A multicenter clinical study: personalized medication for advanced gastrointestinal carcinomas with the guidance of patient-derived tumor xenograft (PDTX)

Establish patient-derived tumor xenograft (PDTX) from advanced GICs and assess the clinical value and applicability of PDTX for the treatment of advanced gastrointestinal cancers. Patients with advanced GICs were enrolled in a registered multi-center clinical study (ChiCTR-OOC-17012731). The performance of PDTX was evaluated by analyzing factors that affect the engraftment rate, comparing the histological consistency between primary tumors and tumorgrafts, examining the concordance between the drug effectiveness in PDTXs and clinical responses, and identifying genetic variants and other factors associated with prognosis. Thirty-three patients were enrolled in the study with the engraftment rate of 75.8% (25/33). The success of engraftment was independent of age, cancer types, pathological stages of tumors, and particularly sampling methods. Tumorgrafts retained the same histopathological characteristics as primary tumors. Forty-nine regimens involving 28 drugs were tested in seventeen tumorgrafts. The median time for drug testing was 134.5 days. Follow-up information was obtained about 10 regimens from 9 patients. The concordance of drug effectiveness between PDTXs and clinical responses was 100%. The tumor mutation burden (TMB) was correlated with the effectiveness of single drug regimens, while the outgrowth time of tumorgrafts was associated with the effectiveness of combined regimens. The engraftment rate in advanced GICs was higher than that of other cancers and meets the acceptable standard for applying personalized therapeutic strategies. Tumorgrafts from PDTX kept attributes of the primary tumor. Predictions from PDTX modeling closely agreed with clinical drug responses. PDTX may already be clinically applicable for personalized medication in advanced GICs.


3
Background Gastrointestinal cancers (GICs), common malignant conditions that involve the gastrointestinal tract and organs, are responsible for a significant portion of malignant tumors. Globally, there were 4.7 million new GICs cases and 3.4 million related deaths in 2018, accounting for 26.1% of the incidence and 35.2% of mortality for malignant tumors (Ferlay et al. 2019;Bray et al. 2018). In China, the prevalence is even worse, with 1.84 million new cases and 1.44 million deaths, and the corresponding incidence and mortality rates for malignant tumors are 43.1% and 50.4%, respectively (Bray et al. 2018).
In the past decade, the advancement of targeted therapy and immunotherapy has revolutionized many oncology fields, including for GICs (Huynh et al. 2020). However, currently patients that can gain clinical benefit from targeted therapy or immunotherapy only account for a small portion in advanced GICs and standard chemotherapy is still the cornerstone for patient care. For example, although the objective response rate (ORR) of anti-HER2 target therapy was 30% and 38% in two Phase II trials (Hainsworth et al. 2018;Sartore-Bianchi et al. 2016), HER2 overexpression accounts for only 2-6% of patients with colorectal cancer (Seo et al. 2014). Patients with deficient mismatch repair (dMMR) were found to have a better response to the immunotherapy in several cancers (Baretti and Le 2018), having a 40% overall response rate (Le et al. 2015), but the prevalence of dMMR in Stage III-IV GICs is only around 10% (Lorenzi et al. 2020). So far, no targeted therapy or immunotherapy in the GICs performs as well as does those for non-small cell lung cancer and melanoma, and chemotherapy is still the mainstay for the management of advanced GICs. The choice of chemotherapy regimen is mainly based on practice guidelines and the experience of clinicians. However, regardless of therapeutic methods, chemotherapy alone or combined with other therapies, a low response rate is always an obstacle. This is especially true after patients have undergone multiple lines of therapies, usually with a response rate of less than 10% (Izumchenko et al. 2016), they can hardly get therapeutical benefits from further chemotherapy, which considerably weakens the value of chemotherapy in advanced GICs. One main reason for the failure of standard chemotherapy regimens for GICs is the intertumoral heterogeneity and individualized biological and molecular characteristics of these cancers; Therefore, personalized medication is required.
To improve the effectiveness of therapeutic drugs, the stratification of patients is necessary. Many approaches have been developed for this purpose, including genetic testing, immunohistochemistry tests, and in vitro or in vivo preclinical models that may involve immortalized cells, organoid technology, patient-derived tumor xenograft (PDTX), and others. Among them, PDTX is a robust platform that recapitulates the extensive heterogeneity and retains the molecular diversity of the original tumors, and effectively captures responses to therapies in patients (Malaney et al. 2014;Izumchenko et al. 2017;Gao et al. 2015). The PDTX model has been shown to be the most effective, safe, and reliable clinical test platform. It has been used extensively in preclinical drug evaluation, biomarker identification, and drug screening for personalized treatment (Izumchenko et al. 2017;Hidalgo et al. 2014;Krepler et al. 2017;Yang et al. 2019;Okada et al. 2018;Tentler et al. 2012).
To assess the clinical value of PDTX in advanced GICs therapies, a prospective multi-centered exploratory study was conducted. Patients with various advanced GICs were recruited, and preclinical PDTX models for real-time personalized medication were tried. Comprehensive factors, including engraftment rate, factors that affect the success rate of engraftment, congruence between tumorgrafts and original tumors at the histopathological level, genomic variants, and the accuracy of drug prediction, were examined.

