Summary

Colorectal cancer's complexity necessitates a deeper understanding of its molecular subtypes for personalized treatment. While invasive procedures like biopsies are currently used for profiling, noninvasive PET imaging shows promise. Recent advancements in PET tracers offer superior predictive capabilities compared to traditional methods. By examining tumor genetics, microenvironment, and metastatic evolution, a study sheds light on PET imaging's potential for patient stratification. Ultimately, integrating PET imaging with molecular profiling could revolutionize colorectal cancer diagnosis and treatment.

Materials and methods

GEMM Mouse models were derived from nine alleles as described previ[1]ously: VillinCreER, Apcfl/þ (23), KrasG12D/þ (24), Rosa26N1icd/þ (25), Mlh1fl/fl (26), BrafV600E/þ (27), Tgfbr1/Alk5 fl/fl (28), Tgfbr2fl/fl (29), and Trp53fl/fl (30). We utilized six GEMM: Apcfl/þ (A), Apcfl/þ KrasG12D/þ (AK), BrafV600E/þ Mlh1fl/fl Tgfbr2fl/fl (BMT), Apcfl/þ KrasG12D/þ Trp53fl/fl Tgfbr1/Alk5 fl/fl (AKPT), Apcfl/þ KrasG12D/þ Trp53fl/fl Rosa26N1icd/þ (AKPN), and KrasG12D/þ Trp53fl/fl Rosa26N1icd/þ (KPN; refs. 26, 31, 32). Tumors were induced in male and female mice aged 6–12 weeks via VillinCreER recombination and tamoxifen injection. Clinical signs such as anemia, hunching, or weight loss indicated tumor development. RNA transcriptional analysis was conducted on GEMM, while the KPN model was imaged using [18F]FDG PET/MRI.

Six mouse-derived colon cancer organoid lines from five GEMM were cultivated in growth factor-reduced Matrigel with specific culture conditions. Tumor allografts were induced in CD-1 nude male mice, while orthograft tumors were induced in male C57BL/6 mice. Tumor growth was monitored clinically and via colonoscopy, with allografts growing for approximately 2 weeks and orthografts for about 3 to 4 weeks before PET/MR imaging. Additionally, transcriptional profiling of primary colon tumors from genetically engineered mice was conducted, followed by CMS classification and gene set enrichment analysis to characterize the tumors. The study provided a comprehensive overview of tumor modeling and molecular characterization methods, laying the groundwork for further investigations into colon cancer biology and therapeutic approaches.

Results from the nanoScan PET/MRI

PET/MRI imaging sessions were conducted on colon tumor-bearing mice using a nanoScan PET/MRI  (Mediso Medical Imaging Systems) scanner while under isoflurane anesthesia. A total of 30 allograft-bearing mice underwent four imaging sessions each, utilizing different PET imaging biomarkers such as [18F]FDG, [18F]FET, [18F]FLT, and [11C]ACE. Additionally, eight orthograft tumor-bearing mice underwent two imaging sessions each with [18F]FDG and [18F]FET. The radiotracer doses administered to the mice were carefully adjusted to ensure a uniform injection volume, with an average injected dose of 15.95 ± 2.1 MBq for [18F]FDG, 15.97 ± 1.4 MBq for [18F]FET, 13.8 ± 3.59 MBq for [18F]FLT, and 252.94 ± 69.2 MBq for [11C]ACE. PET scans were reconstructed using Tera-Tomo 3D software, and quantitative assessment was performed by manually drawing volumes-of-interest around tumors on MRI scans. Standardized uptake values (SUV) were calculated using the average SUVmean and the SUVpeak values from the five hottest VOI pixel values.

To evaluate the effectiveness of PET tracers ([18F]FDG, [18F]FET, [18F]FLT, and [11C]ACE) in distinguishing different colon organoid subcutaneous models (BMT, AKPN, AK, AKPT, KPN), we analyzed their performance using ROC curves. The BMT subcutaneous model showed the highest distinctiveness (ROCmean = 0.80), followed by AKPN, AK, AKPT, and KPN models. [18F]FDG exhibited the highest discriminatory power (ROCmean = 0.73), followed by [18F]FET, [18F]FLT, and [11C]ACE. Interestingly, different PET tracers varied in their effectiveness for different models, with [18F]FDG and [18F]FET performing best overall. Specific PET tracers showed superior separation abilities for certain model-tracer combinations, such as [18F]FDG/BMT, [18F]FET/KPN, and [18F]FLT/BMT.

