Oral Mucosal Organoids as a Potential Platform for Personalized Cancer Therapy

The oral cavity, pharynx, and larynx are lined by a stratified mucosa that protects the underlying structures. These epithelia are keratinizing or nonkeratinizing, depending on the anatomic location (1). Neoplasias commonly arise in this epithelium, with a worldwide incidence of more than half a million patients a year (2). Well-known risk factors are alcohol and tobacco (3). Treatment of head and neck squamous cell carcinomas (HNSCC) is difficult, partly because of their anatomic location that complicates surgery, and partly due to the highly variable treatment response. Advanced cases require combinations of surgery, radiotherapy, and chemotherapy. Taken together, this results in relapse rates of more than 50% (4). Currently, there are no reliable models to predict therapy outcome and guide treatment decisions.

In vitro studies of this epithelium have relied on tumor-derived two-dimensional (2-D) cell lines (5) and on primary keratinocyte cultures (6). Success rates to establish HNSCC-derived cell lines range from 11% to 33% (7). For primary keratinocyte cultures, keratinocytes are grown on feeder cells (mouse fibroblasts) in 2-D and have a limited lifespan. In vitro drug screens of 2-D lines have been used to characterize variability in drug response among tumors and as a tool to understand resistance mechanisms of tumor cells (8, 9). In an attempt to overcome limitations of these 2-D systems (10, 11), HNSCC lines have been replated in a three-dimensional (3-D) format. Compared with 2-D cell lines, these 3-D models better recapitulate characteristics of in vivo HNSCC (12, 13). In addition to in vitro models, HNSCC xenograft mouse models were introduced more than 30 years ago (14). Moreover, transgenic mouse models have been developed to understand molecular drivers of HNSCC tumorigenesis (15). Although these models provided important insights into HNSCC, they are lacking the potential for a personalized approach. Recently, two studies have reported on the generation of 3-D HNSCC cultures, giving an initial description of this technology (16, 17).

The oral mucosa can be subject to viral infection (18). Herpes simplex virus (HSV) is among the most commonly encountered viral infections of the oral cavity (18). HSV is known to infect keratinocytes and give rise to herpes labialis (cold sores; ref. 19). Infection with human papillomavirus (HPV) is associated with oropharyngeal HNSCC and characterizes a genetically distinct subgroup of tumors with better prognosis than HPV⁻ HNSCC (20). In vitro culture systems to study the interaction of keratinocytes with HPV are limited to the use of immortalized cell lines, or to primary cells that can only be cultured short-term (21).

Protocols to grow organoids from adult human tissues have been described for single-layered (simple) or (pseudo)stratified epithelia, such as those that line the colon, intestine, liver, pancreas, stomach, esophagus, prostate, lung, breast, and fallopian tube, and cancers derived thereof (22). Tumor organoids have previously been shown to phenocopy the tumor from which they are derived, allowing in vitro drug responses to be linked to genetic alterations present in the original tumor (22). A recent study on gastrointestinal cancers documented a strong correlation between patient clinical outcome and the response of the corresponding tumor organoids (23).

Here, we set out to establish organoids from the epithelial lining of the oral cavity, larynx, and pharynx. Next, we apply these conditions to derive tumor organoids from patients with HNSCC and to explore their potential to aid personalized therapy.

To propagate organoid formation, we tested a range of media compositions described in previously published protocols for growth support of oral mucosa. Conditions that were successful to grow mouse tongue epithelium (see Supplementary Fig. S1A and S1B) were refined on human material obtained from surgical resections. In brief, the epithelial layer was microdissected from the surgical specimen to remove fat and muscle, digested in 0.125% trypsin, and filtered, and the resulting cell suspension was then plated in Basement Membrane Extract (BME), a Matrigel equivalent (Fig. 1A). Within the first few days after plating, organoids grew out from single cells or small cell clumps (Supplementary Fig. S2A; Supplementary Movie S1). Over time, these organoids developed into dense structures, often with keratinized centers (Fig. 1B). On average, organoids could be passaged within 10–14 days. After the first passages, organoids typically expanded exponentially, being passaged every 10 days with a split ratio of 1:5 (Supplementary Movie S2). We received surgical material from 40 patients, of which 26 (65%) grew out into organoids. Established organoid lines could be expanded long-term (> 15 passages) and could be cryopreserved and recovered successfully. We assessed proliferative potential over 5 passages in expansion medium (Fig. 1C) and observed unabated exponential growth. Scanning electron microscopy revealed that organoids are composed of tightly connected cells (Supplementary Fig. S2B). Immunohistochemical (IHC) staining for basal cell marker TP63 and proliferation marker MKI67 of paraffin-embedded organoids showed that the organoids recapitulate the tissue of origin. Proliferative MKI67⁺/TP63⁺ basal cells were located in the external organoid layer touching the basal membrane substitute BME, recapitulating the tissue of origin (Fig. 1D). Differentiation marker KRT13 was detected in cells in the interior of the organoid. Characteristics of keratinocytes such as abundant tonofilaments and desmosomes were observed using transmission electron microscopy (Supplementary Fig. S2C). Upon withdrawal of growth factors from the medium, differentiation increased, as shown by a >45-fold increase in KRT13 expression, and halted proliferation (decreased expression of MKI67; Fig. 1E; Supplementary Fig. S2D). Finally, we assessed the genetic stability of normal tongue epithelium–derived organoid by metaphase spread analysis and observed normal numbers of chromosomes at passage 6 (median 46 ± 0.42, n = 26) and at passage 16 (median 45 ± 0.34, n = 33; Fig. 1F).

We explored the use of this model to study viral infection with HSV type 1 (HSV1), a virus known to infect keratinocytes (19) and to give rise to herpes labialis (cold sores). Using fluorescence microscopy, we followed infection of organoids with tdTomato-labeled HSV (ref. 24; Supplementary Fig. S3A). Using live imaging, spreading of the infection in organoids was observed two days after initial infection (Fig. 1G; Supplementary Movie S3). After two weeks in culture, infection had spread throughout entire organoids (Fig. 1H). Infection of organoids resulted in an increase in viral DNA, which could be inhibited by the addition of acyclovir (viral tyrosine kinase inhibitor) in all three organoid lines tested (Supplementary Fig. S3B).

As HPV is known to contribute to oncogenesis of a subset of HNSCC tumors (3, 25), HPV16 particles were used to infect oral mucosa organoids (26). Viral replication was quantified by increase in HPV DNA levels (Supplementary Fig. S3C). After splitting of the organoids, an increase in HPV DNA could be observed, implying lasting infection (Supplementary Fig. S3D). Finally, transfer of filtered supernatant taken 12 days postinfection from infected organoids resulted in reinfection, proving virion production in organoids (Supplementary Fig. S3E). Taken together, we conclude that oral mucosa–derived organoids allow infection with HSV1 and HPV16, validating the organoid model as an in vitro model for mucosal pathology.

We next set out to grow organoids from patient-derived HNSCC samples, obtained from either surgical resections or biopsies. We successfully established tumoroids from 31 patients, ranging in age from 48 to 91 (average age at diagnosis: 69). Tumoroids were established from tumors originating in the oral cavity (floor of mouth, tongue, and gingiva/alveolar process), pharynx, larynx, salivary gland, nasal cavity, and neck (Fig. 2A). Patient clinical data corresponding to established organoid lines can be found in Supplementary Table S1. Of the 31 established tumoroid lines, 16 were fully characterized molecularly at the date of first submission (data on all others will be added when these become available). These lines were named T1, T2, T3, etc., whereas the corresponding normal epithelium–derived lines were termed N1, N2, N3, etc. The success rate to establish organoids from tumor tissue was approximately 60%. Tumoroids grew either as dense structures (similar to the normal wild-type epithelial organoids) or as cystic structures (Supplementary Figs. S4 and S5). Tumoroids derived from different patients showed different morphologies, as based on brightfield microscopy (Supplementary Fig. S4) and hematoxylin and eosin (H&E) staining (Supplementary Fig. S5). Comparison of organoids with the original tumor tissue and adjacent normal epithelium of the same patient revealed tumor-specific histopathologic changes that were retained in culture (Supplementary Fig. S6).

