Establishment and characterization of patient-derived head and neck cancer models from surgical specimens and endoscopic biopsies

Head and neck squamous cell carcinoma (HNSCC) is heterogeneous in etiology, phenotype and biology. Patient-derived xenografts (PDX) maintain morphology and molecular profiling of the original tumors and have become a standard “Avatar” model for human cancer research. However, restricted availability of tumor samples hindered the widespread use of PDX. Most PDX-projects include only surgical specimens because reliable engraftment from biopsies is missing. Therefore, sample collection is limited and excludes recurrent and metastatic, non-resectable cancer from preclinical models as well as future personalized medicine. This study compares the PDX-take rate, -growth, histopathology, and molecular characteristics of endoscopic specimens with surgical specimens. HNSCC samples (n = 55) were collected ad hoc, fresh frozen and implanted into NOD.Cg-PrkdcscidIl2rgtm1Wjl/SzJ mice. Engraftment was successful in both sample types. However, engraftment rate was lower (21 vs. 52%) and growth delayed (11.2 vs. 6.7 weeks) for endoscopic biopsies. Following engraftment, growth kinetic was similar. Comparisons of primary tumors and corresponding PDX models confirmed preservation of histomorphology (HE histology) and molecular profile (Illumina Cancer Hotspot Panel) of the patients’ tumors. Accompanying flow cytometry on primary tumor specimens revealed a heterogeneous tumor microenvironment among individual cases and identified M2-like macrophages as positive predictors for engraftment. Vice versa, a high PD-L1 expression (combined positive score on tumor/immune cells) predicted PDX rejection. Including biopsy samples from locally advanced or metastatic lesions from patients with non-surgical treatment strategies, increases the availability of PDX for basic and translational research. This facilitates (pre-) clinical studies for individual response prediction based on immunological biomarkers.


Background
Head and neck squamous cell cancer (HNSCC) is the 7 th most common cancer worldwide and associated with a poor outcome [1][2][3][4][5]. Despite aggressive surgery, radiation-and chemotherapy, ~ 50% of patients die, while survivors suffer from pain, dysphagia and dysphonia [6]. Today, recurrence, and treatment response are difficult to predict, because of inter-and intratumoral heterogeneity in etiology, phenotype, and biology. Preclinical models must represent this heterogeneity to identify predictive biomarkers and develop effective personalized medicine. Patient-derived xenografts (PDX), generated by implantation of human cancer tissue into immunodeficient mice, are considered as the gold standard for preclinical cancer research [7][8][9][10][11]. In early passages, PDXs faithfully recapitulate the original tumors' cellular, molecular and histopathological structures as well as drug response and clinical outcome [8,12]. Thus, PDX provide an excellent platform for translational research on biomarkers and drug development (including setup of clinical trials) [12][13][14][15].
In lung-, gastrointestinal-, pancreatic-, bladder-and skin cancer PDX-projects, the lack of surgical specimens has led to the use of endoscopic-and needle biopsies [15,22,23]. Engraftment rates were lower for biopsies (33-60%) vs. surgery specimens (40-100%), but PDX formation from minimally invasive procedures was possible. For HNSCC, endoscopic PDX sampling appears technically suitable, because of exophytic tumor growth and diagnostic sampling from both the vital margin and the necrotic center of the lesion. Two recent HNSCC studies regularly used biopsies for PDX formation [24,25]. Lilja-Fischer et al. reported a biopsy take rate of 33% in oropharyngeal cancer [24] while Kang et al. observed 100% engraftment efficacy for biopsies in a small and defined patient cohort [26]. Therefore, HNSCC PDX biopsy engraftment appears to be feasible. However, take rate, time to engraftment and contributing factors of endoscopic biopsies for HNSCC PDX formation remain unclear.
In this study, we describe a practical and straightforward method to establish p16 positive and negative PDX models using endoscopic and surgical tumor samples. Performing a side-by-side comparison, we show that the engraftment rate was lower for biopsies, but PDX had similar growth kinetics once established. With this setup, a PDX library was created including clinical characteristics, pathological analysis as well as molecular and immunological data of the tumor microenvironment.