Patient inclusion
This study was carried out as a registered multi-center clinical collaborative study (ChiCTR-OOC-17012731) from January 2016 to January 2018. The primary inclusion criteria were: (1) patients have advanced GIC confirmed by histopathology and need anti-cancer treatment; (2) fresh tumor tissue could be obtained by surgical resection or biopsy; (3) at least one measurable lesion based on the Response Evaluation Criteria in Solid Tumors, version 1.1 (RECIST 1.1) standard; (4) an Eastern Cooperative Oncology Group (ECOG) performance status score of 0-2; (5) the expected survival time must be ≥ 3 months; (6) in case of postoperative subjects, that the previous anti-tumor drug treatment (chemotherapy and/ or targeted therapy) failed, the failure time from the study was ≥ 4 weeks, and the adverse reactions were essentially resolved. The main exclusion criteria were: (1) participating in clinical studies of other drugs simultaneously; (2) patients with liver, kidney, and bone marrow hematopoietic dysfunctions not meeting the requirements of chemotherapy; (3) patients unwilling to follow the therapies recommended by the researcher based on testing results after previous treatment failures. The main exfoliation criteria: (1) researchers were unable to carry out pharmacodynamic tests; (2) any serious violation of this study protocol or non-compliance with recommended therapeutic regimens; (3) treatment plans altered due to significant 1 3 changes in patient condition. All protocols followed the guide line of local institutional ethics regulations and were approved by the local institutional review board. Informed consent was obtained from all patients enrolled.

Animal care
All tumor transplant recipients were 6-8 weeks old, specific-pathogen-free (SPF) immunocompromised NCG mice (NOD-Prk dcem26Cd52 Il2rgem26Cd22 , Nanjing Biomedical Research Institute) weighing 20 ± 2 g. The research was approved by the Institutional Animal Care and Use Committee (IACUC) of Nanjing Medical University. All mice were housed in SPF animal facilities at room temperature of 25 ± 2 °C and humidity of 40-70% for at least 3 days before the xenografting and in the same conditions afterwards. The sex of the selected mouse was consistent with that of the corresponding patient.

Tissue preparation
Fresh tumor tissues collected via surgery or biopsy were preserved within 15 min of sampling in tissue preservation solution (Hyclone, Marlborough, USA) supplemented with 100 units/ml penicillin and 100 µg/ml streptomycin and were transported to the laboratory under refrigeration. Necrotic components, fats, and blood were removed before the transplant. Xenografting was performed within 24 h after sampling.

