This study was approved by the Ethics Committee of the Sixth Affiliated Hospital of Sun Yat-sen University (L2017ZSLYEC-003). All patients underwent an informed consent process approved by the Hospital Institutional Review Board.

Our study included 5763 CRC patients diagnosed at the Sixth Affiliated Hospital of Sun Yat-sen University from January 2014 to December 2018. In this study, colorectal tumor specimens were fixed in formalin, embedded in paraffin after surgery, and confirmed histologically. The clinicopathologic features of these patients were collected from their medical records. The definition of the rectum is 12 cm or less from the anal verge.

Among the 5763 CRC patients, a sub-cohort composed of 1946 patients with available follow-up records was used to evaluate the prognosis value of PIK3CA exon 9 and 20 mutations. Another sub-cohort composed of 377 stage III patients who had received at least six courses of 5-fluorouracil based chemotherapy treatment after surgical excision was used to evaluate the associations between PIK3CA mutations and CRC recurrence.

In this study, 5763 patients’ primary intestinal tumors were collected for genetic testing. Genomic DNA extraction and Sanger sequencing were followed by the procedure of our previous study (25).

Besides, high-resolution melting (HRM) test was also applied to detect the mutation status of PIK3CA. In brief, the PCR reagent used in the HRM test was LightCycler 480 High-Resolution Melting Master Reagent Kit (Cat no: 04909631001, Roche Diagnostics Indianapolis, USA). The reaction system of the HRM test was: 50-100ng genomic DNA, 10μl 2×Master mix, 2.8μl MgCl₂, and 1μl 10µM of each primer for exon 9 test, or 0.8μl 10µM of each primer for exon 20 test, and the mixtures were added up to 15μl with H₂O. The primers of the exon 9 HRM test were: 5’-GAGACAATGAATTAAGGGAAAATG-3’ and 5’-CACTTACCTGTGACTCCATAG-3’; The primers of the exon 20 HRM test were: 5’-ACCCTAGCCTTAGATAAAACTGAGC-3’ and 5’-TCCATTTTTGTTGTCCAGCCACCAT-3’. The reaction conditions of cycling and melting for exon 9 HRM test were: 95°C for 10 min; 45 cycles of 95°C for 20s, 60°C for 20s and 72°C for 20s; 95°C; for 30s and 37°C for 30s; followed by a high-resolution melt of 74°C to 87°C with 45 acquisitions/°C 40°C for 10s. The exon 20 HRM test’s reaction condition was the same as that of exon 9 except that the annealing temperature was 62°C.

The PIK3CA exon 9 and 20 HRM tests’ detection limits were evaluated with spiked plasmid samples containing wild-type copies and mutant-type copies at various percentages. In detail, the exon 9 mutation type was c.1633G>A, and the initial concentration of mutant and wild-type plasmids were both 9.2×10⁵ copies/μl. The exon 20 mutation type was c.3140A>T, and the initial concentration of mutant and wild-type plasmids were both 4.9×10⁵ copies/μl. PIK3CA mutant plasmid was mixed with its corresponding wild type plasmid at various dilutions, at ratio of 50%, 40%, 30%, 20%, 10%, 8%, 6%, 5%. The minimum detectable dilution serves as the detection limit in this study.

Sanger sequencing was wildly considered a golden standard method for mutation detection. According to the Sanger sequencing results, sensitivities and specificities were calculated for PIK3CA exon 9 and 20 by HRM tests. Only the samples that obtained consistent results from the two methods were included for this study’s subsequent analysis.

For association analysis, the mutation information of KRAS exon 2 and BRAF codon 600 in our cohort was also obtained by the previously described detection method (26).

