Patient characteristics

In this study, our analysis consists of tissues collected from 372 CRC patients. Primary tumor tissues with an equal number of patient-matched normal mucosa tissues were obtained from seven different regions of the colorectum: 63 cecum, 44 ascending, 32 transverse, 28 descending, 75 sigmoid, 38 rectosigmoid, and 92 rectum. We implemented a rigorous procedure to ensure that the normal mucosa tissues were indeed free of malignant cells and belonged to the same anatomical subsite as the matched tumor tissue (Supplementary information 1). We did not observe any statistically significant differences in age (p = 0.997), sex (p = 0.073), and chemotherapy status (p = 0.237) between cases from different anatomic subsites (Table S4). We also compared cancer stages between all subsites and found the overall ANOVA p-value to be significant (p = 0.006). However, after performing Bonferroni post hoc analysis we observed only one significant difference between sigmoid and rectum subsites (p = 0.004) concerning cancer stage (Table S4 and S5). Notably, there was no difference between the remaining subsite comparisons in terms of stages.

Metabolites commonly altered for cancers across colorectum subsites but their association with each subsite is different

Untargeted metabolome profiles were compared between all patient-matched tumors and normal mucosa from the seven subsites, which revealed 26 identified metabolites that were commonly altered in all the subsites (Table S6). We observed significantly reduced levels of hydroxyindoleacetic acid (HIAA), uric acid, linolenic acid, and dopamine sulfate, in tumors across all subsites. The association between these metabolites and CRC differed depending on the subsite location. A negative association with tumors strengthened from cecum to rectum for HIAA (OR ranges from 0.42 to 0.11), uric acid (OR ranges from 0.48 to 0.17), and linolenic acid (OR ranges from 0.58 to 0.37). The highest risk of CRC with a reduced level of HIAA was observed in the cecum. The risk was generally higher in the transverse and ascending subsites (which are part of the proximal colon). The risk was reduced in descending, rectosigmoid, and rectum tissues, but not in sigmoid subsites. The magnitude of risk was lowest in the rectosigmoid, and highest in the cecum, with a 3.6-fold difference in the OR. A reduced level of uric acid was associated with a higher risk in the cecum, ascending colon, and descending colon, as compared to other subsites; the risk was reduced in the sigmoid, rectosigmoid, and rectum. The magnitude of risk was lowest in the rectum and highest in the cecum, with a 2.88-fold difference in the OR. The reduced level of linoleic acid in tumors was associated with a higher risk in the regions from the cecum to descending and then the risk was reduced from the sigmoid to the rectum. Dopamine sulfate showed heterogeneity in the association with CRC (OR ranges from 0.54 to 0.34) in different regions of the colorectum but there was no pattern observed.

A significantly increased level of 15 metabolites was observed in tumors from each of the seven subsites when compared to their matched normal mucosa. The majority of these metabolites had an OR that differed between subsites, suggesting the heterogeneous association of these metabolites among the seven subsites of colorectum. The positive association of metabolites with CRC strengthened from cecum to rectum for kynurenine (OR ranges from 2.06 to 6.60) and reduced nicotinamide adenine dinucleotide (NADH) (OR ranges from 2.06 to 3.72). The risk was higher from the sigmoid to rectum subsites as compared to other subsites. The magnitude of risk was lowest in the cecum and highest in the rectosigmoid, with a 3.2-fold difference in the OR in the case of kynurenine. The magnitude of risk was very similar within proximal subsites and distal subsites, but the ORs were generally substantially higher in distal subsites for NADH. The magnitude of risk for hydroxyproline was lowest in the rectum and highest in the sigmoid, with a 2.6-fold difference in the OR. The ORs were very similar within proximal subsites for hydroxyproline. N-acetyl-L-cysteine was highly heterogeneous with no specific pattern across the subsites. The magnitude of risk was lowest in descending and highest in ascending subsites, with a 3-fold difference in the OR. We also performed a linear regression between CRC subsites and the OR for each metabolite. It was observed that the strength of association between CRC risk and metabolites was significantly linear (Plinear<0.05) and gradually changed from cecum to rectum for NADH, kynurenine, and HIAA.

The ORs for uridine 5’-diphosphate (UDP) and guanosine diphosphate (GDP)-glucose did not differ across the subsites which shows that these metabolites have low heterogeneity for OR along the colorectum. For all the remaining metabolites, the risk association with CRC was generally heterogeneous with no specific patterns from cecum to rectum.