PDTX model establishment
The PDTX model was established similarly to previously described protocols (Izumchenko et al. 2017). Briefly, 2 mm × 2 mm × 2 mm fresh tumor tissues were engrafted into NCG mice (NOD-Prk dcem26Cd52 Il2rgem26Cd22 , Nanjing Biomedical Research Institute) subcutaneously. Tumor volume was measured twice weekly and was calculated according to the following formula: tumor volume = [length × width 2 ]/2. The xenograft (P0) was aseptically resected when the volume reached 800-1000 mm 3 , and extended generations P1-P5 were expanded in a similar way. When the average tumor volume reached 100-150 mm 3 , the mice were randomly assigned (5 mice per group) to control or treatment groups after removing substandard mice (mean ± SEM/3) (SEM, standard error of the mean). Drugs were administered for 3-4 weeks based on clinical regimens.

Histopathological comparison between primary tumor and PDTX
After propagation (volume > 60 mm 2 ), fragments of the first passage of tumorgrafts (P1) were fixed, embedded, sectioned, and stained with hematoxylin and eosin (HE) using a standard procedure. The sections of tumorgrafts were examined by an independent pathologist, and their histopathological classifications were compared with those of their primary tumor counterparts. The concordance between the two results was recorded and analyzed.

Tumor growth rates calculation
Specific growth rate (SGR) was used to quantify tumor growth rate for all tumors based on measuring tumor volume changes. The primary tumor tissue was implanted into mice at the starting time (t1) and in a specific size (V1), and terminated at the time (t2) when the tumor sizes (V2) were measured. Tumor dimensions were measured twice weekly and recorded. SGR was calculated according to the exponential growth equation as follows:

Co-clinical drug screen
In each PDTX model, at least 4 types of drugs or drug combinations (determined by physician) were tested to determine the most effective regimen. Drugs were administrated either by gavage (for oral drugs) or intraperitoneal (IP) injection (for infusion or bolus). Doses were converted according the weights of mice. The therapeutical cycles were the same as those in clinical practice and were evenly distributed in 21 days. The tumor growth inhibition rate (TGI) was used to reflect drug efficacy and was calculated as follows: where Vc is the average volume of tumorgrafts in the control group and Vt is the average volume of tumorgrafts in the treatment group. The cutoff value of TGI was set to 60%, where TGI > 60% was classified as an effective regimen, and lower values, ineffective. Effective regimens were chosen for clinical treatment. If more than one effective regimen was identified, the regimen for treatment was determined by the physician. Clinical responses were classified based on RECIST 1.1. Complete response (CR), partial response (PR), and stable disease (SD) were categorized as positive and progressive disease (PD) as negative. When the testing result of the PDTX was effective, and the clinical response was positive, the result was determined as consistent, and otherwise, inconsistent.

Next-generation sequencing (NGS) analysis
Genomic DNA extraction and sequencing library construction were performed using standard protocols as described (Robinson et al. 2017). Whole-exome sequencing was performed on a NovaSeq 6000 System (Illumina, San Diego, TGI = (Vc − Vt)∕Vc × 100%, 1 3 USA), and the primary base call files were converted into FASTQ sequence files using the bcl2fastq converter tool bcl2fastq-1.8.4 in the CASAVA 1.8 pipeline. The publicly available software FastQC (https:// www. bioin forma tics. babra ham. ac. uk/ proje cts/ fastqc/) was used to assess sequencing quality. For each lane, per-base quality scores across the length of the reads were examined. Lanes were deemed passing if the per-base quality score box plot indicated that > 75% of the reads had > Q20 for bases 1-80. In addition to the raw sequence quality, the alignment quality was also assessed using the Picard package (https:// broad insti tute. github. io/ picard/). The FASTQ sequence files generated were then processed through an in-house pipeline constructed for whole-exome sequence analyses of paired cancer genomes. The sequencing reads were aligned to the reference genome build hg19, GRCh37 using BWA-mem v.0.7.5a (Li and Durbin 2009), and converted into BAM files using SAMtools (version 0.1.18) (Etherington et al. 2015). Duplicates were marked for filtering, and INDELs were realigned using GATK v.3.4.46 IndelRealigner (Do Valle et al. 2016). For somatic SNV and indel variant calling, GATK BQSR was applied to recalibrate base qualities. SNV and indel somatic variants were called using Strelka v.1.0.14 (Saunders et al. 2012). Copy-number aberration was quantified and reported for each gene as the segmented normalized log2-transformed exon coverage ratios between each tumor sample and normal samples.