Comparing IHC stainings of the primary tumor specimens with the corresponding tumoroids revealed that these retained histologic characteristics of the epithelial tumor cells (Fig. 2B). However, as described for other adult tissue–derived organoids, tumoroids contain only the transformed epithelial tumor cells, not the immune, connective tissue, or vessel elements. This was evident by the keratin staining that marks all cells of the organoids, but only the epithelial component of the tumor section (Fig. 2B).

A complication of growing tumoroids from other carcinoma types has been the gradual overgrowth by normal wild-type organoids (27, 28). To confirm the tumor identity of our tumoroids, several approaches were taken. Nutlin-3, an MDM2 agonist, prevents the growth of p53 wild-type cells (29). We observed that 10 of the 14 lines tested (71%) grew in the presence of Nutlin-3 (Supplementary Fig. S7), in agreement with the approximately 75% of HNSCC carrying inactivating TP53 mutations (30). For example, N1 organoids (TP53 wild-type) died in the presence of Nutlin-3, whereas tumoroid line T1 did not (Fig. 2C). TP53 staining on fixed organoid sections confirmed the TP53 status (Fig. 2D): TP53 staining is caused by accumulation of mutant protein, a clinical parameter to determine the presence of mutant TP53 protein (31).

We performed transcriptome analysis of nine normal organoid lines and seven tumoroid lines. As a quality control, two tumor organoid lines (T1 and T3) were sequenced twice in two independent runs. Samples from independent runs clustered together. To determine the contribution of the cancerous state to the total variability in gene expression, principal component analysis (PCA) was performed. PCA revealed segregation of normal and tumor organoids into different clusters, with the exception of tumoroid line T6 (with confirmed mutations in TP53 and CDKN2A) clustering within the normal wild-type organoid group, and normal line N10 not clustering with the other normal samples (Fig. 2E). In addition, we used DESEQ2 analysis to explore the potential to identify or validate tumor biomarkers using HNSCC organoids (32). A search for differentially expressed genes between normal and tumor samples resulted in detection of 857 genes (P_adj < 0.0001). Among the 58 most differentially expressed genes (P < 0.01) we found genes of which the expression has been described to be altered in HNSCC, including KLK6, SLCO1B3, HOXC13, CALB1, EHF, and BCHE (refs. 33–38; Fig. 2F; Supplementary Fig. S8; Supplementary Table S2).

Targeted sequencing (on a panel of 54 oncogenes or tumor suppressors relevant to HNSCC, in one case an extended oncopanel of 140 genes; n = 14; Fig. 3A) or whole-exome sequencing (n = 2; Fig. 3B) was performed on a subset of the HNSCC organoid lines (Supplementary Tables S3 and S4). The most commonly mutated gene, TP53, was genetically altered in 11 of the 16 tumor lines (69%). As expected, no TP53 mutations were detected in the Nutlin-3–sensitive lines T8, T9, T10, T20, and T27. PIK3CA was altered in 7 of the 16 tumoroid lines. KRAS was mutated in four cases, although often mutations were detected at low frequency. BRAF, CDKN2A, FAT1, and PDGFRA were mutated in two cases. Genes affected in one case include ABL, ATR, ESR1, FGFR2, HRAS, MDM2, MET, and VHL. In three cases (T3, T5, and T8), we sequenced both the original tumor and the tumoroid line (see Supplementary Fig. S9) and observed an enrichment of the variant allele frequency (VAF) of the detected mutations. Most likely, this is due to the purely epithelial character of tumoroids, whereas the primary tumor sample also contains immune cells and vasculature and stromal components in addition to the tumor epithelial cells. As T5 and T8 were subjected to whole-exome sequencing, mutations were also detected in genes that were not included in the targeted sequencing panel. Those mutation lists were filtered for those present in genes most commonly affected in HNSCC (39). Using this criterion, we detected pathogenic mutations in 20 (T5) and 4 (T8) HNSCC cancer-associated genes (Fig. 3B). For both tumor organoids, corresponding normal organoids were also sequenced, to confirm the absence of any of the mutations detected in the tumor lines. Subsequently, we scrutinized all single-nucleotide variants (SNV) and small insertions or deletions (Indel) throughout the genome in the tumor and normal tissue as well as in N5 and T5 organoids (Fig. 3C). For both samples, both the tumor tissue and the tumoroids showed SNVs and Indels that were absent from the normal tissue. Moreover, tumoroids largely recapitulated the genetic alterations that were detected in the tumor. Normal organoids and normal tissue lacked these genetic alterations, confirming that these organoids consisted of normal (nontumor) cells.

Chromosome missegregation underlies the aneuploidies frequently observed in human tumors (40). Increased missegregation rates result in the phenotype known as chromosomal instability (CIN), which is commonly observed in cancers, including HNSCC (41). To investigate whether CIN was also present in HNSCC tumoroids, we assessed chromosome segregation in a matched normal and tumor organoid line. Organoids labeled with Histone2B-mNeon were imaged using a spinning-disc confocal microscope to visualize the chromatin during cell division (42). T1 showed elevated levels of chromosome segregation errors when compared with its normal counterpart, N1 (Fig. 4A; Supplementary Movies S4 and S5). The majority of cells in N1 showed no signs of chromosome missegregation (Fig. 4B), whereas T1 showed a variety of segregation errors, including anaphase bridges and binucleated cells undergoing multipolar division (Fig. 4C and D, respectively). We conclude that T1 has acquired CIN during oncogenic transformation. CIN can result in aneuploidy, by the loss or gain of (parts of) chromosomes (40). In agreement, quantification of the number of chromosomes in cells of N1 and T1 using metaphase spreads revealed that this number was much more variable in T1 than it was in N1 (Fig. 4E). Other tumoroid lines also carried aberrant chromosome numbers, with T3 and T4 being tetraploid (Fig. 4F), whereas organoids derived from corresponding normal tissue contained normal numbers of chromosomes.

To assess whether tumorigenic potential of the cultured HNSCC cells was retained, we subcutaneously transplanted tumoroids into mice. Injection of the normal organoids did not result in outgrowth, whereas transplantation in all three tumor lines yielded macroscopically visible tumors after 6 weeks in at least 2 of 3 mice (n = 3 for each organoid line; Fig. 5A). For all tumors, H&E staining revealed stratification and keratinization characteristic of HNSCC (Fig. 5B; Supplementary Fig. S10). Staining for human nuclei showed positive, thus proving the origin of the tumors from the injected human-derived organoids. Comparison of proliferating cells, measured by MKI67 staining, revealed differences in proliferation among the distinct organoid lines (Fig. 5C). These characteristics were retained between mice that were transplanted with the same organoid line (Supplementary Fig. S10). The tumor cells displayed levels of atypia that were regarded as cancerous (assessed by a pathologist). Tripolar mitotic figures and nuclear pleomorphism were observed (Fig. 5D). Moreover, muscle invasion was observed in one case (Fig. 5E). Taken together, this shows that HNSCC organoids retain tumorigenic potential in culture and can form HNSCC with features similar to the parental tumor, upon subcutaneous transplantation into mice.

The genetic alterations found in HNSCC are commonly found in other tumor types, and therapies targeting some of these specific mutations exist. Regardless, with the exception of cetuximab, an anti-EGFR antibody that is used in treatment of HNSCC, no targeted therapies are currently applied in standard care of these patients (43–46). In recent years, it has become clear that mutation status alone does not provide the required specificity or sensitivity to serve as a predictive marker (47). For this reason, we tested a panel of organoid lines carrying different genetic alterations that are regularly found in HNSCC for their in vitro drug sensitivity (Supplementary Fig. S11A).

