personalized TMEs in a
   high-throughput manner, we developed an LCA platform using innovative
   microinjection strategy-based microfluidic technology (Supplementary
   Fig. [99]1). The platform is designed for cell-laden GelMA-Matrigel
   manipulation, mainly consisting of bioink preparation and LCA
   generation processes (Fig. [100]1a, b). A total of 49 clinical tumor
   samples were collected, and tumor-derived cells (LCOs and TME cells)
   from 36 patients were used to fabricate LCAs by using droplet
   microfluidics. We successfully fabricated LCAs in 35 patients with a
   97.2% success rate. (Supplementary Fig. [101]2a-g,
   Supplementary Table [102]2). LCOs and TME cells were encapsulated into
   the optimized GelMA-Matrigel hydrogel for bioink preparation at a
   density of 4 × 10^7 mL^−1 cells. The microinjection module is designed
   for microsample manipulation and includes a silica tube with an inner
   diameter of 500 μm for sucking bioink into the tube. The bioink is
   separated by air from the booster reagent PBS to prevent it from being
   diluted. The end of the tube could be attached to the forming module, a
   T-junction chip where the cell-laden hydrogel is subsequently sheared
   by mineral oil into monodisperse droplets (Fig. [103]1b, Supplementary
   Fig. [104]1). The flow rates of the bioink and oil phase were optimized
   at 1 and 5 mL h^−1, respectively, to form uniform assembloid precursors
   with a size of 400–500 μm encapsulating a certain number of cells
   (e.g., 1500–2500 cells). The droplets are subsequently UV
   photo-crosslinked with controllable UV intensity and form stable
   cell-laden microgels (LCA precursors) that can grow into LCAs with
   patient-specific TMEs after 3 days of culture (Fig. [105]2a–c).

Fig. 2. Establishment of uniform LCAs through a microinjection strategy-based
droplet microfluidic technology.

   [106]Fig. 2
   [107]Open in a new tab

   a Schematic representation of LCA fabrication strategy and
   characterization. b Representative bright field microscopy images of
   LCAs (LC05) at day 0 (LCA precursor) and day 3 post fabrication. The
   right images are the enlarged views circled by the red lines. The
   experiment was repeated in 35 patient samples. c Immunofluorescence
   staining of the EpCAM and α-SMA markers in LCA precursors and LCAs. The
   experiment was repeated in 15 patient samples. d Compressive elastic
   modulus of GelMA-Matrigel hydrogels, lung cancer tumors and the matched
   adjacent normal lung tissues (n = 7 independent samples for 6–0, tumor
   and normal groups, n = 5 independent samples for 6–15 and 6–30 groups).
   e Representative live (green) & dead (red) staining images of LCAs at
   day 0 and day 7 post fabrication. The experiments are repeated in LCAs
   of 6 patient samples. f The normalized cell ability of cells before
   assembling and assembled as LCAs at day 0 and day 7 (n = 3 biologically
   independent samples). g Quantitative analysis of cell proliferation
   ability of LCAs over culture time (n = 3 biologically independent
   cells). h The cellular percentages of LCOs and CAFs in intra-batch LCAs
   (n = 8 independent LCAs). i Histograms of LCA size distribution for
   high cell density (more cells, 10^8 mL^−1, n = 53 LCAs) and low cell
   density (less cells, 10^6 mL^−1, n = 62 LCAs). j Histograms of LCA size
   distribution fabricated with 10 μL of microsamples derived from tumor
   biopsies (LC33 & LC34) (n = 35 independent LCAs). k Representative
   bright-field images of LCAs before freezing and after thawing. The
   experiments are repeated in LCAs of 6 patient samples. l Diameters of
   LCAs before freezing (n = 16 independent LCAs) and after thawing
   (n = 10 independent LCAs). Scale bar, yellow bar, 200 μm; two-sided
   student’s t test is used, data are presented as mean ± S.E.M. Source
   data are provided as a Source Data file.

   To generate LCAs with good mechanical properties and biological
   activity, we chose GelMA and Matrigel composite hydrogels because GelMA
   hydrogels are widely used for their excellent processing capability,
   tunable mechanical properties, and biocompatibility even for immune
   cells^[108]16,[109]33–[110]35, while Matrigel hydrogels can provide a
   favorable tumor microenvironment for patient-derived
   organoids^[111]36,[112]37. The material ratios were optimized to ensure
   good biocompatibility and formability. The analysis of compressive
   mechanical property indicated that the hydrogel consisting of 15% (v/v)
   Matrigel and 6% (w/v) GelMA (termed 6–15) exposed to 90 mW of blue
   light at 405 nm for 40 s showed comparable mechanical properties to
   those of patient lung tumors (19.1 ± 0.4 vs. 27.9 ± 3.9 kPa)
   (Fig. [113]2d, Supplementary Fig. [114]3a). More importantly, the 6–15
   hydrogel maintained good cell viability and cell proliferation compared
   with the 6–0 (6% GelMA) and 6–30 (6% GelMA plus 30% Matrigel) hydrogels
   (Fig. [115]2e–g, Supplementary Fig. [116]3b–d).