Statistical analysis
Student's t-test was applied for the differences in tumor volumes between treated and control animals during drug screening. Associations between patient responses and PDTX outcomes were analyzed using Fisher's exact test. All statistical analyses were two-sided, and P < 0.05 was considered significant.

Engraftment rate and influential factors
The schematic diagram of PDTX establishment and other analytical experiments is illustrated in Fig. 1a. Among 33 primary tumorgrafts, 25 (75.8%) were propagated successfully. Of these, drug testing was accomplished in 17, 9 corresponding patients contributed follow-up results, and 23 of the 25 tumorgrafts were subjected to whole exome sequencing (WES) analysis. We analyzed clinical factors that may have affected the success of engraftment, and we discovered that the engraftment rate was independent of age, gender, tumor location, sampling methods, and pathological stage of tumors ( Table 2).

Histopathology of tumorgrafts
We also analyzed histology of tumorgrafts (P1) by comparing to matched primary tumors (Pa). All paired sections displayed significant morphological similarity, and classifications by two independent pathologists also demonstrated a 100% consistency between Pa and P1. Five pairs of typical images from various types of tumors are shown (Fig. 1b).

Drug efficacy test and comparison to the clinical response
Drug efficacy was tested after tumorgrafts were propagated (Fig. 1a). Among all enrolled cases, 51.5% of them (17 cases) had full drug tests, involving 49 single drug or combined regimens with 28 types of chemotherapeutic regents engaged. The testing period (from sampling to the end of the efficacy test) ranged from 67 to 202 days with a median time of 134.5 days. Follow-up information was obtained from 9 cases involving 10 regimens. Among these, eight regimens in 7 cases (two regimens were applied sequentially in one case, DS-dg-9) displayed an effective result in PDTX models and a corresponding positive clinical response; two PR cases and six SD cases. In the other two cases where patient responses were negative (PD), the PDTX result was correspondingly ineffective (Table 3). The consistency between patients' response and PDTX testing is 100%. Three patients (DS-dg-6, DS-dt-6, DS-dg-9) who continued the guidance of PDTX gained SD with progression-free survival (PFS) of 7, 5 and 6 months.

Outgrowth time and prognosis
The first-round propagation of tumorgrafts (P0, Fig. 1a), was assessed for outgrowth time, the time to reach 60 mm 3 from the engraftment of the primary tumor; outgrowth time usually reflects the potency of tumor's malignancy (Janakiraman et al. 2020;John et al. 2011). The overall outgrowth time was 7-76 days. In the combined therapy group, the average outgrowth time of responsive tumorgrafts (38 ± 14 days) was shorter than that of non-responsive ones (52 ± 10 days), but the difference was not statistically significant (p = 0.1297). No differences in average outgrowth times were discovered between responsive (36 ± 19 days) and nonresponsive (36 ± 14 days) tumorgrafts in the monotherapy  (Fig. 2c). Generally, the positive response rate in the combined chemotherapy group was about double that of the monotherapy group (64.6% vs. 30.2%, Fig. 2d).