To refine our in vitro drug screening assay, we used Nutlin-3 treatment. Exposure to a concentration series allowed quantitative discrimination between sensitive and nonsensitive lines (Fig. 6A). As expected, TP53 wild-type cells were sensitive to Nutlin-3 exposure, whereas TP53-mutant organoids were not (Fig. 6B). The assay was reproducible (n = 3; Supplementary Fig. S11B). To assure quality of the drug screening data, a Z factor score (a measure of assay quality) was calculated for each drug screen in this study (Supplementary Fig. S11C). Average Z factor score was 0.7 (ranging from 0.3 to 0.92), which is consistent with an experimentally robust assay.

Subsequently, we exposed 13 fully characterized tumoroid lines (tumoroid lines T10 and T29 did not survive the robotized drug screening procedure) to cisplatin, carboplatin, and cetuximab, drugs currently used in the treatment of patients with HNSCC. Using in vitro concentrations similar to plasma concentrations reached in patients with HNSCC (48), we observed differential sensitivity of the organoids to these compounds. On the basis of the measured IC₅₀, we made a ranking of the tumoroid lines tested for cisplatin and carboplatin (Fig. 6C). Although IC₅₀ values for cisplatin and carboplatin were different (average IC₅₀^cisplatin was 5.91 μmol/L and IC₅₀^carboplatin was 44.1 μmol/L), correlation between cisplatin and carboplatin sensitivity was observed (Pearson correlation, r = 0.64, P < 0.05; Fig. 6D; Supplementary Fig. S12). This correlation has previously been described in ovarian cancer cells (49). In vitro platinum–DNA adduct formation has shown that both drugs give rise to the same damage, yet cisplatin does so at lower dose (50). Although the mechanism of action is the same, it is suggested that cisplatin should be preferred over carboplatin as a radiosensitizer in HNSCC (51). We confirm that, for the lines tested here, none showed a higher sensitivity to carboplatin than to cisplatin.

Organoids were also exposed to the anti-EGFR antibody cetuximab, and differential responses were observed between different patient-derived organoid lines (Fig. 6E; Supplementary Fig. S12). Recent studies challenge the prognostic value of EGFR overexpression or increased gene copy number for cetuximab response (47). Here, no correlation between EGFR expression and cetuximab response was observed. Finally, we found that organoid lines insensitive to cetuximab commonly carried mutations downstream of EGFR (PIK3CA, KRAS, HRAS, or BRAF). This has important implications for patient inclusion for cetuximab therapy, which currently does not include genetic testing. Of note, RAS/RAF genetic testing is routinely performed in patients with colorectal cancer who are eligible for EGFR-targeting treatments.

Currently, treatment with curative intent for patients with advanced HNSCC consists of surgery with adjuvant radiotherapy, with or without chemotherapy, or radiotherapy with or without chemotherapy with the possibility of surgery as salvage treatment in case of residual/recurrent disease (47). Radiotherapy can also be a part of palliative treatment to reduce pain or tumor invasion. Therefore, we tested the sensitivity of the tumoroids to ionizing radiation. We consistently observed differential responses between the tumoroid lines when exposed to radiotherapy. This suggested that clinical correlations could be studied (Fig. 6F). To assess whether expression profiles could predict responses to these therapies, we performed DEseq analysis, comparing organoid lines with different responses to either cisplatin and carboplatin, cetuximab, or radiotherapy (see Supplementary Fig. S13; Supplementary Table S5). Although differentially expressed genes could be detected in all comparisons, including those previously described associated with resistance to these therapies (52–55), no clear indication for resistance mechanisms could be deduced from performing gene enrichment analysis on these gene sets.

Seven of the patients received (postoperative) radiotherapy, which allowed a correlation with organoid sensitivity to this treatment. Although the number of patients is small, and these HNSCC tumors comprise a heterogeneous group (by anatomic location, timing (primary or adjuvant), and total dose of irradiation given), we used this set to explore the potential of HNSCC organoids as a predictor for treatment response in the future. All patients described here were given radiotherapy with curative intent. An overview of the treatment details is given in Supplementary Table S6. Three of 7 patients relapsed after undergoing radiotherapy. Indeed, the three corresponding organoid lines were among the most resistant when exposed to radiotherapy in vitro (Fig. 6G). Organoid line T3 showed the highest sensitivity to radiotherapy. Indeed, this patient had a lasting response to palliative radiotherapy. The primary tumor of the larynx received 48 Gy in 4 weeks and showed no signs of growth upon physical examination up to 5 months later, corresponding with ongoing local control due to radiotherapy. The patient unfortunately succumbed to lung carcinoma 5 months later. The patient from whom T5 was derived was treated with adjuvant radiotherapy following surgery (stage T4aN0 parotid tumor) because of two prognostic factors that predict high risk of relapse of disease: macroscopic residual disease and perineural growth, observed in the resected tissue. The patient received a dose of 66 Gy over a period of 6 weeks at the parotid area. Nine months after completion of radiotherapy, the patient showed no signs of relapse or progression upon physical examination, showing a clinically good response to radiotherapy. Organoid lines T1 and T2 were not responsive to radiotherapy in the in vitro assay. Indeed, the patients corresponding to these lines showed progressive disease shortly after completing the treatment. Patient T1 presented with tongue squamous cell carcinoma (SCC; stage T2N2b) with extranodal growth and was treated with radiotherapy after resection of the primary tumor with positive margins. The patient received a dose of 66 Gy over the course of 10 weeks. Six months later, the patient complained of pain in the tumor area and locoregional relapse of disease was eventually diagnosed 3 months later, leading to death 3 months after that. Patient T2 presented with an SCC in the larynx (stage T2N0) for which 60 Gy of radiation was given. Four months later the patient presented with complaints. Five months later, a recurrent tumor was diagnosed that invaded the subglottis and required a complete laryngectomy. T25 was the organoid line most resistant to radiotherapy. These in vitro findings correlated well with the clinical history of patient T25. This patient presented with an SCC of the floor of mouth (stage T2N1). After receiving surgery for excision of the primary tumor and a selective neck dissection with negative margins, he developed a recurrence in the neck for which he was treated with radiotherapy. A total dose of 52 Gy was delivered to the neck. Only one month later, extensive regional recurrent disease (including skin metastases) in the neck was diagnosed upon physical examination. One month later, the patient succumbed to disease. Patient T27 was diagnosed with a floor of mouth SCC (stage T3N1) for which adjuvant radiotherapy was given after excision of the primary tumor. Indications for adjuvant radiotherapy included close surgical margins and positive sentinel node. The patient received a total dose of ionizing radiation of 56 Gy and was last seen 2 months after the end of treatment. So far, there are no signs of recurrence. It is too early to tell if this patient will remain in remission. It will be of interest to see whether disease will recur, because based on our in vitro findings, radiotherapy should have been effective for this particular tumor. Finally, although T8 appeared resistant to radiotherapy in the in vitro assay, patient T8 so far has not shown relapse after treatment. The patient received adjuvant radiotherapy after incomplete resection of a gingival SCC (stage T4aN0). A dose of 66 Gy was delivered to the tumor area. Five months later, no signs of recurrence were observed upon physical examination. Standard follow-upis ongoing. It will be of interest to see whether the patient relapses in the coming months, a progression that would be in line with our in vitro findings.

