A Chemosensitivity Study of Colorectal Cancer Using Xenografts of Patient-Derived Tumor-Initiating Cells

This article is featured in Highlights of This Issue, p. 2077

Colorectal cancer is one of the commonest types of cancer worldwide, with approximately 50,000 deaths estimated in 2017 (1). Although cytotoxic and molecularly targeted drugs have improved the patient survival, it is still challenging to treat metastatic colorectal cancer. For example, the overall 5-year survival rate of stage IV patients is only 13% in the United States (2). Because response to a regimen varies widely depending on each patient, personalized optimization of chemotherapy is important not only to give efficacious regimens but also to avoid ineffective ones.

Recently, the patient-derived tumor xenograft (PDX) mice have been adopted as a preclinical model because they preserve the genetic profiles and heterogeneity in the primary tumors (3, 4), although some caveats remain (5). Notably, drug-dosing tests with PDX mice have successfully predicted chemosensitivities of the corresponding patient tumors (6, 7). However, only limited numbers of successful cases have been reported on PDX-guided personalized therapies apparently due to the long time and high cost required (3, 8).

Lately, patient-derived tumor-initiating cells (TIC, aka cancer stem cells) have been cultured in vitro as organoids/spheroids (9–12). These cells retain genetic and morphologic characteristics of their original tumors when propagated in vitro (9, 10). However, these cells behave differently in vitro compared with those in vivo surrounded by tumor microenvironments that can affect chemosensitivities (13). Thus, mouse xenograft models appear to be more reliable in predicting the patient chemosensitivities. Although patient-derived TIC spheroids can form xenograft tumors in immunocompromised mice (10, 12, 14), it remains to be evaluated whether such xenografts can serve as personalized chemosensitivity tests.

In the current study, we have established a panel of patient-derived spheroid xenografts (PDSX) by engrafting colorectal cancer TIC spheroids subcutaneously into nude mice. We then compared them with the conventional PDXs derived from the same tumor cohort regarding the efficiency of tumor formation, timeline for drug sensitivity tests, and their drug responses. Finally, we compared chemosensitivities of PDSXs with those of the corresponding patients along their clinical courses in a retrospective manner.

Human colorectal cancer samples were obtained from patients who underwent resection operations at Kyoto University Hospital (Kyoto, Japan; ref.9). The study protocol was approved by the Ethics Committee of Kyoto University (Kyoto, Japan), and written informed consents were obtained from the patients.

Four- to 6-week-old female nude (BALB/c-nu) and NOD/SCID mice were purchased from CLEA Japan or Charles River Laboratories. All animal experiments were conducted according to the protocol approved by the Animal Care and Use Committee of Kyoto University [title of the protocol: “Chemosensitivity studies of gastrointestinal cancers using patient-derived tumor xenografts.” Approval No. 14546, 15091, 16047, 16654, 17086, and 18080 (2014–2018)].

Collected tumor samples were transferred from the operation room to the laboratory in the ice-cold washing medium (9), and washed with the medium and PBS twice each. Necrotic tissues were removed, and each tumor sample was cut into small cubic fragments (50–100 mm³). One or two of them were used to establish primary cancer spheroids and others for PDXs.

To establish the founder PDXs (P₀ generation), the tumor fragments were implanted directly to each flank of nude mice and/or NOD/SCID mice under isoflurane anesthesia. The tumor size was measured and the tumor volume was estimated using the following formula: tumor volume (mm³) = [length (mm) × width² (mm²)]/2 (15, 16).

Graft implantation was judged as successful when estimated tumor volume reached approximately 1,000 mm³, and considered as failed if no visible tumor mass was recognized for 6 months. To passage the tumors in vivo, we excised and cut them into smaller cubes (50–100 mm³).

Patient-derived colorectal cancer TIC spheroids were cultured as reported previously (9). In short, fragments of excised tumor were minced and digested by collagenase type I (Thermo Fisher Scientific). Then, epithelial cells were collected and suspended in Matrigel (Corning Inc.) and cultured in the cancer medium (9) with or without 50 ng/mL EGF (PeproTech) and 100 ng/mL basic FGF (PeproTech).

To prepare the PDSX mice, TIC spheroids were cultured in 3 wells of a 12-well cell culture plate for injection into a mouse. Spheroids in confluent culture (1–9 × 10⁵ cells) were rinsed with PBS twice and transferred to a 1.5-mL tube together with Matrigel. Spheroid suspensions were adjusted to the total volume of 100 μL each with PBS and injected into nude mice subcutaneously. The tumor size was measured weekly.

