signature=7fd77218b220ffb1b5f9e4fec1a300e6,Tumor grafts derived from patients with head and neck squ...

In this study we transplanted 30 human HNSCC tumors directly into mice and serially transplanted those that engrafted. The overall engraftment rate was 17% with poorly differentiated and node-positive tumors more likely to engraft. There were significant expression differences between the human tumors that subsequently grew in mice and those that did not. The tumors maintained the histologic characteristics of the parent tumor, but were more homogeneous in successive generations. Human stromal components were lost upon engraftment. Gene expression between the PDXs and parent tumors were distinct, but stable in successive generations. The PDX expression pattern for gene probes that were different between parent tumors and human cell lines was quite similar to that of the human parent tumors.

Several PDX models have been developed and characterized, with several common themes emerging. In nearly all cases the histology of the original parent tumor was similar to that of the matched PDX [10–13]. This similarity persisted up to at least passage 12 [15]. In our study, tumors from passage 10 were histologically similar to parent tumors, with a loss of stroma over subsequent generations. Genetic abnormalities in the parent tumor such as specific mutations and gene copy number generally have been maintained in xenografts. In a study of colon cancer xenografts, there was concordance between gene copy number and mutation (NRAS, KRAS, BRAF, PIK3CA) in first- and second-generation xenografts and in the parent tumor [15]. Lung cancer models demonstrated similar karyotype and gene copy number as parent tumors by array comparative genomic hybridization (aCGH) [10].

The engraftment rate has varied greatly in different models, and rates ranging from 11% to 90% have been reported [11, 29]. The highest engraftment rate was attributed to rigorous selection of viable cancer tissue and the relatively greater vascularity of the renal site of engraftment than of the usual subcutaneous site [16], although a subcutaneous colon cancer model had a similar engraftment rate of 87% [15]. Engraftment correlated with less tumor differentiation [11, 30], advanced stage [13], and poor donor prognosis [12, 29, 31, 32] in most, but not all, studies [11, 33]. In the best characterized HNSCC PDX model before ours, the engraftment rate was 11% overall. In that model, poorly differentiated tumors, larger tumors, recurrent tumors, and tumors from lymph node metastases were more likely to form xenografts [30, 33]. Additional PDX models of HNSCC have been developed but have not been well characterized at the molecular level and most also demonstrated low engraftment rates [34–40]. In our study, tumors that were poorly differentiated and node positive were significantly more likely to engraft.

Factors other than differentiation and metastasis to lymph nodes may also influence engraftment rates. Tumors derived from metastatic sites have generally been more likely to engraft than those from primary sites [15, 30], but not always [29]. Surprisingly, cold ischemic time did not influence engraftment [15]. Characteristics that predicted engraftment in NSCLC included squamous histology, poor differentiation, larger tumor size, and KRAS mutation [12]. Likewise, in pancreatic cancer the tumors that grew in mice were more likely to have SMAD4 mutations and an independently developed metastatic gene signature [31, 41]. Efferth et al. examined 20 proteins in NSCLCs that were transplanted into mice and found that increased expression of JUN, FOS, cyclin D, and CDK4 predicted successful engraftment [42]. Our results show that tumors that engrafted had a gene expression profile that was distinct from the profile of those that did not engraft.

Few studies have rigorously examined stroma in PDX models. One of the most comprehensive characterizations of the PDX model showed that, after the first passage, tumor cells were enriched in the xenografts compared to the parent tumors, but the ratio of stroma:tumor was constant in subsequent passages. Human stroma was lost in the early passages [15]. In a breast cancer model, human stroma was replaced by mouse stroma after engraftment and tumor cells were enriched in the xenografts [29]. When NSCLC tumors were engrafted into NOD-scid IL2γnull (NSG) mice, tumor-derived human leukocytes were maintained for up to 9 weeks in the first generation [43]. Similarly, we found that in our HNSCC PDX models, the human stroma was replaced by mouse stroma that was progressively less abundant in subsequent passages.

In the subcutaneous PDX models, metastases have not been described. However, nodal and distant (lung and liver) metastases have developed from orthotopic injection, in patterns consistent with clinical observations in colorectal cancer patients [15]. In a breast cancer model, both local nodal and distant metastases were found at rates that varied from 38% to 100% [29].