Distinct metabolite changes in tumors compared to matched normal mucosa in each anatomic subsite

Subsite-specific metabolite markers associated with CRC were discerned through a comparative analysis of tumors from each of the colorectal subsites with their respective matched normal mucosa. The unique metabolite changes in each subsite are presented in Fig. 1B and Figures S3-S4. Notable metabolite changes were identified for each CRC subsite, and metabolite identification was confirmed using analytical standards. Specifically, in the cecum, two dipeptides; alanyl-leucine, valyl-methionine, and an amino acid alanine, exhibited significant upregulation. In ascending tumors, a marked increase in valerylcarnitine and a decreased level of LysoPE(P-18:0) were observed. Transverse tumors displayed a specific elevation in decanoylcarnitine. Notably, in the descending colon, two carnitine metabolites, hexanoylcarnitine and octanoylcarnitine were elevated. Epsilon-(gamma-glutamyl)lysine exhibited a decrease in tumor compared to normal tissue in the descending colon. Interestingly, in sigmoid tumors, two bile acids, glycocholic acid and glycodeoxycholic acid were downregulated. Rectosigmoid tumors showed a significant upregulation of LysoPE(18:0). In rectal tumors, increased levels of six metabolites; cysteinylglycine, cytidine, cytidine-5’-monophosphate, glucose-6-phosphate, glucosamine-6-phosphate, and mevalonate were noted, alongside decreased levels of six metabolites; ethyl-4-aminobenzoate, phenylacetylglutamine, alanyl glutamine, palmitoleic acid, valyl-serine, and diethyl succinate. Furthermore, DDA and MSE data were processed to predict structures of additional unique metabolites (designated level 3 identifications) in each subsite. Our analysis enabled the identification of additional distinct metabolites in each subsite; 8 in cecum, 3 in ascending, 2 in descending, 9 in sigmoid, 2 in rectosigmoid and 19 in rectum (Figure S4). This enhanced metabolome coverage provides a more comprehensive insight into the unique tumor biology of each subsite. Overall, findings shown in this section underscore subsite-specific metabolic alterations in CRC, revealing potential markers that contribute to a nuanced understanding of the molecular landscape of colorectal tumors.

Comparative metabolite changes in tumors compared to normal mucosa and specificity to right versus left CRC subsites

Next, we identified the metabolites that were uniquely altered in the tumors of proximal subsites (cecum and ascending) as compared to their respective normal mucosa, which were not similarly altered in distal subsites (i.e. descending, sigmoid, rectosigmoid and rectum), and vice versa. The unique metabolite changes in the proximal subsites or distal subsites are presented in Fig. 1C, S3, and S5. Formylmethionine, N-acetylmethionine and isovalerylglycine were significantly upregulated and oleic acid was significantly downregulated in the cecum and ascending tumors when compared to their matched normal mucosa, but there was no change in any of the subsites in the distal regions. Next, we identified the metabolites specifically altered in the tumors of distal subsites when compared to their matched normal mucosa. Nicotinamide adenine dinucleotide (NAD) and tryptophanamide were significantly upregulated in the tumors of all the distal subsites including descending, sigmoid, rectosigmoid, and rectum. Five metabolites, LysoPC (20:4), LysoPC (22:6), LysoPE (22:5), sn-glycero-3-phosphocholine, and pantothenic acid, were specifically upregulated and two metabolites (thymidine and bilirubin) were specifically downregulated in the tumors of the sigmoid, rectosigmoid, and rectum. Furthermore, the level 3 identification of additional metabolites revealed some additional distinct metabolites in proximal or distal subsites; three in cecum and ascending, one in descending, sigmoid, rectosigmoid and rectum, three in sigmoid, rectosigmoid and rectum (Figure S5).

Characterization of the metabolic landscape across the different subsites of colorectum

To systematically chart metabolome composition along the colorectum, we compared metabolite abundances of 223 annotated metabolites in normal mucosa and tumor tissue separately, between seven different subsites of the colorectum. Spatially resolved analysis of the colorectum revealed the heterogeneous metabolic landscape of normal mucosa and tumor tissue of refined subsites. Kruskal-Wallis test was applied to detect the metabolic differences between seven subsites, and the metabolites that were significantly different in at least one pairwise comparison was used to perform hierarchical clustering analysis (Tables S7-S8). The results revealed that metabolic heterogeneity exists among tumors and normal mucosa of the seven CRC subsites. There was a statistically significant difference in the mean abundances of 70 metabolites in at least one pairwise comparison between the normal mucosa of all the colorectum subsites (Figure S6 and Table S7). The major differentiating metabolites included 13 lipid metabolites (seven LysoPCs and six LysoPEs). Other metabolites that varied significantly among subsites were two bile acids (glycocholic acid and glycoursodeoxycholic acid), N-acetylneuraminic acid, kynurenine, succinic acid, lactic acid, and S-adenosylhomocysteine (SAH). Similarly, we identified statistically significant differences in the mean abundances of 39 metabolites in at least one pairwise comparison between the tumor tissues of all the colorectum subsites (Fig. 2A and Table S8). Normal and tumor colorectum tissue had 21 common metabolites (including four LysoPCs, four LysoPEs, and two bile acids (glycocholic acid, glycoursodeoxycholic acid)) that were significantly different between subsites; yet the distribution of these metabolites varied depending on the tissue types (tumor and normal) (Fig. 2A, S6 and Tables S7-S8). For example, LysoPC(18:1) was significantly different between cecum and descending in both the tumor and normal tissues but only in normal tissues between ascending and rectosigmoid.