Paraffin-embedded tissues of PDXs and PDSXs were prepared according to the standard procedures. Primary colorectal cancer sections were obtained from Department of Diagnostic Pathology, Kyoto University Hospital. Sections were stained with H&E or immunostained for MUC2 (Dako, M7313) or CDX2 (BioGenex, AM392-5M).

Oxaliplatin (Wako, 152-02693), irinotecan (Wako, 091-06651), 5-fluorouracil (Wako, 064-01403), and levofolinate calcium (Wako, 035-22871) were prepared in 5% glucose solution. Cetuximab (Erbitux, Merck) was diluted with PBS.

Groups of PDX (P₂–P₄) and PDSX mice were prepared and subjected to drug-dosing tests when estimated tumor volume reached 300 to 500 mm³. We excluded the following kinds of tumors from the tests as outliers: too small (<100 mm³), too large (>900 mm³), abscess-like, naturally shrinking, or those deeply implanted and difficult to be measured.

Mice in each set were distributed into the control and treatment groups (n = 3–6/group). The day of the dosing start was defined as day 1, with the tumor volume and the body weight of each mouse measured twice a week for 3 weeks (days 1–22).

The treatment protocols were designed to reflect clinical regimens, and drug doses for mice were calculated according to the following formula: mouse dose (mg/kg) = human dose (mg/kg) × 37 (hK_m)/3 (mK_m), where K_m indicates human (h) or mouse (m) body surface coefficient (17).

For FOLFOX-like treatment, mice were injected intraperitoneally with oxaliplatin (12 mg/kg) and levofolinate calcium (30 mg/kg) first, followed by 5-fluorouracil (55 mg/kg), weekly for 3 weeks. For irinotecan treatment, mice were injected (intraperitoneally) at 40 mg/kg weekly for 3 weeks. For FOLFIRI-like treatment, mice were injected (intraperitoneally) with irinotecan (40 mg/kg) and levofolinate calcium (30 mg/kg) first, followed by 5-fluorouracil (55 mg/kg), weekly for 3 weeks. For cetuximab treatment, mice were injected (intraperitoneally) at 250 μg per mouse, twice a week for 3 weeks (18). These were less than the maximal tolerated doses (MTD), and approximately 80% of the clinically relevant doses.

The relative tumor volume was obtained by calibration to the initial tumor volume on day 1. To evaluate effects of drug dosing, the T/C (treated/control) and TGI (tumor growth inhibition) values were calculated according to the following formula (7, 16):

TGI (%) = [1 – Δ relative tumor volume (treated)/Δ relative tumor volume (control)] × 100, or = [1 – Δ relative tumor volume (treated)] × 100 for Δ relative tumor volume (treated) when < 0 (i.e., tumor regression), Δ Relative tumor volume = (relative tumor volume on day 22) – (relative tumor volume on day 1).

The χ² analyses, unpaired t test, Tukey multiple comparisons, Sidak multiple comparisons, Spearman correlation analyses, and Pearson correlation analyses were performed using GraphPad Prism ver.6 (GraphPad software. Inc.)

We collected 92 fresh colorectal cancer samples from 89 patients who underwent resection operations at Kyoto University Hospital. Background characteristics of the patients and tumors are summarized in Supplementary Table S1. We successfully established the founder generation (P₀) PDXs in 56 of 92 tumors (61%) and the TIC spheroids in 68 of 92 cases (74%). For 44 tumors, we generated both PDXs and TIC spheroids.

As reported previously (19), we found a correlation between advanced tumor stages and high success rates for P₀ PDXs. Namely, stage II, III, and IV tumors had higher PDX take rates (57%, 71%, and 67%, respectively) than stage I (29%). Notably, the success rates in culturing TIC spheroids were higher than the PDX take rates through all stages (Supplementary Table S1).

Histopathologically, 86 of 92 colorectal tumors (93%) analyzed here were classified as well- or moderately differentiated adenocarcinomas (i.e., low grade in a two-tiered classification; ref. 20). The remaining six included two poorly differentiated and four mucinous adenocarcinomas. This distribution of histopathologic subtypes is similar to that of common clinical cases (21). Interestingly, the latter two subtypes that are generally known as more malignant than low-grade adenocarcinomas showed rather high success rates for both PDXs and PDSXs (Supplementary Table S1).

Although 10 of 92 (11%) cases had been treated by neoadjuvant chemotherapies in this patient cohort, the success rates were not affected substantially in establishing either PDXs or spheroids by the chemotherapies (Supplementary Table S1).