In an attempt to overcome the observed radioresistance, HNSCC organoids that were responding poorly to radiotherapy were reexposed to radiation, but now in the presence of LC161. LC161 is a Second Mitochondria-derived Activator of Caspase (SMAC) mimetic, described to overcome intrinsic cell death resistance by promoting degradation of cellular inhibitor of apoptosis proteins (56). As others have shown an effect of SMAC mimetics in HNSCC cells, either alone or in combination with radiotherapy, we set out to test this in HNSCC organoids (57). Indeed, a radio-sensitizing effect of LC161 was observed in T8. In parallel, T1 and T9 were also tested, but did not show increased sensitivity to radiotherapy in the presence of LC161. In T8 organoids, the addition of SMAC mimetic increased cell death in response to radiation (Fig. 6H). Importantly, LC161 did not radiosensitize the matched N8 organoids; these were more resistant to this treatment. To validate these findings and understand which type of cell death was triggered in response to radiotherapy combined with LC161, the treatment was repeated in the presence of either Z-VAD-FMK (zVAD) or Necrostatin1 (Nec1), which block apoptosis or necroptosis, respectively. It was found that the addition of Nec1, but not zVAD-FMK, could prevent the radiosensitizing effect of the SMAC mimetic (Fig. 6I). These findings show that T8 tumor cells, in line with general tumorigenic mechanisms (58), evade apoptosis, yet that this resistance could be overcome by the addition LC161. These results highlight the potential of organoids to explore alternative (sensitizing) therapies on a personalized level.

In addition to determining radiosensitivity of the tumoroids, we set up an in vitro screening assay combining chemotherapy with radiotherapy. We exposed organoids to a gradient of chemotherapy in the presence (2 Gy) or absence (0 Gy) of radiation. Previously detected differences in sensitivity for these compounds were confirmed (cetuximab sensitivity T5 > T8, cisplatin sensitivity T2 > T1; Fig. 6J and K). We found that combination therapy resulted in increased cell death at lower doses of chemotherapy than when chemotherapy was given as a single agent. To study the effect of chemotherapy in the presence of radiation, we compared organoids treated with chemotherapy and radiotherapy with organoids exposed to radiotherapy alone (T5 and T8 to cetuximab, T1 and T2 to cisplatin; Fig. 6J and K). It seems that in these cases, the effect of chemotherapy itself is not changed by the presence of radiotherapy, although the combination treatment results in additional cell death.

Using the same approach, we studied the radiosensitizing effect of chemotherapy in vitro. In the clinic, such combination therapy is used to treat patients with HNSCC, in both the curative and adjuvant setting. The combination of cisplatin or cetuximab with radiotherapy increases relapse-free survival when compared with treatment with radiotherapy alone in clinical trials (59). Because radiotherapy and chemotherapy are given concurrently to patients, it is impossible to understand whether the effect between these treatments is additive or synergistic. Here, 10 organoid lines (T1–T6, T8, T24, T25, and T27) were exposed to a range of radiotherapy, in either the presence or absence of a toxic dose of cisplatin or cetuximab. Indeed, the combination of chemotherapy and radiotherapy resulted in increased cell death when compared with single-agent treatment. When corrected for the effect of the chemotherapy itself, sensitivity to radiotherapy was increased in the presence of cisplatin in six of ten tested lines, and in four of ten when combined with cetuximab (Fig. 6L and M; Supplementary Fig. S14A). Although not statistically significant, a correlation between the effect of radiotherapy alone and radiotherapy combined with cisplatin was observed. Overall, response to radiotherapy improved in the presence of cisplatin, implying a synergistic effect (Supplementary Fig. S14B). On the basis of the data presented here, this is not the case when radiotherapy is combined with cetuximab (Supplementary Fig. S14C). This suggests that, although additive (Supplementary Fig. S14D and S14E), no synergistic effect of cetuximab and radiotherapy can be observed, at least in vitro.

On the basis of mutations detected in this set of HNSCC-derived tumoroids, we determined the in vitro sensitivity for a range of targeted therapies that are not used in the treatment of patients with HNSCC. As T1, T7, T9, T10, T24, T25, and T29 carry activating mutations in PIK3CA, sensitivity to the PIK3CA inhibitor alpelisib was determined for sequenced tumoroid lines eligible for drug screening (Supplementary Fig. S12). PIK3CA mutation has been investigated as a biomarker for response to PIK3CA inhibitors (60–62). Although some studies imply that activating mutations of PIK3CA can serve as a biomarker for good response, others claim the type of mutation can influence response to these therapeutics. We observe no enrichment for PIK3CA mutations in the organoid lines more responsive to alpelisib (Fig. 7A). Recently, a study linking genetic alterations to alpelisib responses in patients showed that patients carrying H1047R PIK3CA mutations had a more durable response in the clinic, whereas a negative association was found between E545K mutations and alpelisib response (62). When testing all PIK3CA-mutant organoid lines eligible for drug screening, T9 organoids (PIK3CA H1047R, IC₅₀ 0.32 μmol/L) showed increased sensitivity to alpelisib compared with the E545 PIK3CA–mutant organoids T1 (IC₅₀ 3.19 μmol/L) and T7 (IC₅₀ 2.69 μmol/L). However, the IC₅₀ value of organoid line T25, also carrying a H1047R activating mutation in PIK3CA, was 2.01 μmol/L (Fig. 7B). Although still lower than the IC₅₀ values obtained for the E545K-mutant organoid lines, these data support the hypothesis that PIK3CA mutation status cannot predict response to PIK3CA inhibitors in all cases, and functional testing might be required to identify responders.

BRAF V600E mutations were detected in T9 and T10. Therefore, we tested a panel of tumoroid lines for sensitivity to vemurafenib, a BRAF inhibitor (Fig. 7C). As expected, we observed increased sensitivity of BRAF-mutant T9 when compared with the other tumoroid lines tested in this assay.

Finally, we exposed the panel of tumoroids to the PARP inhibitor niraparib, the mTOR inhibitor everolimus, and the FGFR inhibitor AZD4547 (Fig. 7D–F). These compounds were selected on the basis of the genetic alterations described in HNSCC (30). Interestingly, although mutations in PARP, MTOR, and FGFR were not detected in the characterized tumoroid lines, we observed variable sensitivities to these compounds (Supplementary Fig. S12). For all tested compounds, the difference in response between matched normal and tumor organoids was tested (Supplementary Fig. S15). Responses to therapies remained stable over time (Supplementary Fig. S16). Among the lines sensitive to everolimus (Fig. 7D) was tumor line T27, which carries a RET mutation. Inhibition of mTOR signaling has been shown to inhibit the growth of RET-mutant tumor cells in vitro (63). AZD4547 was also found to effectively target tumoroid line T27 (Fig. 7E). This response can likely be explained by the mutation in KDR (also known as VEGFR2A) detected in T27 organoids. Indeed, AZD4547 has been shown to target KDR in addition to FGFR1, although with lower affinity (64). Taken together, these data reveal differences in sensitivity to all compounds tested between different tumoroid lines, which in some, but not all, cases can be explained by detected genetic alterations. Thereby, these data illustrate the potential of this technology to guide personalized therapy in the future.

We present a protocol for long-term expansion of wild-type oral mucosal epithelia and HNSCC in the form of organoids. The growth factor cocktail that is used in this work differs from that for other epithelia. The addition of the GSK3ß inhibitor CHIR and the growth factor FGF2 appears essential for the successful establishment of these organoids. We characterize HNSCC organoids by histology, gene-expression analysis and, in the case of tumoroids, by mutational profiling. Using the technique described here, we can retain normal keratinocytes without the requirement for feeder cells in culture for >15 passages. To document that these organoids are amenable to infectious disease modeling, we performed infection with HSV1 and HPV16. Previous studies have applied 3-D models of fibroblasts, vaginal epithelium, and melanoma cells to study the interaction of HSV1 with epithelial cells (65, 66). Yet to our knowledge a 3-D model to study HSV1 infection in oral mucosa cells has not been reported. Infection with HPV16 has been limited to studies in immortalized cell lines, or to primary cells that can be cultured only briefly (21). Essential for HPV infection is viral access to basal cells of the stratified epithelium, whereas for virion production, differentiated keratinocytes are essential. A model that retains this stratification in vitro and shows productive infection, such as the one described here, should therefore be valuable to research in this field.