Next, the subsites were grouped into three major regions for further analysis: (1) right sided CRC (RCC); cecum, ascending, transverse, and hepatic flexure, (2) left-sided CRC (LCC); splenic flexure, descending, sigmoid, rectosigmoid, and (3) rectum. The metabolites were compared between these three major regions in the normal mucosa and tumor tissues independently. In the tumor tissues, we identified 62 metabolites that were significantly different (FDR < 0.05) between LCC versus RCC, 13 metabolites significantly different between LCC versus rectum, and 54 metabolites significantly different between RCC versus rectum (Fig. 2B). Similarly, in the normal mucosa, 83 metabolites were significantly different (FDR < 0.05) between LCC versus RCC, 33 metabolites were significantly different between LCC versus rectum, and 88 metabolites were significantly different between RCC versus rectum (Figure S7). The metabolites that were significantly different in at least one pairwise comparison was used to perform hierarchical clustering analysis, and the distribution of these metabolites across three regions (RCC, LCC and rectum) is shown in Fig. 2B and S7. We discovered several intriguing metabolic differences between RCC and LCC tumors. Specifically, 11 phospholipid metabolites (five LysoPCs and six LysoPEs) and three bile acids (glycocholic acid, glycodeoxycholic acid, and glycoursodeoxycholic acid) were significantly upregulated in RCC tumors compared to LCC tumors. Comparing tumors from RCC and LCC locations, a purine metabolism pathway was markedly activated, with seven metabolites downregulated (NAD, NADH, NADP+, FAD, ADP, cyclic AMP, and inosine) and one metabolite elevated (uric acid) in RCC tumors. Additionally, we observed downregulation of six metabolites (glucose-6-phosphate, glucosamine-6-phosphate, GDP-glucose, fucose-1-phosphate, UDP, and N-acetylneuraminic acid) involved in the amino acid sugar and nucleotide sugar metabolism pathway, in RCC tumors compared to LCC tumors.

Gradual changes in metabolite abundances from the cecum to the rectum challenge the simple classification of CRC into right-sided and left-sided cancer

We analyzed the distribution of the annotated metabolites (those identified at level 1 and level 2) across all seven subsites of the colorectum in both normal and tumor tissues. The analysis showed that many of the metabolites did not change abruptly and that these markers gradually increased or decreased in abundance from the cecum to the rectum along the colorectum, challenging the conventional classification of CRC into right-sided and left-sided cancers. By linearity test, 50 metabolites were significantly linear from the cecum to rectum in normal mucosa tissues with 23 metabolites showing an increasing gradient and 27 metabolites showing a decreasing gradient. We found that 25 metabolites were significantly linear from the cecum to the rectum in tumor tissues with 10 metabolites showing an increasing gradient and 15 metabolites showing a decreasing gradient. There were 14 metabolites that showed a significant linear trend in both normal and tumor tissue (Figure S8 and Table S9). The distribution of three representative metabolites showing significantly linear trends (Plinear<0.05) from cecum to rectum in each case are shown in Fig. 2C. For instance, glycoursodeoxycholic acid displayed a decreasing concentration gradient from cecum to rectum, with similar trends followed in both colon tumor and normal tissues (Fig. 2C). Glycoursodeoxycholic acid had Plinear<0.05 for both normal mucosa and tumor tissue. Figure 2C shows that adenosine diphosphate (ADP) exhibited a gradual increase (Plinear< 0.05) from cecum to rectum in tumor tissue but remains relatively stable across subsites in normal tissues. N-acetylneuraminic acid showed exactly the opposite trend, gradually increasing (Plinear<0.05) from cecum to rectum in normal tissue but remains relatively similar across subsites in tumor tissues (Fig. 2C). These results also validated our previous study which showed significantly higher levels of glycocholic acid and glycoursodeoxycholic acid in RCC tumor as compared to LCC tumors [12]. Overall, our findings indicate that metabolite concentration gradient exists across the colorectum with these gradients contingent upon the disease status.