Chi-square test, Mann-Whitney U test, and Fisher’s exact test were applied to analyze the association of PIK3CA exon 9 and 20 mutation status with clinical characteristics, including the mutation status of KRAS exon 2 and BRAF^V600E . The analyses were initially evaluated with continuous variables, categories data analysis, and further accessed with logistic regression models to evaluate the association based on estimating the odds ratios (OR) and their 95% confidence intervals (CIs). The significance test was two-sided, and a p-value < 0.05 was considered as statistically significant. All the statistical analyses were performed with SPSS 20.0 packages (SPSS, Chicago, IL, USA).

The annual mutation rates of PIK3CA exon 9 and 20 from 2014 to 2018 were calculated separately, and the mutation rates tendency was analyzed with the joinpoint regression model (Joinpoint 4.6.0.0., Calverton, MD, USA).

Kaplan-Meier survival curves for overall survival (OS) of 1946 patients with available follow-up records were performed with GraphPad Prism 5 (Graph Pad Software Inc., San Diego, CA, USA). The Log-rank test was used to assess the significance, and the p-value less than 0.05 was considered statistically significant. Besides, the logistic regression was used to screen other clinical characteristics related to the recurrence of patients at stage III. These variables were then enrolled to create a nomogram with rms package (version 5.1-4, https://cran.r-project.org/web/packages/rms/) in R statistical software (version 3.4.3).

The study cohort involved 5733 patients ( Figure 1 ), getting the general profiles of CRC patients in China. Some basic clinical features of the cohort were summarized in Table 1 . Briefly, there were more male patients (60.9%) in this cohort. The age of the first diagnosis ranged from 17 to 96 years old. We assigned patients into four groups, younger than 45 years old, between 45 to 49 years old, between 50 to 75 years old, and older than 75 years old, with 13.8%, 9.4%, 67.2%, and 9.6% of the study cohort respectively. For all CRC patients’ tumor sites, 47.4% of tumors were located in the rectum, 30.9% of tumors were located in the left colon, and 21.6% of tumors were located in the right colon. The vast majority of patients included in this study were tubular adenocarcinoma, of which 17.6%, 74.6%, and 7.8% were well, moderately, and poorly differentiated, respectively. Among these 5733 patients, only 3153 patients’ TNM stages information was available, in which 10.9%, 37.6%, 34.9%, and 16.5% of patients were assigned to stage I, stage II, stage III, and stage IV.

The associations of PIK3CA exon 9 and 20 mutations and clinicopathologic characteristics were shown in Table 1 and Table S1 . In exon 9, mutations were more likely to occur at right colon (13.4%) than at other sides (6.8% at rectum, 9.1% at left side; p-value < 0.001), while less likely to present at poor differentiation tumors (4.8%) than at well and moderate differentiation tumors (9.5% and 8.9% respectively; p-value = 0.012). In TNM stage analysis, the exon 9 mutation frequency was higher in stage II (18.3%) than the other three stages (12.8% at stage I, 12.4% at stage III, and 15.5% at stage IV, p-value = 0.001, Table S1 ). In exon 20, mutations were more likely to occur in younger patients (7.2% in patients younger than 45 years old and 6.1% in patients between 45 to 49 years old) than those older than 50 years old (4.1% in patients between 50 to 75 years old and 3.8% at older than 75 years old, p-value < 0.001), as well as more likely to occur at the right colon (9.0%) than other sites (both 3.5% at the rectum and left colon, p-value < 0.001). In TNM stage analysis, exon 20 mutations were significantly more common at stage II (11.4%) than in other stages (5.5% at stage I, 5.9% at stage III, and 6.5% at stage IV; p-value < 0.001, Table S1 ).

Among all the 5733 patients, only ten patients mutated in both exons simultaneously, which indicated that exon 9 and exon 20 of PIK3CA are mutually exclusive in CRC ( Table 1 ). KRAS exon 2 mutations were associated with PIK3CA exon 9 and exon 20 mutations (p-value < 0.001 for both exons), while BRAF^V600E mutation did not show any association with both PIK3CA exons.