Materials and methods
Tumor sample preparation HNSCC samples were donated from consecutive patients undergoing an endoscopic biopsy or surgery at a University Medical Center from 06/2018 through 01/2020. Patients with pathologically proven or suspicion of HNSCC and the following characteristics were eligible: 1) HNSCC of the oral cavity, oropharynx, hypopharynx, larynx and neck lymph node metastases; 2) size of the tumor > 2 cm; 3) primary disease or recurrence; 4) > 18 years of age. Written informed consent was obtained according to the local Ethics Committee (reference number A2018-0003) and the guidelines for the use of human material. Blood sampling (2 × 7.5 ml Heparin) was performed prior to endoscopy/surgery. Endoscopic biopsies were sampled using endoscopic scissors as required for the routine diagnostic procedure. Surgery specimens were obtained by open (e.g. pharyngotomy, laryngectomy) or transoral approach (e.g. laser, radio frequency).
Immediately after dissection, the samples were sent to the Institute of Pathology (in room temperature NaCl 0.9%) for instantaneous H&E section. The pathologist removed tumor tissue for routine diagnostics, necrotic/ fibrotic areas and provided macroscopically vital tumor tissue for the experimental laboratory. All specimens of at least 5 × 5 × 5 mm 3 were accepted and processed within 120 min of ischemia time using a clean bench. Tumor samples were cut into fragments of 3 × 3 × 3 mm 3 and split: Two pieces were snap frozen and stored in liquid nitrogen for molecular analysis. The remaining fragments were frozen viable (FCS, 10% DMSO) and stored at -80 °C for xenografting (Fig. 1).

PDX generation
After melting of the frozen samples, DMSO was removed and the tumor fragments were embedded in Matrigel (Corning ® Matrigel ® Basement Membrane Matrix, Wiesbaden, Germany) for 10 min. Meanwhile, the NSG-mice were anesthetized using Ketamin/Xylazin (dose: 90/6 mg/ kg bw). After verification of sufficient anesthesia, both hind flanks were shaved, iodine disinfected and incised (3 mm). Tumor fragments were implanted subcutaneously and wounds were closed using simple interrupted sutures (Ethicon 6-0, Johnson & Johnson GmbH, Neuss, Germany) followed by iodine disinfection. Mice were placed under a heating lamp during recovery from anaesthesia and received analgesia (Metamizol 1250 mg/l, in drinking water) (pre-and) post-surgery to reduce pain.
After recovery, the flank tumors were measured weekly using a caliper. If no tumor growth occurred for 6 months, mice were euthanized. When flank tumors grew and reached a maximum size of 1.5 × 1.5 × 1.5 cm 3 , euthanasia and tissue collection was performed. The tumor was minced as described for the primary sample. Briefly, two fragments were used for molecular and histological analysis and the remaining fragments were viable frozen for further passaging. Tumor nomenclature was as followed: HNSCC [serial number] P [passage P0/P1] M [mouse M1/M2]. The patients' serial numbers are maintained throughout all PDX passages.

Molecular pathology
Nucleic acids and proteins were isolated from snap frozen primary tumors and PDX [mean weight: 16.6 mg]. Workflow of sample processing, implantation and PDX validation. HNSCC patients that underwent diagnostic endoscopy or surgery were included in the study. Following resection/biopsy, the sample underwent H&E diagnostics for squamous cell carcinoma. The pathologist chose a part of viable cancer for fresh frozen fragments (within 120 min post removal). Next, thawed 3 × 3 × 3 mm 3 fragments were implanted subcutaneously into NSG mice. Finally, the engraftment efficacy, histology and molecular pathology were analyzed. Created with biorender.com

Statistics
Statistical evaluation was performed using GraphPad PRISM software, version 5.02 (GraphPad, San Diego, USA). Values are reported as the mean ± SD. After proving the assumption of normality, differences between biopsies and surgery specimens were calculated using the unpaired Student's t-test. If normality failed, the non-parametric Mann-Whitney U test was applied. Multiple comparisons were done using one way ANOVA on ranks (Bonferroni's Multiple Comparison Test). Spearman nonparametric correlation was used to calculate correlations between individual parameters (twotailed P value). The criterion for significance was taken to be p < 0.05.