Typical cases
Here we illustrate the performance of PDTX in guiding clinical medications with two typical cases. Case 1 A 63-year-old male was diagnosed with colon adenocarcinoma at stage IV with liver metastasis. PDTX was established using a specimen from a metastatic lesion via biopsy, and four regimens involving seven drugs were screened. Two regimens, including the one (cetuximab, oxaliplatin, 5-fluorouracil and calcium folinate) concurrently used for patient treatment, showed strong TGI at day 9 ( Fig. 3a-d). Consequently, the concurrent regimen was extended for 6 cycles. Liver lesions were reduced after 3 cycles of treatment (Fig. 3e, f), carbohydrate antigen 19-9 (CA19-9) decreased from 11,019.0 to 3918.0 U/ml, and carcinoembryonic antigen (CEA) decreased from 11366 to 4146 µg/L after 5 cycles of treatment (Fig. 3g, h). The clinical evaluation of patient's response was SD, consistent with the PDTX prediction.
Case 2 A 54-year-old female was diagnosed with moderately differentiated pancreatic adenocarcinoma at stage IV with liver metastasis. A specimen from the primary lesion via surgical resection was used for PDTX, and four regimens involving seven drugs were screened. The combination of gemcitabine with TS-1 (tegafur, gimeracil, and oteracil potassium capsules) showed a TGI of 80.5% (Fig. 3i-l) and was chosen for medication. The therapy continued for 3 cycles, and the metastatic lesion reduced significantly after 2 cycles of treatment. CA19-9 decreased from 1945 to 63.1 U/mL and CEA decreased from 80 to 9.6 μg/L. The clinical evaluation was PR (Fig. 3m-p).