Features which were significantly associated with PIK3CA mutations in Table 1 or considered to have important effects in clinical practice were involved into logistic regression analysis, and the results were shown in Table S2 . In both exons, mutations were more likely to present at right colon (in exon 9, OR = 1.86, 95% CIs = 1.53-2.26; in exon 20, OR = 2.72, 95% CIs = 2.12-3.50; p-value < 0.001 for both exons), as well as in stage II (in exon 9, OR= 1.46, 95% CIs = 1.20-1.78; in exon 20, OR = 2.01, 95% CIs = 1.55-2.60; p-value < 0.001 for both exons). In exon 9, mutations were less likely detected in poor differentiated tumor (OR = 0.51; 95% CIs = 0.32-0.81, p-value = 0.004). Higher PIK3CA exon 20 mutation frequency was associated with patients younger than 50 years old (OR = 1.70, 95% CIs = 1.31-2.21, p-value < 0.001).

The mutation rates of PIK3CA among Chinese CRC patients from 2014 to 2018 were shown in Figure S1 and Table S3 . The mutation rate of PIK3CA exon 9 shows a slightly ascending tendency from 6.2% in 2014 to 8.9% in 2018 ( Figure S1A . APC = 7.95, p-value > 0.05). The mutation rate of PIK3CA exon 20 remained nearly stable, as 5.00% in 2014 and 4.7% in 2018, ( Figure S1B . APC = 0.30, p-value > 0.05). Combined both exons, the mutation rate of PIK3CA shows a gradually increasing tendency from 11.0% in 2014 to 13.5% in 2018 ( Figure S1C . APC = 5.08, p-value > 0.05).

PIK3CA exon 9 and 20 mutations were tested in all 5763 samples with Sanger sequencing and the HRM test ( Figure 1 ). Sanger sequencing failed to generate analyzable results from 16 samples, which can be detected by the HRM test. In the rest of 5747 samples, 14 had inconsistent results. The detection limits of PIK3CA exon 9 and exon 20 HRM tests were both 5% ( Supplementary Figures S2 and S3 ), which is more sensitive than Sanger sequencing (20%). The specificities of PIK3CA exon 9 and exon 20 HRM tests were 99.77% and 99.96%, respectively, and the sensitivities of these two detecting methods were both 100% (Chinese patent, NO. 201610524420.9).

Among the final cohort with 5733 samples, 13.4% (771/5733) patients mutated in PIK3CA with 58 types of mutations ( Table 2 ). Among these 771 mutated patients, 501 mutated at exon 9, 260 mutated at exon 20, and only 10 mutated at both exons simultaneously.

Among 58 mutation types, the 10 most frequent mutation types were: E545K (32.0%), H1047R (23.6%), E542K (17.6%), H1047L (3.9%), Q546K (3.5%), E545G (2.5%), Q546R (1.8%), E545A (1.8%), M1043I (1.7%) and H1047Y (1.0%). However, the other 48 types of mutations were only detected in 1.4% (81/5733) of CRC patients, which represented 10.5% (81/771) of PIK3CA mutated patients.

Among 58 mutation types, only 12 types of mutations could be detected by ARMS-based commercial PIK3CA mutation detection kits, which comprised 78.6% (606/771) of mutated patients.

In this study, we also annotated these somatic mutations following the instruction of ACMG guides on the interpretation of sequence variants (45). There were 4, 30, 21, and 3 mutation types in PIK3CA that were classified as Tier I, II, III, and IV variants in colorectal cancer, respectively ( Table 2 ).