Figure 2. Distinct intermodel heterogeneity in PET imaging signatures. A, In the experimental imaging protocol, five colon cancer organoid models and four PET tracers were used to determine imaging signatures. Details of all mice used in these studies are presented in Supplementary Table S1. B, Representative transverse PET images from each model and tracer. The [18F]FDG PET/MR images of the KPN subcutaneous model are reproduced again in Figs. 4B and D and 5B for comparison against other tumors at different sites and stages. C, Imaging signature heatmap showing mean tracer uptake, models with highest tracer update highlighted with black outline (representation of the data matrix analyzed with two-way ANOVA). D, Correlation matrix of each tracer uptake based on Pearson correlation coefficient. E, Heatmap illustrating correlation of PET tracer uptake with gene expression in the Molecular Signatures Database (MSigDB) hallmark dataset, sorted by hierarchal clustering

Moreover, specific PET tracers showed superior separation abilities for certain model-tracer combinations, such as [11C]ACE/AKPN with AUC values of 0.82 ± 0.09 (P = 0.02), 0.79 ± 0.09 (P = 0.05), 0.98 ± 0.02 (P < 0.001), and 0.88 ± 0.08 (P = 0.02), respectively. Previous studies suggest that distinct oncogenes, like KRAS, drive specific phenotypic features. To explore this, we grouped and compared models based on their genetics, finding that the Kras mutation had the most distinctive driver phenotype, followed by Apc, Tgfbr1, and Notch, while the loss of p53 was the least distinguishable genetic alteration. When separating the driver genes, [18F]FLT exhibited the best performance, followed by [18F]FDG, [18F]FET, and [11C]ACE. These findings demonstrate the potential of PET imaging in characterizing cancer heterogeneity and identifying specific genetic alterations.

Figure 3. PET imaging can distinguish different colon subcutaneous organoid cancer models and individual driver genes. A, The data processing workflow for comparing PET radiotracer discriminatory power and the model/gene uniqueness. B, Separation matrix and statistics of the area under the ROC curves for each tracer and model. C, Red highlighted box showing boxplot and ROC curves for [18F]FDG in the BMT (n¼ 6 subcutaneous organoid allografts) compared with other models (n ¼ 19), each point represents a mouse. Numbers inside bars show sample size, n. Data compared using unpaired t test. D, Separation matrix and statistics of the area under the ROC curves for each tracer and gene. Tgfbr1/Alk5 fl/fl and Tgfbr2 fl/fl are combined as TGFb for this analysis. E, Red highlighted box showing boxplot and ROC curves for [18F]FLT in the Kras (n ¼ 18) compared to other subcutaneous models (n ¼ 6), each point represents a mouse. Numbers inside bars show n. Data compared using unpaired ttest. Error bars in C and D represent SD. , P < 0.05;  , P < 0.01; , P < 0.001 for unpaired ttests and AUC ROC. Each analysis stands on its own; no multiple comparison testing was used. See extended datasets in Supplementary Fig. S4

Additionally, radiotracers beyond [18F]FDG, such as [18F]FET and [18F]FLT, proved beneficial in discerning between different cancer models and genetic alterations. Furthermore, we investigated whether imaging variances related to transcriptional profiles were influenced by consensus molecular subtypes (CMS). Despite dividing the models into CMS classes (CM4 and CMS2/3) and conducting ROC analysis, none of the tested radiotracers were able to differentiate based on the CMS classifiers, as illustrated in Supplementary Figures S5 and S6.