   In addition, uniform LCA precursors with good intrabatch consistency in
   terms of size, cell composition and distribution could be generated
   using 6–15 hydrogels in this platform (Fig. [117]2h, Supplementary
   Fig. [118]3e, f). Even 10 μL hydrogels containing 10^6 ~ 10^8 cells
   mL^−1 could be successfully manipulated to generate ~200 uniform LCAs
   with sizes of 400–500 μm (~0.05 μL per LCA) within 1 min (Fig. [119]2i,
   Supplementary Fig. [120]3g, h). Encouraged by the results, we
   successfully generated uniform LCAs directly using the limited number
   of cells derived from tiny tumor needle biopsies (Fig. [121]2j,
   Supplementary Fig. [122]3h). This suggested broad application of this
   platform in the rapid fabrication of cancer assembloid models even
   using microsamples such as biopsies that can be easily obtained from
   patients with intermediate and advanced tumor stages^[123]38.

   The ability to bank such assembloids will improve the utilization of
   LCAs and provide researchers with the opportunity to generate living
   biobanks, which will substantially contribute to basic and
   translational research in a wide range of areas^[124]39. Therefore, we
   performed a thawing test for cryopreserved LCAs. LCAs could
   successfully reconstitute their biological properties after being
   frozen for 2 months. The morphology and diameter of LCAs before
   freezing and after thawing showed great similarity, and cells assembled
   inside the LCAs maintained good viability and proliferation after
   thawing (Fig. [125]2k, l), suggesting that the LCAs could be
   cryopreserved as a biobank for further applications.

LCAs maintain the heterogeneous histology and TME features of parental tumors

   Tumors have the features of inter- and intratumor heterogeneity,
   including but not limited to cellular and histological
   heterogeneity^[126]40,[127]41. The LCAs derived from the same patient
   or different patients showed heterogeneous morphology, suggesting the
   maintenance of inter- and intratumor heterogeneity of patients
   (Supplementary Fig. [128]4a). To further assess whether the LCAs
   resemble their corresponding parental tumors at the histological level,
   we performed histological analyses. The hematoxylin and eosin (H&E)
   staining results showed that the LCAs had similar histological features
   to their parental tumors (Fig. [129]3a, Supplementary Data [130]1).
   Stromal cells (red arrows) were observed wrapping around the tumor
   cells (black arrows) and forming junctions with each other as indicated
   by the arrows. LCAs derived from adenocarcinomas (ACs) of different
   patients maintained the intertumoral heterogeneity of cancer cell
   differentiation degree, and the expression patterns of EpCAM,
   cytokeratin 7 (CK7) and Ki67 were also retained in the LCAs
   (Fig. [131]3b, Supplementary Fig. [132]4a, b). It is worth noting that

Fig. 3. LCAs maintain the heterogeneous histology and TME features of
parental tumors.

   [133]Fig. 3
   [134]Open in a new tab

   a H&E staining images of LCAs and their corresponding parental tumors
   (LC06, LC17 and LC07). Black arrows indicated the tumor cells and the
   red arrows indicated the stromal cells. b Concordant expression of CK7
   and EpCAM in parental tumors (LC05, LC14 and LC63) and their derived
   LCAs. c Immunofluorescence staining of human α-SMA and EpCAM in tumor
   fragments and corresponding LCAs cultured for 1 week and 2 weeks. d
   Immunofluorescence staining of human CD3 and EpCAM in tumor fragments
   and LCAs (LCA37 and LCA66). For (a–d) each experiment was repeated
   independently with similar results for 3 times. Scale bar, yellow bar,
   200 μm, green bar, 20 μm.

   LCAs replicated the heterogeneous expression of CK7 and Ki67 markers
   within LC14 tumors, which indicated that LCAs also recapitulated the
   intratumoral heterogeneity of the original tumor tissues.

   To further characterize the TME in LCAs, we performed
   immunohistological analysis using specific markers of TME cells. We
   observed heterogeneous cells within the LCAs, including the EpCAM^+
   tumor cells, α-SMA^+ CAFs and CD45^+/CD3^+ immune cells (Fig. [135]3c,
   d, Supplementary Fig. [136]4c–e), which indicated that LCAs could
   recapitulate a certain tumor microenvironment of the parental tumors.
   In particular, LCAs of 5 patients generated by droplet microfluidics
   demonstrated a more uniform shape, cell distribution and TME
   maintenance than those generated by coculture in U-bottomed ultralow
   attachment microplates (ULAs), which are often used to generate 3D
   spheroid tumor models^[137]42 (Supplementary Fig. [138]4f, g). Ki67 and
   hypoxia probe staining showed that tumor cells proliferated well
   without an obvious tumor hypoxia zone within the LCAs (Supplementary
   Fig. [139]4h, i). The small size of our LCAs (400~500 μm) may
   contribute to oxygen and nutrient delivery, and we could mimic tumor
   hypoxia gradients by increasing the size of LCAs according to the need
   for research^[140]20.

   Overall, these results demonstrated that our LCAs could maintain the
   TME and heterogeneous histology of corresponding parental tumors.

LCAs maintain the transcriptomic and genomic signatures of parental tumors

   To determine whether LCAs maintained the transcriptomic landscape of
   their corresponding parental tumors. We performed bulk RNA sequencing
   (RNA-seq) on samples from 4 patients (LC14, LC28, LC51 and LC52),
   including tumor samples, matched adjacent normal lung tissues for
   references, and the corresponding LCAs derived from the tumor organoids