In order to evaluate the prognosis value of PIK3CA mutation in CRC patients, 1946 patients with available follow-up information were collected for survival analysis (clinicopathologic characteristics were shown in Table S4 ). The follow-up started from the day of surgery and ended on August 30, 2019. The median follow-up for the 1946 cohort is 16 months (0-64 months). Their overall survival rates were analyzed with the Kaplan-Meier method. In this follow-up cohort, 447 patients carried exon 9 mutations, 238 patients carried exon 20 mutations, and the remaining 1270 patients had wild-type PIK3CA gene. During the following-up, 178 patients died, among which 138 patients died from colorectal cancer or related diseases. No significant difference was detected between patients with and without PIK3CA mutation (Log-rank test, p-value = 0.289; Figure 2A ), exon 9 mutation (Log-rank test, p-value = 0.241; Figure 2B ), and exon 20 mutation (Log-rank test, p-value = 0.772; Figure 2C ), respectively. For further analyses, patients were divided into four subgroups according to their TNM stages at the first diagnosis and conducted the survival analysis separately ( Figures 2D–F ). We found that only patients at stage IV who carried exon 20 mutations experienced significantly shorter OS than wild-type patients (median follow-up: 14 months; median survival for PIK3CA wild-typed and exon 20 mutated patients: 40 months vs. 23 months; Log-rank test, p-value = 0.002; Figure 2F ). To assess the PIK3CA exon 20 mutations’ influence on the survival of stage IV patients, the COX regression model was applied. In comparison with patients carried PIK3CA exon 20 wild-type tumors, those with PIK3CA exon 20 mutated stage IV patients showed a decrease in OS (univariate HR = 2.52, 95% CIs = 1.38-4.59; p-value = 0.003, Table 3 ). In the multivariate COX regression model, PIK3CA exon 20 mutation was associated with a significant decrease in OS of stage IV CRC patients (HR = 2.72, 95% CIs = 1.47-5.09; p-value =0.012, Table 3 ).

In this study, the impact of PIK3CA mutations on 5-fluorouracil based chemotherapy treatment was evaluated in stage III patients who had received consistent chemotherapy in our hospital (clinicopathologic characteristics were shown in Table S5 ). In total, 377 patients were included in this sub-cohort, and 76 (20.2%) were observed to have disease progression (median follow-up: 13 months; median time-to relapse: 13 months). Univariate and multivariate logistic analyses were applied to screen independent factors relative to disease recurrence. In univariate and multivariate analysis, PIK3CA exon 9, and tumor sites were related to stage III patients’ disease relapse (p-value < 0.05 for all, Table 4 ), indicating that these variants were independent predictive markers in recurrence. Besides, PIK3CA exon 20 was found to be on the cusp of conventional statistical significance (OR = 2.15, 95% CIs = 0.99-4.65, p-value = 0.053) in univariate analysis. In multivariate analysis, it showed significantly associated with recurrence in stage III patients (OR = 3.89, 95%CIs = 1.66-9.10, p-value = 0.002). The same situation also occurred in patients’ gender and tumor differentiation ( Table 4 ).

A nomogram incorporated these five variables of stage III patients’ disease recurrence was established ( Figure 3A ), and the concordance index of this nomogram was 0.685, which serves as a reasonable accuracy for prediction. The calibration curves also showed high coherence between the observed and predicted disease relapse in the nomogram ( Figure 3B ).

We collected a cohort consist of 14 stage IV patients who received cetuximab treatment ( Table 5 ) from the 5733 patients. All of these 14 patients carried wild-typed KRAS, NRAS, and BRAF. Five patients carried PIK3CA exon 9 mutations, and the rest nine patients carried wild-typed PIK3CA. For wild-typed PIK3CA patients, most of them (7/9) were evaluated as stable disease (SD), one patient was evaluated as partial response (PR), and one patient was evaluated as progressive disease (PD). For PIK3CA exon 9 mutant patients, four patients were evaluated as SD, and one was evaluated as PR. The disease control rate (DCR) was 88.9% (8/9) in wild-typed patients and 100% (5/5) in mutant patients. The progression-free survival (PFS) time ranged from 1.2 to 9.4 months in wild-type patients and from 1.3 to 12.2 months in mutant patients. The median PFS of the two groups did not reach a significant difference (3.6 months vs. 2.3 months, Log-rank test, p-value =0.513).