Figure 4. Imaging signatures depend on tumor context. A, The generation of mouse models. B, Transverse and coronal PET/MRI slices images showing [18F]FDG uptake in subcutaneous and orthotopic organoid models and GEMM of KrasG12D/þ Trp53fl/fl Rosa26N1icd/þ (KPN) colon cancer. KPN subcutaneous images reproduced here from Fig. 2B for comparison with KPN orthograft and GEMM. Tumors are outlined with a white dotted line. C, Standard uptake peak values (SUVpeak) PET quantification from images in B (n ¼ 4 mice/model). Data compared using ANOVA followed by Fisher least significant difference test. D, Transverse and coronal PET/MRI slices images showing [18F]FDG uptake in subcutaneous and orthotopic KPN and Apcfl/þ KrasG12D/þ Trp53fl/fl Tgfbr1 fl/fl (AKPT) organoid models of colon cancer. Tumors are outlined with a white dotted line. KPN images reproduced here from Figs. 2B and 4B and for comparison to AKPT. E, SUVpeak quantification from images in panel D (Numbers inside bars show sample size, n). Data compared using two-way ANOVA, with the results of injection site shown. Details of all mice used in these studies are presented in Supplementary Table S1. F, Representative GLUT-1 immunohistochemistry of tumors from mice shown in panel E. Black scale bars represents 100 mm. G, H-score of GLUT-1 immunohistochemistry from mice shown in panel E. Box and whisker plots show range, median and interquartile range. Error bars in panel C and E represent standard deviation. Data compared using two-way ANOVA, with the results of injection site shown,  P < 0.05,   P < 0.01,  P < 0.001.

Imaging characteristics in cancer are not solely influenced by genetic factors but also by the tumor microenvironment. To explore this, we used the same genetic model (KPN) in various contexts: subcutaneous, orthotopic, and autochthonous. Interestingly, higher-fidelity models displayed increased [18F]FDG uptake, with GEMM > orthotopic > subcutaneous, indicating an environmental influence on PET imaging outcomes. Comparing KPN and AKPT models, we found higher [18F]FDG uptake in orthografts compared to subcutaneous tumors, with variations observed between genotypes. Further validation through GLUT-1 staining suggested enhanced glucose uptake in orthografts, underscoring the significant role of the tumor microenvironment in shaping PET imaging outcomes. Additionally, the contribution of genes and environment to imaging signatures varied based on tumor genetics, highlighting an intricate interplay between genetic and environmental factors in influencing PET imaging phenotypes.

Figure 5. PET imaging phenotypic difference between primary and metastatic tumors. A, The generation of the KrasG12D/þ Trp53fl/fl Rosa26N1icd/þ (KPN) and KPN liver metastasis organoid lines and subsequent implantation. One pair of lines, generated from a matched mouse primary tumor and liver metastasis, which were then propagated and injected subcutaneously into recipient mice (n ¼ 5). B, Transverse and coronal PET/MRI slice images showing uptake of four PET tracers ([18F]FDG, [ 18F]FET, [18F]FLT, [18F]ACE) in subcutaneously implanted KPN primary and KPN liver metastasis organoids. KPN primary tumor-bearing mice are the same four PET ([18F]FDG, [18F]FET, [18F]FLT, [18F]ACE) images (primary) as displayed in Figs. 2B and 4B and D. C, Standard uptake peak values (SUVpeak) PET quantification from images in B. Sample size (n) is displayed on the bars. Error bars represent SD. Data compared using unpaired t test. Details of all mice used in these studies are presented in Supplementary Table S1. D, Representative GLUT-1 IHC and Lat-1/Slc7a5 ISH. Black scale bars represent 50 mm (, P < 0.05; , P < 0.001; see also Supplementary Fig. S8).

Conclusion

In conclusion, the research underscores the multiple influences on the imaging phenotypes of colon cancer, including genotype, model, site, and stage, suggesting a complex interplay of intrinsic and extrinsic tumor mechanisms. Their results support the use of PET as a valuable tool to provide differential molecular diagnostic insights. They achieved remarkable accuracy in distinguishing between cancer models and genetically diverse tumors, even in the presence of complex driver genes. Notably, the tumor microenvironment has become a key determinant of imaging phenotypes, sometimes surpassing genetic disparities. Radiotracers such as [18F]FLT and [18F]FET require further research to assess Kras mutation, Apc loss and tumor staging. As a non-invasive addition to conventional biopsy-based molecular diagnostics, PET-based phenotyping enables a comprehensive assessment of the entire organism, facilitating the identification of intra-patient heterogeneity and dynamic changes.

 Consequently, their study represents a significant advance toward noninvasive, multi-tracer profiling of cancer subtypes, promising to advance precision medicine strategies in the field of colon cancer treatment.

Full article on Clin Cancer Res