We report the initiation of an HNSCC-derived organoid biobank, which is being expanded significantly. Like our other biobanks, these will be made available to the research community. The biobank was used to test sensitivity to a panel of drugs. This drug panel included both drugs currently used in the treatment of patients with HNSCC and targeted drugs that based on their molecular targets appeared relevant to be tested in this setting. The fact that a variable response to cisplatin, carboplatin, cetuximab, and radiotherapy was observed in vitro suggests that this method records interindividual differences. This implies that tumor-derived HNSCC organoids hold potential to guide personalized therapy. Vlachogiannis and colleagues recently showed the predictive potential of similar drug screens using organoids derived from metastasized gastrointestinal cancers (23).

The high take rate and the fast growth rate of the HNSCC tumoroids makes them particularly suited for personalized approaches. In current clinical practice, treatment options for HNSCC are dictated by site, stage, and patient factors. There is a need for more tailored treatment. For this reason, we explored the potential to correlate organoid responses to patient outcome. Clinical data were available for 7 patients who presented with tumors in different anatomic locations. They received different total dosages of radiotherapy, while all were therapy-naïve at the start of treatment. Although numbers should still be considered small, we find that clinical response of patients treated with radiotherapy can be correlated to in vitro responses of the corresponding organoids. To investigate whether such a correlation truly exists, more cases are required, especially considering the heterogeneity of this disease and other variables that influence patient outcome (total delivered dose, if surgery is performed and, if so, if resection is successful, ability to finish treatment regimen as prescribed, etc.). To follow up on these initial findings, we are embarking on an observational study to link patient outcome to organoid responses in vitro (ONCODE-P2018-0003). We anticipate including approximately 80 patients. Organoids will be exposed to cisplatin, carboplatin, or cetuximab, and combined with irradiation such as occurs in clinical practice. Here, we already show the possibility of combining chemotherapy and radiotherapy in vitro. The observational trial will elucidate whether these in vitro results hold predictive potential for patient responses.

Cetuximab is used in the treatment of patients with HNSCC. Fifty percent to 90% of tumors overexpress EGFR (47), and 15% carry gene amplification of EGFR (30). Initially, it was expected that differing levels of EGFR expression could explain patient response to anti-EGFR therapy. However, it has been shown that neither EGFR overexpression nor increased gene copy number can serve as a prognostic biomarker for cetuximab response (47). In agreement, the cetuximab sensitivity of the tumoroids could not be correlated by their EGFR expression levels. It will be interesting to see whether this correlation can be observed in larger panels of patient-derived organoids. A direct, functional test such as offered by patient-derived tumoroids may prove valuable.

Because chemotherapy and radiotherapy are given concurrently to patients, it is difficult to evaluate the effect of these single treatments at the individual patient level. As shown here, organoids can be used to study the effect of these treatments as single agents, but also can be used to study how chemotherapy influences radiotherapy efficacy and vice versa. As such, our results suggest that cetuximab does not act as a radiosensitizer in patient-derived HNSCC cells. Regardless, compared with radiotherapy alone, the combination of cetuximab and radiotherapy results in increased cell death in vitro, fitting with clinical data (67).

In addition, screening for sensitivity to other targeted drugs may be of value when the tumor recurs. The targeted therapies tested in our studies (everolimus, niraparib, alpelisib, AZD4547, and vemurafenib) yielded differential responses. Although phase I and II clinical trials are ongoing to test a number of these drugs in patients with HNSCC (68–70), implementation remains limited. Organoids might serve as an in vitro tool to select the right therapy for the right patient. Alternatively, studying the characteristics of those organoids that are responding in such screens might help identify biomarkers that distinguish responders from nonresponders. As an example, in this study we show that PIK3CA-mutant tumoroid lines show variable responses to alpelisib. PIK3CA E545K–mutant lines showed a similar response to the other PIK3CA wild-type tumoroids in our panel, whereas H1047R-mutant lines showed increased sensitivity to this agent (although the extent of sensitivity differed for the two H1047R-mutant lines tested here). In patients, differences in alpelisib responses could indeed be linked to the type of PIK3CA mutation, where H1047R-mutant patients showed more durable responses than patients harboring a E545K mutation (62). On the contrary, other studies have implicated PIK3CA mutation as a predictor for alpelisib response, regardless of the specific mutation (60, 61). Our data support the hypothesis that the type of PIK3CA mutation might be relevant for the response to alpelisib. In organoids, the presence of PIK3CA activating mutations did not correlate with in vitro responses to alpelisib. These differences in sensitivity between lines carrying identical PIK3CA mutations might be explained by other genetic alterations, interfering with the dependence on this pathway for cell survival.

In vitro responses observed in this study could in some, but not all, cases be explained by the genetic alterations detected in the tumoroid lines. For example, the response of tumoroid line T27 to AZD4547 can be explained by the mutation in KDR, as AZD4547, designed as an FGFR inhibitor, is known to also target KDR with high affinity (64). Interestingly, in the panel tested here, there are tumoroid lines showing comparable sensitivity to this therapeutic agent, although at this point we cannot identify genetic alterations explaining this sensitivity. These results indicate that, at least in some cases, functional tests such as those performed here might be more informative than genetic screening to predict response to therapeutic agents.

In an attempt to detect general therapy resistance mechanisms, we performed differential gene expression on a panel of organoid lines that responded either well or poorly to cisplatin/carboplatin, cetuximab, or radiotherapy. Under current conditions (sequencing depth, number of samples included in the analysis, etc.), no indications for resistance mechanisms could be deduced from this data by gene enrichment analysis. Differently expressed gene lists did contain genes previously described to be associated with therapy resistance. For example, genes enriched in radiotherapy-resistant organoid lines included PGK1 and GPX3. For both genes, high expression has previously been associated with poor response to radiotherapy or DNA-damaging therapies (52, 53). Expression of IGFBP5 and SLC16A3, both enriched in cetuximab-resistant organoid lines, have previously been reported to be biomarkers for poor response to cetuximab (54, 55).

Finally, extension of this system to the generation of immune-tumoroids, tumor-derived organoids combined with immune cells, might make this system suitable to test responses to immunomodulating antibodies. Such drugs are currently evaluated in patients with HNSCC (71). Again, prognostic biomarkers are largely lacking. Initial studies demonstrate that tumoroids can be cocultured with immune cells (derived from either peripheral blood or tumor microenvironment) and can provide functional readouts (17, 72).

The collection of patient data and tissue for the generation and distribution of organoids has been performed according to the guidelines of the European Network of Research Ethics Committees (EUREC) following European, national, and local law. The Biobank Research Ethics Committee of the University Medical Center Utrecht (TCBio) approved the biobanking protocol: 12-093 HUB-Cancer according to the University Medical Center Utrecht (UMCU) Biobanking Regulation. All donors participating in this study signed informed-consent forms and can withdraw their consent at any time, leading to the prompt disposal of their tissue and any derived material, as well as the cessation of data collection. Available organoids will be cataloged at www.hub4organoids.eu and can be requested at [email protected].