Fig. 2

A Hierarchical clustering analysis of metabolite abundances from colorectum tissue samples. Abundances for metabolites significantly differentiating in at least one pairwise comparison between colorectum subsites are shown across seven colorectum subsites in tumors (39 metabolites). Abundances were averaged from each subsite. Metabolites highlighted with blue asterisk (19 metabolites) were commonly altered in both normal and tumor tissues. B Hierarchical clustering analysis of metabolite abundances from colorectum tissue samples. Abundances for metabolites significantly differentiating in at least one pairwise comparison between three major regions are shown across the left side colon, right side colon, and rectum in tumors. Abundances were averaged from each subsite. Metabolites highlighted with blue asterisk (25 metabolites) were commonly altered in both normal and tumor tissues. C Distribution of metabolite abundances showing concentration gradient exist from cecum to rectum in both tumors and normal mucosa. Linear regression analysis was performed between metabolite abundances and seven subsites. Three representative metabolites showing significantly linear trend (Plinear<0.05) from cecum to rectum. The concentration gradient is similar in both normal colon and colon tumor (glycoursodeoxycholic acid), concentration gradient is different between normal colon and colon tumor (N-acetylneuraminic acid and adenosine diphosphate). Each subsite is presented in a unique color. Red-Cecum, green-Ascending, blue-Transverse, cyan- Descending, light violet-Sigmoid, yellow-Rectosigmoid and dark violet-Rectum. D Subsite specific unique metabolites markers significantly (p < 0.05) associated with CRC prognosis. Hazard ratio (HR) for each metabolite (log2-transformed abundance) was calculated by Cox proportional hazard regression analysis between log2 abundances of individual metabolites and 5-year overall survival in each subsite, adjusting for age, sex, chemotherapy and stage. Error bars represent 95% confidence intervals (CIs). The unique survival metabolites for each subsite are color coded differently. Cecum-blue, Ascending-pink, Transverse-green, Descending-yellow, Sigmoid-purple, Rectosigmoid-grey, Rectum-ice blue. GDCA - Glycodeoxycholic acid, GUDCA -Glycoursodeoxycholic acid, 5-HIAA - 5-Hydroxyindole-3-acetic acid, N-acetylneuraminic acid, PAA - Phenylacetylglycine, PAGln - Phenylacetylglutamine, NMDA - N-Methyl-D-aspartic acid, E2-3G - Estradiol-3-glucuronide, gamma-Glu-Glu - Gamma-glutamylglutamate, g-Glu-epsilon-lys - Epsilon-(gamma-Glutamyl)lysine, SAM - S-Adenosyl-methionine, NAAG - N-Acetylaspartylglutamic acid

Metabolites associated with survival were different and unique to each subsite

Cox proportional hazard regression analysis was performed to identify the association between log2 abundances of individual metabolites and 5-year overall survival in each subsite, adjusting for age, sex, chemotherapy, and stage. A two-sided p < 0.05 was considered statistically significant. Notably, we observed inter-subsite differences in survival markers and distinct metabolites were associated with survival in each subsite (Fig. 2D). We identified unique metabolite markers significantly (p < 0.05) associated with 5-year overall survival, with 13 metabolites in the cecum, 10 metabolites in the ascending, 2 metabolites in the transverse, 2 metabolites in the descending, 6 metabolites in the sigmoid, 28 metabolites in the rectosigmoid and 14 metabolites in the rectum that showed exclusive associations (Fig. 2D). Next, we grouped the subsites and identified the association between metabolites and 5-year overall survival in RCC, LCC, and the rectum regions. We identified unique metabolite markers significantly (p < 0.05) associated with 5-year overall survival in each of these regions, with 6 metabolites in RCC, 15 metabolites in LCC, and 19 metabolites in the rectum region (Figure S9).

Development of an interactive colorectal cancer metabolome platform

To facilitate further exploration of the extensive data generated in this study, we designed an interactive, publicly accessible online platform to enable researchers and healthcare professionals to probe the CRC metabolic landscape across seven subsites (https://colorectal-cancer-metabolome.com/yale-university). This user-friendly interface permits the investigation of metabolite distributions, diagnostic and survival heterogeneity, and unique metabolite markers of each subsite, with highlights for data visualization and statistical analysis. Furthermore, the interactive nature of the database allows for user-driven exploration and hypothesis generation, empowering deeper insights into CRC heterogeneous biology and illuminating future research directions in the underlying biological mechanisms driving colorectal cancer progression and response to therapy.