Patient and sample characteristics
Fifty patients were included (male:female: 41:9) with high performance status (ECOG 1.06 ± 0.88) with a median age of 64.0 years (range 46 to 82) ( Table 1 and Fig. 1). Most patients were smokers (68%, ≥ 10 py), without critical alcohol consumption (54%). The majority of patients presented with locally advanced disease (T4a/4b) and four with a recurrence after (chemo)-radiation. Tumors were localized in the oropharynx (n = 19), oral cavity (n = 17), larynx (n = 11) and hypopharynx (n = 5) ( Table 1). Neck metastases were collected from four patients (primary + related metastasis: HNSCC40/40II/40III and HNSCC45/46; metastasis only: HNSCC08 and HNSCC48). One case was a HNSCC lymph node metastasis of unknown primary (CUP) (HNSCC37). Thirteen of the tumors were p16 + (determined by p16 immunoreactivity, representative images of p16 + and p16 − cases are given in Fig. 2). To analyze whether p16 positivity is the result of HPV infection, molecular HPV testing was done (Supplementary Table 1). HPV was confirmed in 12 cases, whereas one case (HNSCC26) was finally classified as p16 + /HPV − . The mean sample size of the tumor piece was 1.2 ± 1.6 cm 3 . In some cases of endoscopic biopsies, several pieces were obtained. Biopsy samples were smaller than surgical specimens (mean size: 0.9 cm 3 vs. 1.3 cm 3 ). The ischemia time was the same for biopsies and surgery samples (< 120 min, mean: 89.5 ± 33.6 min).
Detailed information on clinical follow-up, including adjuvant treatment, is summarized in Table 1. So far, four patients died because of progressive disease (HNSCC21, HNSCC31, HNSCC48, and HNSCC51).

PDX engraftment rate of biopsies and surgical specimens
Fourty eight cryopreserved individual HNSCC tumors were implanted subcutaneously into NSG mice (Table 1). Six HNSCC cases did not undergo PDX engraftment because of limitations in tumor quantity/quality and unspecific tumor type (CUP). Additional four HNSCC cases were lost because of impaired health of laboratory animals. Finally, 44 cases were included in data acquisition: 23 biopsy samples and 21 surgery samples.
Engraftment was obtained in 16 cases, yielding an overall efficacy of 36.4%. The engraftment rate was higher for surgical specimens (52.4 vs. 21.7%) and engraftment time was shorter (6.9 ± 2.4 vs. 10.6 ± 3.8 weeks; p < 0.05; Fig. 3A). After successful engraftment, PDX had comparable growth kinetics: Surgical specimens reached the maximum size in 9.5 ± 4.8 weeks and biopsies in 12.2 ± 4.6 weeks (Fig. 3A). Some aggressive cases grew in several flanks/implantation sites, while other cases did not engraft at all (Fig. 3B). Therefore, the positive flank related engraftment rate was 59.7% (Table 2). Again, the number of individual PDX/tumor was higher for surgical specimens (77.3%, Fig. 3B). Still, positive engraftment was seen in 37.5% of biopsy samples.
Notably, four PDX were established from p16 + /HPVdriven HNSCC cases (HNSCC06, HNSCC26, HNSCC42, and HNSCC54); three of them were taken from biopsies. Hence, the p16 status did not impact engraftment efficacy. Additional clinical parameters such as age, smoking, tumor localization, TNM stage and grade did not correlate with the engraftment. The same was true for the sampling related factors ischemia time and sample size ( Table 3).
Serial passaging of the first seven established PDX models into P1 NSG mice was successful. All PDX grew after passaging and showed growth acceleration (time to tumor resection: 11.4 weeks vs. P0: 13.8 weeks) (Fig. 3C).

Histomorphology is preserved in PDX
All PDX closely resembled their primaries (Fig. 4). Tumor architecture, growth pattern, cytological features and stromal architecture were principally preserved. Besides, tumor differentiation (i.e., poor, moderate, or well) largely matched between patients and PDX models (Fig. 4). The PDX reflected intratumoral and intertumoral heterogeneity. Sometimes there were minor differences in tumor morphology (degree of keratinization) between individual mice: HNSCC13 P0 M1 showed strong keratinization, whereas in HNSCC13 P0 M2 the number of keratinizing areas was moderate (Fig. 5).
Regarding the biopsies and surgical specimens, no differences in preservation of morphology could be detected.