Patient material was collected from pathology material in Advanced DMEM/F12 (Life Technologies, catalog no. 12634-034), supplemented with 1× GlutaMAX (adDMEM/F12; Life Technologies, catalog no. 12634-034), Penicillin–streptomycin (Life Technologies, catalog no. 15140-122), and 10 mmol/L HEPES (Life Technologies, catalog no. 15630-056). This medium was named Advanced DMEM +/+/+. For collection of patient material, 100 μg/mL Primocin (Invivogen, catalog no. ant-pm1) was added to the +/+/+ medium. For normal tissue samples, excess fat or muscle tissue was removed to enrich for epithelial cells, and tissue was cut into small fragments. Random pieces of approximately 5 mm³ were stored at −20°C for DNA isolation. Some pieces were fixed in formalin for histopathologic analysis and IHC, and the remainder was processed for organoid derivation. Fragments were incubated at 37°C in 0.125% Trypsin (Sigma, catalog no. T1426) in +/+/+ until digested. Every 10 minutes, the tissue suspension was sheared using 1 mL pipette. Digestion was monitored closely to prevent excess incubation in trypsin. Incubation was performed for a maximum of 60 minutes. When complete, Trypsin was diluted by addition of 10 mL +/+/+. Suspension was strained over a 100 μm EasyStrainer filter (Greiner, catalog no. 542000) and centrifuged at 1,000 rpm. The resulting pellet was resuspended in ice-cold 70% 10 mg/mL cold Cultrex growth factor–reduced BME type 2 (Trevigen, 3533-010-02) in organoid medium. Droplets of approximately 10 μL were plated on the bottom of preheated suspension culture plates (Greiner, catalog no. M9312). After plating, plates were inverted and put at 37°C for 30 minutes to let the BME solidify. Subsequently, prewarmed organoid medium was added to the plate. For the first week, 10 μmol/L Rho-associated kinase (ROCK) inhibitor Y-27632 (Abmole Bioscience, catalog no. M1817) was added to the medium to aid outgrowth of organoids for the primary tissue. For mouse-derived organoids, tongue tissue was obtained from control mice used under Instantie voor Dierenwelzijn (IvD)–approved projects. Subsequent processing of tissue was identical to processing of human tissue.

HNSCC and normal epithelium-derived organoids were grown in Advanced DMEM +/+/+. Organoid medium contained 1 × B27 supplement (Life Technologies, catalog no. 17504-044), 1.25 mmol/L N-acetyl-l-cysteine (Sigma-Aldrich, catalog no. A9165), 10 mmol/L Nicotinamide (Sigma-Aldrich, catalog no. N0636), 50 ng/mL human EGF (PeproTech, catalog no. AF-100-15), 500 nmol/L A83-01, 10 ng/mL human FGF10 (PeproTech, catalog no. 100-26), 5 ng/mL human FGF2 (PeproTech, catalog no. 100-18B), 1 μmol/L Prostaglandin E2 (Tocris Bioscience, catalog no. 2296), 0.3 μmol/L CHIR 99021 (Sigma-Aldrich, catalog no. SML1046), 1 μmol/L Forskolin [Bio-Techne (R&D Systems) catalog no. 1099], 4% R-spondin, and 4% Noggin (both produced via the r-PEX protein expression platform at U-Protein Express BV). Mouse organoids were maintained similar to human organoids, but were grown in +/+/+, containing B27, 25 mmol/L N-acetyl-l-cysteine, 10 mmol/L Nicotinamide, 2% RSPO, 50 ng/mL EGF, and 10 ng/mL FGF10. Organoids were split between 7 and 14 days after initial plating. For passaging, organoids were collected from the plate by disrupting the BME droplets with a P1000, collecting and washing in 10 mL +/+/+. Pellet was resuspended in 1 mL of TrypLE Express (Life Technologies, catalog no. 12605-010) and incubated at 37°C. Digestion was closely monitored and suspension was pipetted up and down every 5 minutes to aid disruption of the organoids. TrypLE digestion was stopped when organoids were disrupted into single cells by adding 10 mL +/+/+. Cells were subsequently resuspended in ice-cold 70% BME in organoid medium and plated at suitable ratios (1:5 to 1:20) to allow efficient outgrowth of new organoids. After splitting, 10 μmol/L Y-27632 was always added to aid outgrowth of organoids from single cells. Medium was changed every 2–3 days and organoids were split once every 1–2 weeks.

Organoids were cultured as normal. For differentiation of the organoids (as shown in Fig. 1C), organoids were split to single cells, left to grow one week on organoid medium, and then put on +/+/+ for one week before collection. On the day of collection, organoids were collected from tissue culture plates and washed twice in 10 mL +/+/+. RNA was extracted using RNeasy Mini Kit (Qiagen, catalog no. 74104) according to protocol. RNA amount was measured using NanoDrop. For quantification of EGFR expression, organoids were split to single cells, left to grow 5 days on organoid medium, and then put on organoid medium with lower EGF concentration (0.63 ng/mL).

RNA was processed as described previously, following the protocol of CEL-Seq (73, 74). Paired-end sequencing was performed on the Illumina Nextseq500 platform, High Output 2 × 75 bp run mode. Read 1 was used to identify the Illumina library index and CEL-Seq sample barcode. Read 2 was aligned to the hg19 human RefSeq transcriptome using BWA (75). Reads that mapped equally well to multiple locations were discarded. Around 2 million reads were mapped per sample. Samples with low number of reads were removed. Sample annotation and barcodes can be found on the GEO submission of this data. The remaining samples were normalized and analyzed by the DESeq2 package (32). For visual comparison between samples, regularized log transformed (rlog) values were used.

For cDNA synthesis, RNA was incubated with 50 μg/mL Oligo (dT) 15 Primer (Promega, catalog no. C1101) in water for 5 minutes at 70°C. Subsequently GoScript Reverse Transcriptase (Promega, catalog no. A5003) was used according to protocol to produce cDNA. qPCR reactions were performed in 384-well format using IQ SYBR green (Bio-Rad, catalog no. 1708880) in the presence of 0.67 μmol/L forward and reverse primer and cDNA transcribed from 25 ng RNA. For qPCR, samples were incubated for 2 minutes at 95°C and for 40 cycles at: 15 seconds at 98°C, 15 seconds at 58°C, and 15 seconds at 72°C. Results were calculated by using the ΔΔC_t method. Expression was calculated relative to expression in tongue tissue (total RNA, human normal tongue tissue, AmsBio, catalog no. R1234267). Melt peak analysis was performed to assure that primer had no aspecific binding. Primers used were the following:

• Human p63 forward: GACAGGAAGGCGGATGAAGATAG,

• Human p63 reverse: TGTTTCTGAAGTAAGTGCTGGTGC,

• Human Ki-67 forward: GAGGTGTGCAGAAAATCCAAA,

• Human Ki-67 reverse: CTGTCCCTATGACTTCTGGTTGT,

• Human KRT13 forward: GACCGCCACCATTGAAAACAA,

• Human KRT13 reverse: TCCAGGTCAGTCTTAGACAGAG,

• Human KRT4 forward: CTCTTTGAGACCTACCTCAGTGT,

• Human KRT4 reverse: GGCTGCTGTGCGTTTGTTG,

• Human EGFR forward: AGGCAGGAGTAACAAGCTCAC,

• Human EGFR reverse: ATGAGGACATAACCAGCCACC,

• Human Actin forward: TGCGTGACATTAAGGAGAAG,

• Human Actin reverse: TGAAGGTAGTTTCGTGGATG,

• Human GAPDH forward: GGAGCGAGATCCCTCCAAAAT,

• Human GAPDH reverse: GGCTGTTGTCATACTTCTCATCG.

DNA was isolated using Reliaprep gDNA Tissue Miniprep System (Promega, catalog no. A2052) according to protocol. DNA concentration were measured using NanoDrop.

Single cells (100,000) were plated in 50 μL of BME. After one week in culture, all organoids were collected and disrupted into single cells. Cells were counted and total cell number was determined. Counting was performed four times. By calculating the number of cells at day 7, which all came from the 100,000 cells plated at day 0, a multiplication factor could be determined for each week. Using this, a theoretical total number of cells could be calculated by multiplying the total cell number of the previous week with the multiplication factor of that week. Subsequently, 100,000 of the counted single cells were plated in 50 μL BME. This procedure was repeated for five weeks.