Discussion
The incidence of GICs continues to increase worldwide. Many patients are already at advanced stages when diagnosed, and therefore, the prognosis for those patients is generally not favorable. Patients' responses to chemotherapy diverge considerably due to the heterogeneity of the disease and individual differences. Standardized chemotherapy regimens for all patients based on unified guidelines may cause first-line-drug-insensitive patients to miss their best therapeutic window, especially those patients with advanced disease.
PDTX is an ideal model for testing drug susceptibility that has been studied in a broad spectrum of solid tumors, including common cancers such as colorectal cancer, lung cancer, breast cancer, and rare cancers like adenoid cystic carcinoma and cholangiocarcinoma. The engraftment rate of PDTX differs depending on the tumor types, with a general success rate around 50%. Among all GICs, the engraftment rate of colorectal cancer is usually higher than others at about 64-89% (Zhu et al. 2015). In this study, we tried to use a standard method to establish the PDTX model with specimens from 33 patients who were diagnosed with advanced GICs. The overall engraftment rate was 75.8%, which exceeded the basic requirement of a 60-70% engraftment rate for personalized medication (Hidalgo et al. 2014). Statistical analysis showed that the success rate was independent of age, gender, sampling method, and stage, indicating that the standard PDTX technique might be universally applicable in advanced GICs. Importantly, the small amount of specimen sampled by endoscopic biopsy, puncture and aspiration showed no difference in engraftment rate compared to the large amount of specimen obtained by surgical resection. This has significant implications for advanced GICs because many such patients have lost the opportunity for surgery.
Many factors could affect the engraftment rate, including conditions for sampling and transportation, the mouse strain used for PDTX, tumor types, the amount of specimen, and the technology used for modeling. There were 8 cases where PDTX failed to establish in this study. Among these 8 cases, 4 were male and 4 were female, 4 were surgical specimens and 4 were puncture specimens. All patients were at stage IV and showed no differences in the factors that could affect engraftment. Three unsuccessful cases were from pancreatic cancer, accounting for 30% of the pancreatic cancer cases in this study. The rest of the cases were from esophageal cancer, colon cancer, liver cancer, rectal cancer, and gallbladder cancer, one case for each type. Preliminary analysis suggested the tumor type was not the cause of unsuccessful engraftment. A flaw in engraftment technique could be responsible, but this would need to be examined in more detail and validated with more samples. A PDTX model can faithfully preserve the histological characteristics, molecular diversity, heterogeneity and microenvironment of the primary tumor making it the closest approximation of a patient's physiology among all tumor models. Zhu et al. (2015) successfully established 63 PDTX models in NOD/SCID mice using gastroscopic biopsy tissues from 185 patients with gastric cancer. The results showed that the histopathological characteristics of the PDTX model were highly consistent with those of the primary tumor. The consistency between PDTX and the primary tumor of Lauren classification was 88.9% (56/63), the cell differentiation degree was 90.5% (57/63), and the HER2 mutation was 95.2% (60/63). The mice used in this study were the NCG (NOD-Prk dcem26cd52 il2regm26cd22) strain, which is deficient in T cells, B cells and NK cells. It is the most immune-compromised commercialized strain and therefore is the most suitable strain for human cell or tissue transplantation. All 25 PDTX models were confirmed to be tumor tissue by pathology, and they were consistent with the primary tumor at the pathological and the morphological levels.
The PDTX model can accurately reflect the patient response, and therefore predicts the patients' susceptibility to certain chemo drugs, which can be used to optimize clinical regimens for personalized medication. Izumchenko (Izumchenko et al. 2017) and others established 578 PDTXs from multiple types of tumors, of which 237 cases were subjected whole genome sequencing to confirm the genetic consistency between the primary tumor and tumorgraft. They also compared the clinical efficacy of 129 regimens in 92 patients with the predictions of PDTX and the result showed a consistency of 87% (112/129). In this study, the clinical follow-up results of 10 regimens in 9 patients were obtained. The clinical responses and PDTX predictions of 8 regimens were all positive, and the other 2 regimens were negative; the clinical responses and PDTX predictions results were completely consistent. However, only 4 of the 10 regimens in PDTX were exactly the same as the clinical regimen. Among the other six regimens, five (a11016, 130111, 110160 and 0101006) of them had additional drugs in the clinical regimens compared to the PDTX model, and the remaining one had one drug removed in the clinical regimen. Although the final clinical regimens are often adjusted depending on the physician's decisions, and the predictions from PDTX were consistent with the patient response, such alterations still should be avoided when designing future studies.
The PDTX model can screen multiple drug regimens simultaneously and thereby give an important reference for formulating individualized medication plans. Especially when recurrence and metastasis occur, the results of previously tested drugs provide an important reference for guiding subsequent-line therapy. Hidalgo et al. (2011) carried out a clinical study in 2011 on PDTX for guiding the chemotherapy for advanced refractory cancer. PDTX models were successfully established in 12 cases, including 7 cases from GICs. 232 regimens involving 63 drugs were tested, and 17 treatment plans suggested by PDTX modeling were engaged on the patients in 11 cases with 15 plans achieved long-term PR. Subsequently, multiple studies (Bürtin et al. 2020;Bresnahan et al. 2020;Corso et al. 2019;Kondo 2020;Gitto et al. 2020) reported PDTX-guided personalized medication that covered both chemotherapy and targeted therapy, illustrating the advantage of PDTX for therapy optimization. In this study, 3 patients (130111, 110150 and 110160) also continued the medication following the guidance of PDTX and obtained SD.