Tumor microenvironment of patient tumors and PDX
Immunohistochemistry and flow cytometry revealed correlations between PDX engraftment, PD-L1 expression, and macrophage infiltration (Fig. 6). Immunohistological quantification of PD-L1 using the combined positive score (PD-L1 positivity in tumor and tumorinfiltrating immune cells), showed a trend towards better engraftment for low combined positive score (CPS ≤ 10); whereas cases with high CPS were more likely to be  (Fig. 6A).
Correlation of the innate immune cell compartment with PDX engraftment showed that engraftment was more frequent in cases with high numbers of CD14 + CD163 + macrophages (Pearson r: -0.757; p < 0.05) and low numbers of CD14 + CD204 + macrophages.

Molecular pathology is preserved in PDX
The mutational profile was studied in 13 patient samples and corresponding PDX. The overall number of genomic alterations in the cancer hotspot panel was low: TP53 (61.5%) and KDR (38.4%) were the most affected genes. The direct comparison of the molecular fingerprint identified few discrepancies in most commonly affected genes (Fig. 7). In the PDX of HNSCC02, a KRAS mutation (c.114 T > C; variant allele frequency (VAF): 7.3%) and SMARCB1 mutation (VAF: 5.2%) was detected. Vice versa, the patient tumor sample harbored an EGFR mutation (c.2361G > A; VAF: 30.5%) that was lost in the PDX. In HNSCC01, a KRAS mutation (c.114 T > C, VAF: 32.7%) was exclusively seen in the PDX. It can be assumed that intratumoral heterogeneity explains these findings best.  that (pre-treatment) endoscopic biopsies contain sufficient viable tumor for PDX growth, in a clinically relevant time frame [21,30]. Therefore, a comparative analysis on engraftment efficacy of tumor biopsies and surgical specimens was conducted. While both biopsy and surgery specimens were suitable for PDX formation in vivo, the engraftment rate was substantially lower for biopsies.
Recently, PDX engraftment from biopsies has been tested for many cancer entities. The success rates were 3 to 90% using fine-and core needle laparoscopic biopsies [15,22,23,31,32]. In line with our findings, direct comparisons with surgery sampling showed mainly lower biopsy engraftment rates. For HNSCC, Lilja-Fischer et al. implanted mainly HPV + oropharyngeal cancer biopsies with a success rate of 33%, which is similar to our biopsy engraftment [24]. In conflict with these results, Kang et al. reported a superior HNSCC biopsy engraftment of 100% compared to 16% of surgery samples, yet in a very small patient cohort [25]. However, both groups did not specify whether the biopsies were diagnostic tonsillectomies, lymph node metastasis extirpations or endoscopic biopsies. Take rates approaching 100% are more typical for metastatic lymph node biopsies, which are not part of routine HNSCC diagnostics. Reviewing the present results in the scope of previous research, we expect engraftment rates of ~ 30% for biopsies and 50-70% for surgical specimens in HNSCC. However, comparisons of engraftment rates are impaired by limited standardization: Depending on the authors, the engraftment rate has been reported after one, two or three passages. More differences comprise the mouse strain (complex immunodeficient vs. athymic), the number of animals/tumor and the implantation technique (orthotopic vs. subcutaneous vs. kidney capsule vs. muscle). More standards are highly desirable to reduce laboratory animal expenditure. Mutations were classified in the following categories: non-synonymous-variant is given in the box (in case of the same mutation in the primary and PDX: light blue, difference between primary and PDX: red/blue); synonymous: orange; benign, VUS: light-orange; neutral: grey. The latter is defined as "passenger" mutation that does not play a role in HNSCC; wt -wildtype: light grey. Created with biorender.com A critical factor for engraftment efficacy is tissue ischemia and cryopreservation of tumor samples prior to engraftment that impair viability and thus PDX formation [33][34][35][36][37]. Tissue ischemia has been controlled by pathological diagnostics, processing, and viable freezing within 120 min. Immediate tissue transfer into mice is supposed to enable the highest yield of viable cells, but it also requires a complex infrastructure including timing of surgery, availability of laboratory personnel and accessibility to mice at a specific age (usually < 3 month). We therefore decided to cryopreserve fresh frozen tumor samples prior to engraftment and our engraftment efficacy was comparable to previous research using direct implantation. The standardized cryopreservation enables implantation of several tumor samples simultaneously and re-implantation of frozen backup samples if initial PDX formation fails. Freezing also facilitates reanimation of early passaged tumors at later time points. To date, all individual PDX cases were cryopreserved in P0 and seven have been replanted into NSG mice. All of them grew successfully in P1.