In short, we procured collections of both in vivo PDXs and in vitro TIC spheroids from a cohort of patients with colorectal cancer.

To determine the efficiency in preparing xenografts from our TIC spheroids, we injected nude mice with 21 spheroid lines subcutaneously and successfully developed tumors of 18 lines (at the rate of 86%; Supplementary Table S2). We coined these xenografts as PDSXs to distinguish them from the conventional PDXs.

For drug-dosing tests with PDXs, it required serial in vivo transplantations (more than two passages) to prepare a large number of PDX mice. In contrast, the whole set of PDSX mice could be established simultaneously once the spheroid cultures were expanded in vitro (5–20 passages). Overall, the cumulative success rate for PDXs was estimated to be approximately 43%, whereas that for PDSXs was 64% (Supplementary Fig. S1). These results indicate that the PDSX method is more efficient than the conventional PDX.

To investigate whether the PDSXs retain the primary tumor morphology, we compared histopathology of PDSXs with that of the primary tumors and PDXs. Notably, their epithelial characteristics such as gland formation and cell differentiation were essentially identical to those of the primary tumors and PDXs for the respective cases (Supplementary Fig. S2), although mice formed less tumor stroma than humans as reported (15). Likewise, primary tumor expression of markers such as CDX2 and MUC2 was also reproduced in PDSXs (Supplementary Fig. S2).

In PDXs, not all engrafted tumor fragments expanded, which was likely caused by tumor tissue heterogeneity regarding the TIC contents. Notably, PDSXs showed much smaller variances in tumor volume among individual xenografts than PDXs, despite that they derived from the same primary tumors (Supplementary Fig. S3). The mean coefficients of variation (CV; standard deviation divided by the mean value) for PDSXs and P₁ PDXs were 0.29 and 0.53, respectively (P < 0.05). The large CVs in PDXs (especially, in early passages) were problematic in the preparation of PDX mice for drug sensitivity tests. With PDSXs, on the other hand, the low CVs increased the statistical power in drug-dosing tests even with relatively small number of host mice (e.g., n = 3–4/group; ref. 22). Accordingly, we concluded here that the PDSX model was more efficient and therefore suitable for drug efficacy evaluation than the conventional PDX.

To compare the reliability in drug-dosing tests between PDX and PDSX models, we prepared test mice transplanted with samples from 4 patients with colorectal cancer (Fig. 1A). We had to exclude 23 of 94 (24%) PDXs as outliers because of the large individual variances in growth rates described above. On the other hand, we eliminated only 9 of 85 (11%) PDSXs according to the same criteria as applied to PDXs.

As the first example of drug dosing, we used groups of PDXs and PDSXs derived from a colon cancer case (HC13T) that turned out to carry heterozygous BRAF^V600E mutation. As anticipated, the results of drug sensitivity tests showed that cetuximab (an anti-EGFR Ab) treatment was ineffective in both PDX and PDSX mice (T/C = 120% and 114% with PDXs and PDSXs, respectively; Fig. 1B and C; Supplementary Table S3). On the other hand, a FOLFOX-like regimen significantly decreased the tumor growth (T/C = 48% and 57% with PDXs and PDSXs, respectively; Fig. 1B and C; Supplementary Table S3). Notably, PDXs showed much larger variances in tumor growths than PDSXs; even the outliers had been eliminated before dosing (Fig. 1B and C; Supplementary Table S4). Accordingly, the dosing data with PDSXs led to stronger statistical differences than those with PDXs (Fig. 1B and C). Essentially the same results were obtained with tumors from 3 additional patients (Supplementary Fig. S4; Supplementary Tables S3 and S4). Regarding therapy responses, the mean values for PDSXs correlated well with those for PDXs (in T/C and TGI, Spearman correlation; R = 0.67 and 0.90, respectively; Supplementary Table S3). These results indicate that the difference between PDSXs and PDXs did not affect drug sensitivities substantially. Importantly, however, all datasets of PDSXs had only half as large CVs as those of PDXs with a significant statistical difference (the mean CVs were 19% and 31% for PDSXs and PDXs, respectively, P < 0.01; Supplementary Table S4).