For imaging experiments, cells were incubated with 1 × 10⁷ PFU HSV-dTomato virus in the culture medium. The virus was a kind gift of Prashant Desai (John Hopkins University, Baltimore, MD). For DNA quantification, organoids were split using TrypLE. On the third day after splitting, organoids were incubated with 1 × 10⁷ PFU HSV-dTomato virus in suspension for 6 hours. After washing with 10 mL +/+/+, organoids were plated (1,500 organoids in 20 μL BME per well) in 48-well format. Organoids were kept in organoid medium, with or without 10 μmol/L acyclovir (Sigma). For DNA collection, the BME drop was collected together with culture medium and added to 10 mL +/+/+ in a 15 mL falcon tube. After centrifugation, medium was removed and pellet was stored at −20°C until gDNA extraction. For DNA quantification, qPCR reactions were performed in 384-well format using IQ SYBR Green Mix (Bio-Rad) in the presence of 0.67 μmol/L forward and reverse primer and 2% of total DNA isolated from 1,500 organoids. After gDNA extraction, qPCR was performed with the following primers to detect HSV DNA: forward: 5′-ATCAACTTCGACTGGCCCTT-3′ and reverse: 5′-CCGTACATGTCGATGTTCAC-3′. PCR program used: 2 minutes at 95°C and for 40 cycles at: 15 seconds at 98°C, 15 seconds at 60°C, and 15 seconds at 72°C. Increase in DNA content was calculated relative to noninfected wells.

HPV16 virions were produced as described previously (26). Upon fractionation of the supernatant containing the virus, fractions with highest titer (as determined by quantitative PCR on HPV DNA) were pooled and subsequently used for infection experiments. Organoids were split using TrypLE and plated at a density of 1,500 cells/well, in 20 μL BME drops. After the addition of culture media, HPV-containing supernatant was added to the wells. During the course of the experiment, medium was refreshed every 2–3 days. DNA isolation and DNA quantification were performed as described for HSV infections, except primers used were: forward: 5′-CTACATGGCATTGGACAGGA-3′ and reverse: 5′-GGTCACGTTGCCATTCACTA-3′. For reinfection experiments, supernatant taken from organoids cultured for 12 days after HPV infection was collected and filtered with 0.45-μm pore filter. Subsequently, this was added to the uninfected organoids.

Oncopanel sample prep and analysis was performed as described previously (76) and sequenced with the mpliSeq Cancer Hotspot Panel V2+ (for details: https://www.umcutrecht.nl/getmedia/c39cd469-a4de-4ae9-9a52-0b8ed6761311/CHPv2Plus_NGS.pdf.aspx). For sample T27, an extended oncopanel was used, where 140 genes were checked with 3,817 probes targeting a region of 1,080,437 bp. For library preparation, SureSelectXT Library Prep Kit was used following the SureSelectXT Target Enrichment System for Illumina Version B.2 protocol. For cluster generation, the library is loaded into a flow cell where fragments are captured on a lawn of surface-bound oligos complementary to the library adapters. Each fragment is then amplified into distinct, clonal clusters through bridge amplification. Total reads were above 50,000,000 (52 < GC% > 50).

Whole-exome sequencing data were mapped against human reference genome GRCh37, and variants were called using the IAP pipeline (https://github.com/UMCUGenetics/IAP). To obtain high-quality somatic mutation catalogs, we filtered out variants with evidence in their corresponding normal samples, overlaps with the single nucleotide polymorphism database v137.b3730, and the variants did not reach our quality measurements (base coverage of 10×, VAF of 0.1, GATK phred-scaled quality score of 100 for base substitutions, 250 for indels and mapping quality of 60 for indels). Indels that were present within 100 bp of a called variant in the control were excluded. Only autosomal variants were considered. The scripts used for the filtering are available at: https://github.com/UMCUGenetics/SNVFI, and https://github.com/ToolsVanBox/INDELFI. Nonsynonymous mutations (missense mutation, start loss, stop gain, in-frame insertion/deletion, and frame shift) in the genes checked in the OncoPanel were reported as driver mutations. Whole-exome sequencing data has been deposited in the European Genome–Phenome Archive (EGA; https://www.ebi.ac.uk/ega/home); accession number EGAS00001003628.

Two days prior to start of the drug screen, organoids were passaged and disrupted into single cells using TrypLE. Single cells were plated in 70% BME in organoid medium as for regular splitting. Two days later, organoids were collected from the BME by the addition of 1 mg/mL dispase II (Sigma-Aldrich, catalog no. D4693) to the medium of the organoids. Organoids were incubated for 30 minutes at 37°C to digest the BME. Subsequently, organoids were filtered using a 70-mm nylon cell strainer (BD Falcon), counted, and resuspended in 5% BME/growth medium (12,500 organoids/mL) prior plating in 40 μL volume (Multi-drop Combi Reagent Dispenser, Thermo Scientific, catalog no. 5840300) in 384-well plates (Corning, catalog no. 4588).

The drugs were added 1 hour after plating the organoids using the Tecan D300e Digital Dispenser (Tecan). Nutlin-3 (Cayman Chemical, catalog no. 10004372), Niraparib (Selleckchem, catalog no. S2741), AZD4547 (ApeXbio, catalog no. A8250), everolimus (LC Laboratories, catalog no. E4040), vemurafenib (Selleckchem, catalog no. S1267), and alpelisib (LC Laboratories, catalog no. A4477) were dissolved in DMSO. Cisplatin (Sigma, catalog no. C2210000), carboplatin (Sigma, catalog no. C2538), and cetuximab (obtained from hospital pharmacy) were dissolved in PBS containing 0.3% Tween-20, which was required to dispense these drugs using the HP printer. All wells were normalized for solvent used. DMSO percentage never exceeded 1%, PBS/Tween-20 percentage never exceeded 2%. Drug exposure was performed in triplicate for each concentration shown. For a layout of the drug screen, see Supplementary Fig. S12.

One hundred twenty hours after adding the drugs, ATP levels were measured using the CellTiter-Glo 3-D Reagent (Promega, catalog no. G9681) according to the manufacturer's instructions, and luminescence was measured using a Spark multimode microplate reader (Tecan). Results were normalized to vehicle (100%) and baseline control (Staurosporin 1 μmol/L; 0%). For each line, when viability did not go above 70% or below 30%, an additional screen was performed for that particular drug with an adjusted dose of this drug for this organoid line. Screen quality was determined by checking Z factor scores for each plate following this formula:

Drug screens with a Z factor of <0.3 were not used and repeated. Kill curves were produced using GraphPad software and lines were fitted using the option “log (inhibitor) versus normalized response-variable slope.”

Organoids were disrupted into single cells using TrypLE, and plated at a density of 6,000 single cells in 30 μL BME drops in a 48-well plate. Two days later, cells were irradiated. For each radiation dose, a separate plate was used. Plates were sealed air-tight and irradiated with a single fraction of 0–8 Gy using a linear accelerator (Elekta Precise Linear Accelerator 11F49, Elekta). The plates were positioned on top of 2-cm polystyrene and submerged in a 37°C water bath. After radiation, medium was changed. Four days later, readout was performed as described previously.

For testing a range of chemotherapy concentrations in the presence or absence of radiotherapy, drug screens were performed as described in the section “In vitro drug screen.” Two identical plates of organoids were treated with the desired chemotherapy. One day after the start of treatment, one of these plates was irradiated with a total dose of 2 Gy. This dose was chosen as a sublethal dose based on the radiotherapy-only experiments that were previously performed.

To test a range of radiation doses in the presence or absence of cisplatin, cetuximab, LC161, zVAD-FMK, or Necrostatin 1, organoids were disrupted into single cells using TrypLe, and plated at a density of 6,000 organoids per 20 μL BME in 96-well plates. One day later, chemotherapy treatment was started. On the basis of chemotherapy screens that were previously performed, the sublethal doses of 5 μmol/L cisplatin and 30 μg/mL cetuximab were chosen. On the basis of literature, 1 μmol/L LC161, 10 μmol/L zVAD, and 10 μmol/L Necrostatin were used. One day later, cells were irradiated as described above. Four days later, readout was performed.