Additional to the higher engraftment rate, surgery samples showed an accelerated in vivo growth. After tumor formation, the growth kinetics was the same in both groups. Handling time, specimen size, p16 status, Ki-67 index, patients' clinical characteristics and implantation (3 × 3 × 3 mm 3 fragments) did not differ between the study groups. As sampling quality does not appear to be the primary reason, it can only be speculated on the underlying causes for impaired biopsy engraftment. One plausible explanation may be sampling of less proliferating tumor cells in biopsies because of intratumoral heterogeneity [32]. Likewise, the tumor microenvironment at the invasive margin may have played a role, too. Biopsies are usually taken from exophytic tumor areas to avoid bleeding and tumor cell dispersion: These areas are exposed to air, toxins and microorganisms. Also, basal proliferating areas remain untouched. In contrast, PDX of surgical specimens originate from central areas of the tumor to ensure safe pathological margin diagnostics. If the surgery samples are large and margins safe (e.g. laryngectomy) (I) viable cells from the cancer invasion front may be selected and (II) specific areas (center vs. margin) of the tumor may be punched to consider heterogeneity.
To improve engraftment rates, orthotopic implantation of HNSCC biopsy pieces might provide a good alternative. Also, metastasis and accurate mimicry of the original tumors' environment can be anticipated. Another way to increase the number of PDX models from biopsies is a preselection of the tumor type (e.g. focusing only on highly malignant/aggressive cases) or ex vivo enrichment of tumor cells via tumor-surface antigen-based sorting/separation (such as EpCAM or cancer-initiating/ stem cells), followed by co-implantation with non-malignant stromal cells. Finally, grafting of three-dimensional patient-derived tumor organoids or even circulating tumor cells is increasingly applied in preclinical research and may help to ameliorate success rates prospectively [38].
Even though surgery samples have many advantages, including non-resectable advanced cancer biopsy samples in preclinical models is crucial. Since these tumors are usually treated by radiation or chemotherapy, drug response prediction would be highly desirable. Additionally, 75% of the p16 + PDX originated from biopsies in this study. HPV-related HNSCC PDX underline the importance of biopsy xenografting in cancer entities, which are mostly treated by irradiation therapy. Improvement of biopsy engraftment could also enable repetitive PDX generation from pre-therapeutic endoscopy, surgery, post-radiation endoscopy and cancer recurrence to study cancer progression and clonal selection.
Previous studies described an association of clinical and pathological features between tumors with rapid and slow PDX growth. In pancreatic cancer, for instance, rapid growth was significantly associated with male gender and lymph node metastases [21]. Likewise, rapid PDX growth was associated with a poor outcome in melanoma and HNSCC patients [10,30]. A recent study described an association of tumor mutational burden and reduced E7/p16 INK4A levels with p16 + HNSCC PDX and organoid engraftment [27]. Hence, successful and rapid PDX establishment is suspected to be predictive for increased risk of recurrence and poor outcome. Since the biobank has been established recently, survival rates could not be calculated. However, previous data on the correlation of PDX engraftment and patient outcome can be confirmed at least partially: One PDX was established from a chemoradiation refractory patient (HNSCC48) who deceased shortly after salvage surgery. This PDX grew rapidly and growth accelerated after serial passaging.