To ensure the reproducibility of the two methods, we performed drug sensitivity tests of HC13T xenografts with independently prepared additional set of both PDXs and PDSXs. Notably, the first- and second-set relative tumor volumes of PDSXs were almost identical between the corresponding test groups (e.g., control 1 and 2 in Fig. 2C and D). However, those of PDXs showed much wider variances (e.g., control 1 and 2 in Fig. 2A and B). In addition, there was a statistically significant difference between the mean tumor volumes of PDX cetuximab 1 and 2 data (P < 0.01; Fig. 2A and B). We obtained similar results with tumors from 2 more patients (HC5T and HC17T), underscoring higher reproducibility of the PDSX model than PDX (Supplementary Fig. S5).

Thus, the PDSX method enabled us to decrease individual variances in xenograft growth rates and to evaluate colorectal cancer chemosensitivity even with such small numbers of mice as three to four in each dosage group.

To exploit PDSXs for personalized treatments, we compared drug responses of PDSXs with those of the patients in a retrospective manner. We prepared panels of PDSXs from 7 patients who had been treated with several chemotherapeutic regimens for liver or peritoneal metastasis after resection of their primary colorectal cancers. We performed a total of nine drug-dosing tests that matched the regimens given to the corresponding patients (Supplementary Table S5). The clinical courses for 3 individual patients are summarized below (Figs. 3–5), and four more are presented in Supplementary Figs. S6–S9:

Patient 1 (HC1T; Fig. 3): This patient was treated by S-1 (a 5-FU prodrug) + oxaliplatin (i.e., a FOLFOX-like regimen) for the metastatic lesions, which led to a stable disease (SD; Fig. 3A, B-b and B-c). In PDSX mice, a FOLFOX-like dosing caused a moderate response with a significant difference from the control group (T/C = 50%, TGI = 82%; P < 0.01; Fig. 3C; Supplementary Table S5). Thereafter, the patient was treated with irinotecan + cetuximab, which resulted in another SD (Fig. 3A, B-d and B-e), although the treatment was interrupted for a month due to an accidental knee bone fracture (Fig. 3A). When we performed the same irinotecan + cetuximab regimen on the PDSX mice, it caused a moderate suppression of tumor growth, indicating that the primary tumor was responsive to this therapy (T/C = 37%, TGI = 93%; P < 0.01; Fig. 3D; Supplementary Table S5). Accordingly, both sets of these clinical responses in this patient to the regimens were reproduced well in the PDSX dosing experiments.

Patient 2 (HC6T; Fig. 4): This patient was treated with FOLFOX + panitumumab (an anti-EGFR antibody) regimen that caused dramatic decreases in the tumor volume [partial response (PR)] and serum carcinoembryonic antigen (CEA) level (Fig. 4A, B-b and B-c). Following the postoperative FOLFOX + panitumumab treatment, he was switched to FOLFIRI + bevacizumab regimen. Although the treatment led to a condition evaluated as an SD at first, the CEA level was increasing steadily and the size of metastatic lesions increased during 6 months of treatment (Fig. 4A, B-c and B-d). In the PDSX model, transplanted tumors were sensitive to FOLFOX + cetuximab (T/C = 42%, TGI = 122%; P < 0.01; Fig. 4C and D; Supplementary Table S5). On the other hand, they did not respond to the FOLFIRI regimen well (T/C = 76%, TGI = 50%; P < 0.05; Fig. 4C and D; Supplementary Table S5). Accordingly, the marginal effects of the FOLFIRI-based treatment for the patient appeared to be an innate characteristic of the primary cancer, which was reflected in the PDSX data.

Patient 3 (HC50T; Fig. 5): On the other hand, there was only one set of PDSX drug-dosing results that did not appear to match with the clinical evaluation (RECIST) of a chemotherapy regimen. In this case, the patient was treated by (FOLFIRI-like) S-1 + irinotecan + bevacitumab regimen. The efficacy of S-1 + irinotecan regimen was clinically assessed as a progressive disease (PD) with a 21% increase in the metastatic tumor volume determined by MRI during the treatment (Fig. 5A,. 5B-a and B-b). Despite the clinical assessment of PD above, the PDSXs treated with FOLFIRI were reduced in size significantly (T/C = 55%; TGI = 107%; P < 0.05; Fig. 5C and D; Supplementary Table S5)

Upon close reexamination of the clinical course (Fig. 5A), the above assessment was likely affected by the timing of MRI examinations. Consistent with the sharply decreased levels of CEA and CA19-9 after the overshoots that were caused by resistance to the preceding chemotherapy with XELOX (Fig. 5A), the results of ¹⁸F-FDG–PET images taken one month after the respective MRI photos showed markedly reduced liver glucose metabolism in the metastatic lesions (compare Fig. 5B-c and B-d). Even if the S1 + irinotecan + bevacitumab regimen was efficacious, it was likely to have taken some time before the metastatic tumors shrunk to the volume smaller than that in the previous MRI. In summary, the apparent discrepancy between one of the clinical assessments and PDSX results was of the transitional nature during the clinical course, and therefore did not discredit the PDSX evaluation method.