Organoids were infected with lentivirus encoding mNeon-tagged histone 2B and a puromycin-resistant cassette (77). After selection, organoids were plated in BME in glass-bottom 96-well plates and mounted on an inverted confocal laser scanning microscope (Leica SP8X), which was continuously held at 37°C and 5% CO₂. Over 16 to 20 hours, approximately 10 H2B-mNeon–expressing organoids were imaged simultaneously in XYZT mode using a ×40 objective (N.A. 1.1), using minimal amounts of 506 nm laser excitation light from a tunable white light laser. Time interval was approximately 3 minutes (2:30–3:20 minutes). Cell divisions were scored, judged, and counted manually.

For all in vivo work, ethical approval was gained prior to the start of this project by the Central Authority for Scientific Procedures on Animals (CCD) and the local animal experimental committee at the Hubrecht Institute (IvD-HI-KNAW-HI 17.10.11). Five days before transplantation, organoids were disrupted to single cells and plated as usual. On the day of transplantation, organoids were disrupted into single cells and resuspended in 50% BME/organoid medium at a density of 33.33 million cells per mL. Cells (2.5 million) were subcutaneously injected in NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ mice between 6 and 12 weeks of age. Six weeks after injection, mice were sacrificed by cervical dissociation, and tumors were excised and fixed overnight in 4% formaldehyde.

Tissues or organoids were fixed in 4% paraformaldehyde overnight, dehydrated, and embedded in paraffin. Sections were subjected to H&E as well as IHC staining. The details on primary antibodies used for IHC staining on organoids and primary tissue are given in Supplementary Table S7. Staining for TP40, TP53, MKI67, and KRT5 on the tumoroids (Fig. 2) was performed at the pathology department of the UMCU (Utrecht, the Netherlands).

Two days after splitting, organoids were treated with 0.1 μg/mL Colcemid (Gibco 15212012) for 17 hours in organoid medium. After that, organoids were disrupted into single cells using TrypLE and processed as described previously (78). Metaphase spreads were mounted with DAPI-containing VectaShield (Vector Laboratories, catalog no. H-1200) and imaged on a DM6000 Leica microscope.

Organoids were collected and BME was removed using Cell Recovery Solution (Corning). To fix organoids, 1 mL of 1% (v/v) glutaraldehyde (Sigma-Aldrich, G5882) in PBS was added. After overnight fixation at 4°C, organoids were transferred onto 12-mm coverslips (Corning, catalog no. 354085). Samples were dehydrated by consecutive 10-minute incubations in 2 mL of 10% (v/v), 25% (v/v), and 50% (v/v) ethanol-PBS, 75% (v/v), and 90% (v/v) ethanol-H₂O (2x) followed by 50% ethanol-hexamethyldisilazane (HMDS), and 100% HMDS (Sigma-Aldrich, catalog no. 379212). Coverslips were removed from the 100% HMDS, air-dried overnight at room temperature, and mounted onto 12-mm specimen stubs (Agar Scientific). After gold-coating to 1 nm using a Q150R sputter coater (Quorum Technologies), samples were examined with a Phenom PRO table-top scanning electron microscope (Phenom-World).

Organoids were placed in BME on 3-mm diameter and 200-μm depth standard flat carriers for high-pressure freezing and immediately cryoimmobilized using a Leica EM high-pressure freezer (equivalent to the HPM10), and stored in liquid nitrogen until further use. They were freeze-substituted in anhydrous acetone containing 2% osmium tetroxide 0.1% uranyl acetate at −90°C for 72 hours and warmed to room temperature, 5°C per hour (EM AFS-2, Leica). The samples were kept 2 hours at 4°C and 2 additional hours at room temperature. After acetone rinses (4 × 15 minutes), Epon resin infiltration was performed during 2 days (acetone: resin 3:1, 3 hours; 2:2, 3 hours; 3:1, overnight; pure resin: 6 hours + overnight + 6 hours + overnight + 3 hours). Resin was polymerized at 60°C during 96 hours. Leica Ultracut UC6 ultramicrotome was used to cut sections that were mounted on Formvar-coated copper grids and stained with 2% uranyl acetate. Sections were observed in a Tecnai T12 Spirit equipped with an Eagle 4kx4k camera (FEI Company), and large EM overviews were collected using the principles and software described by Ravelli and colleagues (79).

E. Driehuis is an inventor on multiple patents related to organoids. D.A. Tuveson is on the scientific advisory board at Leap Therapeutics and Surface Oncology, reports receiving commercial research grants from Fibrogen and ONO, has ownership interest (including stock, patents, etc.) in Leap Therapeutics and Surface Oncology, andis a consultant/advisory board member for Surface Oncology and Leap Therapeutics. H. Clevers is an inventor on multiple patents related to organoids. No potential conflicts of interest were disclosed by the other authors.

Conception and design: E. Driehuis, S. Spelier, L.A. Devriese, P.J. Peters

Development of methodology: E. Driehuis, S. Kolders, S. Spelier, P.J. Peters

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): E. Driehuis, S. Kolders, K. Lõhmussaar, S.M. Willems, L.A. Devriese, R. de Bree, E.J. de Ruiter, J.H. van Es, V. Geurts, G.-W. He, R.H. van Jaarsveld, J. Vivié, M.M.J.M. Zandvliet, N. Iakobachvili, P. Sridevi, O. Kranenburg, D.A. Tuveson, P.J. Peters

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): E. Driehuis, S. Spelier, K. Lõhmussaar, S.M. Willems, L.A. Devriese, R.H. van Jaarsveld, R. Oka, M.J. Muraro, N. Iakobachvili, R. van Boxtel, P.J. Peters

Writing, review, and/or revision of the manuscript: E. Driehuis, S.M. Willems, L.A. Devriese, R. de Bree, A.P.A. Hendrickx, H. Clevers

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): E. Driehuis, S. Kolders, S.M. Willems, R. de Bree, E.J. de Ruiter, J. Korving, H. Begthel, J.H. van Es, M.M.J.M. Zandvliet, A.P.A. Hendrickx, O. Kranenburg

Study supervision: G.J.P.L. Kops, P.J. Peters, A. van Oudenaarden, H. Clevers

We thank Tulay Bayram for support regarding ethical regulatory affairs. We acknowledge Anneta Brousali, Petra van der Groep, Alexander Constantinides, and Anne Snelting of the Utrecht Platform for Organoid Technology (U-PORT; UMC Utrecht) for patient inclusion and tissue acquisition. We would like to thank the Molecular Diagnostics Department of the UMCU, Department of Pathology. Special thanks to Marja Blokland, Carmen Vooijs, and Wendy de Leng for help with targeted sequencing of the organoids. We thank Prashant Desai of John Hopkins University for contributing fluorescently labeled HSV virus. We thank Dohun Pyeon and Sharon Kuss-Duerkop of Michigan State University for contributing HPV16 virus, and for their guidance while setting up these infections. We would like to acknowledge the members of the Microscopy Core Lab at M4I Maastricht University for their scientific and technological support regarding high-pressure freezing and sample preparation for electron microscopy, with a special mention of Carmen López-Iglesias and Hans Duimel. We thank Yorick Post, Frans Schutgens, and Joep Beumer for critically reading this manuscript. This work was funded by the Oncode Institute (partly financed by the Dutch Cancer Society) and by the gravitation program CancerGenomiCs.nlfrom the Netherlands Organization for Scientific Research (NWO) and a ZonMw grant (116.006.103). Research supported by an SU2C-DCS International Translational Cancer Research Dream Team Grant (SU2C-AACR-DT1213). Stand Up To Cancer (SU2C) is a division of the Entertainment Industry Foundation. Research grants are administered by the American Association for Cancer Research, the Scientific Partner of SU2C.

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.