Clinically, the effectiveness rate of a single drug is usually low and it is difficult to find an effective single-drug regimen. In this study, we found a statistically significant negative correlation between the effectiveness of monotherapy and tumor mutation burden (TMB). A biomarker developed in recent years, TMB has become one of the most important biomarkers in immuno-therapy. For example, in June 2020, the FDA approved a TMB-associated pan-cancer indication for pembrolizumab. However, whether TMB can be used as a biomarker for other kinds of drug treatment is still largely unexamined and could be a useful subject for future studies.
In this study, we also found that some unconventional drugs were highly efficacious in the PDTX model. For example, mitomycin C showed effective inhibition of tumorgrafts in 4 of the 5 PDTX models, although the drug was not administered to patients. As an antineoplastic drug, mitomycin C was originally used for upper gastrointestinal cancer (such as esophageal cancer), anal cancer, breast cancer, and superficial bladder cancer, but its indications are still expanding. For example, in April 2020, the FDA approved the application of mitomycin C in low-grade urothelial carcinoma. For patients with advanced cancer, the regimens recommended in the guidelines are usually of limited effectiveness. When there is no standard treatment available, doctors usually carry out individualized treatment according to literature or personal experience. However, by testing multiple drugs in parallel, PDTX provides a much more efficient and evidence-based basis for selecting drugs.
Invasiveness and the malignant proliferation of tumors are often associated with poor prognosis. The growth rate of a tumorgraft can reflect the proliferative ability of the primary tumor. The outgrowth rate of tumorgrafts was also examined and was correlated to drug efficacy. The results showed that the faster the growth rate of the transplanted tumor, the lower the possibility of single drug inhibition, and the greater the need to combine drug regimens to inhibit growth. However, the results were not statistically significant, and more research is needed to determine a pattern.
PDTX modeling is a significant development in translational medicine, because it allows for the personalization of cancer treatment. However, there are still many challenges in clinical practice. First, PDTX modeling requires fresh tumor tissue, and the tissues need to be properly transported and transplanted in time. If the patient cannot have surgery or a biopsy, or no facilities are available for transportation and transplantation of specimens, the model cannot be established. Secondly, and more importantly, in the context of advanced stages of GICs, the therapeutic window is narrow, whereas an elongated testing cycle of 3-6 months is obligatory for PDTX modeling. Patients with rapidly progressing disease will have lost their opportunity for drug therapy before PDTX testing results are available. Therefore, the clinical utility of PDTX is restricted and a tailored approach to PDTX may be needed under these circumstances. One potential modification is using one mouse per model per treatment scheme. In this design, averaged results from multiple mice per treatment are replaced by the result from just one mouse, and the long period of tumorgraft propagation can be substantially reduced since there is no requirement for a massive amount of tumor tissue. This approach has already been used in industrial-level drug screenings and exploratory trials (Gao et al. 2015;Einarsdottir et al. 2018;Ny et al. 2020). However, the applicability of this design to the clinical setting still needs to be fully validated. Another adaptation could be using first-line therapy following guidelines while second-line or multipleline drugs are being assessed simultaneously using PDTX. Using this scheme, the timing issue would be attenuated at the cost of potentially reduced effectiveness rate of first-line therapy. However, a precise and effective regime of nonstandardized therapies for the subsequent treatment can be accurately determined. A third challenge is that the PDTX model is unapplicable for evaluating the efficacy of immunosuppressants and immunomodulators since the mice used are severely immune-compromised. Human immunity is the focus of clinical oncology in recent years with many efforts being made to develop immunotherapy in various cancers, including GICs. Although most GICs do not respond well to immunotherapy (Abdul-Latif et al. 2020), progress has been Fig. 3 Typical cases of PDTX in advanced GICs. a-h Advanced CRC; i-p advanced PDAC. a and i Schematic time frame of modeling in each case. b and j Drug regimens tested in each case. c, d and k, l Growth inhibition of tumorgrafts for each drug regimen. e, f and m, n Biomarker changes before and after treatment. g, h and o, p CT scans of tumors before and after treatment. Red circles indicate the locations of tumors. CEA carcinoembryonic antigen, CA19-9 carbohydrate antigen 19-9, TGI tumor growth inhibition rate made in MSI-high colorectal cancer and gastric cancer with the FDA-approved immune checkpoint inhibitors pembrolizumab and nivolumab in metastatic or refractory settings. Despite that progress, a single immunotherapeutic agent as monotherapy is unlikely to be successful for GIC treatment, and many trials for combined regimens are ongoing now. Unfortunately, PDTX is inherently unsuitable for these new therapies because the severe immuno-deficiency in the host and the incompatibility between human and murine immune system. One solution could be humanized mice for PDTX, a complex, emerging topic (Byrne et al. 2017) which is beyond the scope of this study.
One major limitation of this study lies in the small sample size, and a large-sample-sized study to validate conclusions in this research will be our next endeavor. It would also be useful to further examine the consistency between primary tumors and tumorgrafts in PDTX models, making additional comparison such as immunohistochemistry and/or molecular biomarkers for specific types of tumors. New biomarkers and/or targets could also be explored. Communication with clinicians should be further addressed to ensure consistency between laboratory and clinical protocols in future studies.

Conclusion
To summarize the above points, PDTX modeling is applicable for the personalized medication in advanced GICs. PDTX faithfully preserves the heterogeneity of primary tumors, accurately predicts drug response, rapidly stratifies patients, and effectively guides the optimization of treatment regimens. PDTX modeling can reduce ineffective chemotherapy for GICs and allows real-time personalized medication.