Accordingly, all nine sets of the drug-dosing results with PDSXs well reflected their corresponding clinical outcome (Supplementary Table S5). Namely, the therapeutic regimens that were efficacious in the patients (PR or SD by RECIST) were also effective in treating PDSXs (Supplementary Table S5). In addition, there were strong correlations between the results of PDSX dosing tests (T/C) and patient outcome, as well as those and DOR (duration of clinical response; Supplementary Fig. S10A and S10B). Correlations were also confirmed between TGI and the patient outcome as well as DOR (Supplementary Fig. S10C and S10D).

Taken together, these results support the robust reliability of PDSXs in “paraclinical” evaluation of chemosensitivity (see Discussion).

One of the most practical limitations of the PDX method for personalized medicine is the long and unpredictable time it takes to prepare enough numbers of PDX mice for drug-dosing tests. For example, it took 5 months or longer to set up a group of P₂ generation PDXs in our study. Thereafter, it took approximately 35 days for P₂ PDXs to expand to the size of 300 to 500 mm³ (i.e., appropriate for drug dosing; Supplementary Fig. S11, top). In contrast, it took us only 2 to 3 months to prepare a group of approximately 20 PDSX mice (Supplementary Fig. S11, bottom). These results indicate that the PDSX method is more advantageous than PDX not only in the accuracy of drug sensitivity testing but also in saving time for preparation of the tumor-bearing mice. Accordingly, it should be easier to feed back the dosing data to the bedside before deterioration of the patient conditions.

The current results demonstrate three advantages of PDSXs over conventional PDXs in clinical application to chemosensitivity tests. First, the PDSX method showed a higher efficiency (i.e., take rate) than the conventional PDX method in the dosing-test mouse preparation, which should provide the personalized clinical services to a wider cohort of patients with colorectal cancer. The take rates for conventional PDXs were reported as 60% to 80% at the founder generation (3, 4), although the tumor take was not always defined precisely (23). The cumulative rate after three passages of colorectal cancer was reported to be 43% in nude mice (24), which is similar to the currentresults (43%; from primary tumors to P₁ PDXs). Only a few pieces of data were reported on the tumor take rate regarding spheroids/organoids injected into immunodeficient mice (14, 25). In this study, we established TIC spheroid lines for 74% (68/92) of primary tumors and generated PDSX mice for 86% (18/21) of the spheroid lines. One of the reasons for this high success rate of PDSX formation was likely because we expanded TICs in culture and thereafter injected them in larger numbers than those contained in the PDX transplants. Regarding intratumor heterogeneities, our PDXs and spheroids that derived from separate subregions of the same tumor were almost identical as others reported (5, 26, 27).

Second, the PDSX method allowed us to reduce variances in the tumor growth rate. As noted above, TIC spheroids consisted of only proliferating cancer epithelial cells. In case of PDXs, on the other hand, the engrafted tumor fragments upon passages were likely to be heterogeneous regarding the number of live TICs and their microenvironment (e.g., the extent of differentiation and coexisting stromal cells, respectively). In fact, we had significantly fewer mice excluded in the PDSX groups than in PDX upon preparations for drug-dosing tests, indicating that PDSX method is less wasteful in xenograft formation.

Third, and most importantly, PDSXs gave chemosensitivity data statistically more significant than PDXs in drug-dosing tests. Accordingly, it is expected that the clinical responses can be predicted more reliably with PDSX mice than with PDX. To date, many methods have been proposed that predict patient responses to cancer chemotherapeutics (28). These include mutational analyses, gene expression signature analyses, as well as in vitro sensitivity tests with cancer organoids similar to spheroids (14, 29). For example, colorectal cancer that retains intact Ras signaling with wild-type RAS genes responds to EGFR inhibitors at a high probability (25, 30). However, there is always a sizable fraction in the patient population that does not respond to the indicated regimen(s). Although it is beneficial to each patient if a regimen is efficacious, the patient will lose precious time and QOL if not. Thus, more reliable prediction methods are awaited. To this end, direct tumor grafts to immunodeficient mice (i.e., PDXs) have been proposed as a straightforward method of drug sensitivity evaluation personalized to the respective patients (3, 4).