Several studies have compared gene expression in PDXs with that in the original parent tumors in lung, breast, and colon cancers. These studies show mixed results which are similar to ours; we found that the expression patterns of the PDX and parent tumors were distinct in 2 cases, but one parent tumor and PDX pair did cluster very tightly together. Unsupervised hierarchical clustering of 17 human NSCLC tumors and their matched heterotransplants revealed that 9 of the 17 animal tumors clustered with the primary tumors from which they were derived; 5 of the 8 primary tumors that did not cluster with their matched heterotransplants contained less than 10% tumor tissue [18]. In a breast cancer model, using expression of a set of 1291 intrinsic genes and unsupervised clustering, all xenografts and parent tumor pairs clustered together [29]. Unsupervised clustering of about 40 patient tumors with matched early- and late-stage xenografts separated parent tumors from xenografts, but there were no major differences between early- and late-stage xenografts. Ontogeny analysis of genes that were differentially expressed between tumors and xenografts revealed genes encoding for extracellular matrix components and immune modulators consistent with the loss of human stroma in the xenografts [15]. Gene expression analysis in breast cancer xenografts demonstrated that the majority of probes were not differentially expressed between the parent tumor and its xenograft. In this study, gene ontology analysis demonstrated that the genes that were differentially expressed corresponded to the stromal compartment [17]. Likewise, an independent group identified stromal and extracellular matrix genes as differentially expressed in xenografts and parent tumors [44].

Gene expression was analyzed in 3 small cell lung cancer (SCLC) biopsy specimens, primary xenografts derived from the tumors, cell lines derived from the xenografts, and those same lines reimplanted into mice. When compared to normal lung, the tumors, xenografts, and cell lines had an expression pattern that was specific for SCLC. However, the gene expression patterns changed when the cancer cells were grown on plastic, and this expression pattern did not fully recover when the lines were subsequently re-injected in vivo (i.e., secondary xenografts). The expression patterns of the primary xenografts more closely resembled those of the parental tumors than those of the cell lines or the secondary xenografts [45]. Likewise, in our model, the genes that were expressed differently in cell lines grown in plastic vs. parental tumors were maintained in PDX tumors, suggesting that the PDX model more closely reflects the biology of the parental tumor than do cell lines.

The ultimate goal of developing a PDX model is to recapitulate tumor biology for the development of effective cancer therapies. When xenografts are treated with therapeutic agents used in patients with that tumor, similar response rates are usually observed, suggesting that the PDX model is superior to cell culture models in vitro or in vivo for drug development. When mice bearing colorectal cancer xenografts were treated with cetuximab, the response rates mirrored those seen in clinical trials of cetuximab, and tumors bearing KRAS mutations were resistant to cetuximab, as are most KRAS-mutant colon cancers in human patients [14]. Similar results were demonstrated in a separate group of xenografts derived from colorectal cancer in which irinotecan and cetuximab as single agents had response rates in the population of xenografts that were similar to those found in human populations [15]. When mice with NSCLC xenografts were treated with chemotherapy regimens that are commonly administered to NSCLC patients, response rates were similar to those in patients [16]. Fifteen colorectal cancer xenografts responded to cytotoxic chemotherapy agents that are active in colon cancer in human patients [17].

Data linking drug sensitivity in specific patients to their paired xenografts are sparse. In a pancreatic cancer model, there was a positive correlation between responses of the xenografts to gemcitabine and time to progression for the corresponding 7 patients who were treated with gemcitabine [31]. The most comprehensive and only prospective published study established xenografts from 14 patients with advanced metastatic disease and screened the tumors for sensitivity to 63 drugs. Two of the patients’ xenografts were not sensitive to any drugs tested, but the other 12 patients underwent therapy based on the xenograft data. In 9 of the 12 patients, the cancer responded as predicted by this screening, while the two cancers whose xenografts were resistant did not respond to standard therapy [46]. Although these numbers are too small for rigorous statistical analysis, the correlation is striking and clearly indicates that the PDX model may be useful to predict drug sensitivity prospectively. We successfully utilized the HOSC1 model to determine the in vivo efficacy of JAK and Src inhibition, demonstrating that our model may be useful for developing novel cancer therapeutics [26].

The biggest limitations of our study were the low engraftment rate and lack of adequate tissue for analysis of all parental tumors. Many resected HNSCC tumors are small, and the residual tissue after standard pathologic analysis is completed is not adequate for research studies. The use of NSG rather than the Nude mouse may improve the engraftment rate and allow for the retention of human fibroblasts and lymphocytes [47–49]. Nonetheless, our study is the most comprehensive analysis and development of the PDX model for HNSCC to date.