Another important factor in HNSCC research is the tumor microenvironment, since HNSCC naturally shows high immune cell infiltration. Fuji et al., described the impact of the tumor microenvironment on engraftment efficacy of colorectal cancer specimens and concluded that tumor-infiltrating lymphocytes can inhibit engraftment by exerting suppressive effects on tumor growth [39]. The tumor microenvironment was scanned with immunohistochemistry and flow cytometry in the present work. Both methods confirmed increased immune activity in p16 + tumors as part of the immune response to the viral infection compared to p16 − tumors. Regarding the influence of the TME on PDX growth, immunohistochemistry revealed that low CPS scores were associated with high engraftment rates. Flow cytometric TME analysis identified (tumor associated) CD14 + CD163 + and CD14 + CD204 + macrophages as critical determinants for PDX propagation. Both subtypes are strongly associated with the M2-like phenotype. In breast and lung cancer, circulating CD14 + CD204 + cells are representative for an advanced tumor stage and contribute to metastasis [40,41]. In esophageal cancer, the high infiltration of CD163 + macrophages is significantly associated with chemoresistance [42]. Because of the value of tumor microenvironment analysis for PDX engraftment and biomarker identification, flow cytometry could become a method of choice for quantitative examination of the immune infiltration. Flow cytometry is timesaving and measures multiple cell types simultaneously in small tumor samples (compared with immunohistochemistry). A major constraint of flow cytometric tumor microenvironment phenotyping is the lack of spatially resolved measurements regarding leukocyte localization within the tumor and its interaction with different cell types, such as tumor cells and cancer-associated fibroblasts. A very recent study integrated histomorphological patterns of immune cell infiltration and mRNA expression data of immune genes in HNSCC. In this study, the status of tumor-infiltrating lymphocytes in the intra-epithelial and stromal compartment was identified as non-redundant biomarker in HNSCC that should be evaluated separately [43]. In the present study, immunohistochemistry was combined with flow cytometry and it suggests the latter as an important add-on tool for tumor microenvironment analysis in preclinical models. Finally, this work has several limitations. One major limitation is the partial molecular characterization using the 50 gene comprising Illumina Cancer Hotspot Panel. However, HNSCC is characterized by heterogeneous (epi-)genetic alterations and several comprehensive studies did not find reliable molecular biomarkers for PDX engraftment. As the focus was on the tumor microenvironment; applying flow cytometry and immunohistochemistry could identify macrophages differentiation and PD-L1 immune status as potential biomarkers for PDX engraftment. A second limitation relates to tumor heterogeneity: Implantation, histology and cancer sequencing were performed from one sample. Joining punches from different areas may preserve heterogeneity in future studies. Next, the PDX suffer from a selection bias in favor of advanced cancer. However, most HNSCC cancers are already advanced on diagnosis and most T1/2 lesions are sufficiently treated by transoral surgery. Therefore, advanced disease models are more valuable for preclinical research. Still, the results on implantation of specimens from small biopsies may also enhance research on PDX from early-stage lesions.
This study has implications for HNSCC PDX-based research: We have established and characterized new p16 + and p16 − in vitro and in vivo models from primary, recurrent and metastasized HNSCC. The new PDX represent morphology and molecular alterations of the patients' tumors. Using endoscopic biopsies, this new PDX models comprise advanced disease patients that were not eligible for surgery. In this context, the correlation between PDX growth and clinical outcome could lead to the identification of clinical biomarkers.
Even though the engraftment rate is lower, PDX from biopsies are essential to include aggressive recurrent and metastasized carcinomas -that cannot be treated by surgery-in preclinical research. Since PDX engraftment from biopsies is feasible, routine implantation of endoscopic biopsies should be implemented in HNSCC research projects. As a result, biopsy PDX might play a crucial role in personalized therapy for HNSCC patients in advanced tumor stage.

Conclusion
This study describes the successful establishment of patient-derived xenograft models from head and neck cancers obtained from endoscopic biopsies and surgery resection specimens. Additionally to the higher engraftment efficacy of the latter compared to the former, intratumoral M2-like macrophages as well as a low PD-L1 expression on tumor and immune cells were identified as independent positive predictors for engraftment. Vice versa, the p16 status had no impact on engraftment efficacy. Finally, PDX models from both sources were successfully transferred and expanded thus broadening the resources for preclinical drug response analyses.