Our current results provide a rationale for PDSXs as a more improved method than PDXs, overcoming the drawbacks of the latter. Namely, the results of PDSX drug-dosing tests demonstrated a strong correlation with the clinical responses (Figs. 3–5; Supplementary Figs. S6–S9), and paralleled with those of PDXs (Supplementary Table S3). Therefore, these results suggest a strong predictive power in chemosensitivity when applied to personalized prospective studies. To this end, our PDSX method also helps expedite preparation of test mice for drug-dosing tests. Namely, it took only 2 to 3 months to setup groups of PDSXs for testing several regimens. Notably, however, some types of cancer may be excluded that take very rapid downhill courses such as pancreatic cancer. It can be more practical to test the sensitivity of spheroids in vitro for such types. Along this line, in vitro cultures of spheroids/organoids have been proposed to be used in drug screening as well as in personalized chemosensitivity tests (9, 31). Such in vitro sensitivity tests may provide quicker, although limited, answers for a class of chemotherapeutics that directly target cancer cell proliferation. In other words, there are occasions in which xenografts can provide more practical prediction of the clinical courses as exemplified by irinotecan sensitivity in Fig. 4. Although irinotecan resistance was reported to correlate with the expression level of DNA topoisomerase I (32), our expression analysis did not provide enough data that allowed personalized predictions.

Taken together, the PDSX method should allow us to design personalized chemotherapy regimens for patients with advanced colorectal cancer in a prospective manner. We would like to propose introduction of personalized PDSX-based chemosensitivity tests as “personalized paraclinical prediction” (PPP) trials. For testing already established regimens, they can be designated as PPP phase 3.5X trials (X for xenograft) because the drugs had completed phase III clinical trials. When the tests can be performed using in vitro culture of spheroids/organoids, they may be PPP phase 3.5V (V for in vitro). If candidate drugs for repurposing are used, they can be PPP 1.5X or 1.5V trials, and so on. We encourage further discussion among those who participate in cancer chemotherapy development.

Notably, PDSXs have some limitations common to PDXs because both lack key immune responses and have different stromal microenvironment from that of human hosts. Technical improvements to humanize the mouse immune system and mimic the patient tumor microenvironment in mice appear to be in progress, although multiple difficulties still remain (4, 33, 34). Despite such limitations, PDX has been one of the best preclinical models owing to their predictive power for a variety of chemotherapeutics (3, 4).

In conclusion, we have demonstrated that PDSXs are reliable “paraclinical” models for personalized colorectal cancer chemotherapies. Our methods of TIC spheroids and PDSXs are straightforward, reliable, and well formulated. They also meet the timeline in most colorectal cancer clinical courses. In addition, PDSXs may be applied to chemotherapies of not only colorectal cancer, but also other types of cancer when the primary tumors are available. It is worth noting that applications of the PDSX method can be extended to evaluations of off-label uses of therapeutics and compassionate uses of yet-unapproved drugs.

No potential conflicts of interest were disclosed.

Conception and design: H. Maekawa, H. Miyoshi, T. Yamaura, M.M. Taketo

Development of methodology: H. Maekawa, H. Miyoshi

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): H. Maekawa, H. Miyoshi, T. Yamaura, Y. Itatani

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): H. Maekawa, H. Miyoshi, M.M. Taketo

Writing, review, and/or revision of the manuscript: H. Maekawa, H. Miyoshi, K. Kawada, Y. Sakai, M.M. Taketo

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): Y. Itatani, K. Kawada, Y. Sakai

Study supervision: H. Maekawa, H. Miyoshi, M.M. Taketo

We thank the Medical Research Support Center, Graduate School of Medicine, Kyoto University for the technical support. We also thank the members of the Division of Gastrointestinal Surgery, Department of Surgery for help in collecting surgical specimens. We are grateful to T. Sakurai and H. Haga of the Department of Clinical Pathology at Kyoto University Hospital for their pathologic diagnosis. We are also grateful to S. Matsumomo, T. Horimatsu, M. Kanai, and M. Muto of the Department of Clinical Oncology, Kyoto University Hospital for clinical discussions on chemotherapy. This work was supported by the Program for Creating STart-ups from Advanced Research and Technology (START, ST261001TT) from Japan Science and Technology Agency (JST); Practical Research for Innovative Cancer Control (JP18 ck0106195) from Japan Agency for Medical Research and Development (AMED), and Kyoto University Venture Incubation from Kyoto University Office of Society-Academia Collaboration for Innovation (to M.M. Taketo).

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.