Abstract
Purpose
To study the somatic molecular profile of the epidermal growth factor receptor (EGFR) pathway in advanced CRC (aCRC), its relationship to prognosis, the site of the primary and metastases, and response to cetuximab.
Experimental Design
We used Sequenom and Pyrosequencing for high-throughput somatic profiling the EGFR pathway in 1,976 tumours from patients with aCRC from the COIN trial (oxaliplatin and fluoropyrimidine chemotherapy ±cetuximab). Correlations between mutations, clinico-pathological, response and survival data were carried out.
Results
Sequenom and Pyrosequencing had 99.0% (9961/10063) genotype concordance. We identified thirteen different KRAS mutations in 42.3% of aCRCs, two BRAF mutations in 9.0%, four NRAS mutations in 3.6% and five PIK3CA mutations in 12.7%. 4.2% of aCRCs had microsatellite instability (MSI). KRAS and PIK3CA exon 9, but not exon 20, mutations co-occurred (P=8.9×10−4) as did MSI and BRAF mutations (P=5.3×10−10). KRAS mutations were associated with right colon cancers (P=5.2×10−5) and BRAF mutations with right (P=7.2×10−5) and transverse colon (P=9.8×10−6) cancers. KRAS mutations were associated with lung-only metastases (P=2.3×10−4), BRAF mutations with peritoneal (P=9.2×10−4) and nodal-only (P=3.7×10−5) metastases, and MSI (BRAFWT) with nodal-only metastases (P=2.9×10−4). MSI (BRAFWT) was associated with worse survival (HR=1.89, 95% CI 1.30-2.76, P=8.5×10−4). No mutations, subsets of mutations, or MSI-status were associated with response to cetuximab.
Conclusions
Our data support a functional co-operation between KRAS and PIK3CA in colorectal tumourigenesis and link somatic profiles to the sites of metastases. MSI was associated with poor prognosis in advanced disease, and no individual somatic profile was associated with response to cetuximab in COIN.
INTRODUCTION
Worldwide, over a million people are diagnosed with colorectal cancer (CRC) each year. CRC is among the cancers showing the greatest improvement in survival (1) and this is due, in part, to advances in drug therapy. Approximately two thirds of patients now receive chemotherapy either as a component of curative treatment or to extend survival with incurable disease, and there is good randomised controlled trial (RCT) evidence of effectiveness in both of these settings (2-4). Patients with advanced CRC (aCRC) have shown further benefits with monoclonal therapies targeting the epidermal growth factor receptor (EGFR). For example, cetuximab increased median survival by 4.7 months compared with best supportive care alone in patients with KRAS-wild type tumours who had exhausted standard cytotoxic therapy options (5).
EGFR acts as the gate-way for multiple downstream intracellular signalling pathways including the RAS-RAF-MAP and PI3K-PTEN-AKT pathways. Through these, EGFR regulates multiple cellular processes including apoptosis, growth, proliferation, differentiation and migration (6). Cetuximab binds to the extracellular binding domain of EGFR, thereby preventing its ligands binding to the receptor and triggers receptor internalisation, thus inhibiting downstream signalling. Response to cetuximab has been suggested to be limited to patients with CRCs wild-type for KRAS. Given the high-frequency of CRCs that are KRAS mutant, guidelines are now in place which recommend testing of KRAS mutation status prior to treatment with anti-EGFR agents (7). However, recent data have indicated that not all somatic mutations within KRAS are refractory to cetuximab, and patients with G13D have longer overall survival (OS) and progression free survival (PFS) after treatment, as compared to patients with other KRAS mutations (8,9). Other genes within the EGFR pathway may also affect response to cetuximab, with efficacy likely to be dependent upon an absence of somatic mutations in BRAF, NRAS and exon 20 of PIK3CA (10).
During embryologic development, the right colon (cecum, ascending colon, proximal two-thirds of the transverse colon) arises from the midgut, and the left colon (distal one-third of the transverse colon, descending and sigmoid colon, rectum) from the hindgut. Differences exist in the macroscopic pathology, histopathology and molecular biological patterns between right-sided (RCCs) and left-sided colon cancers (LCCs) (11-17). Interestingly, RCCs have been associated with peritoneal metastases and LCC with hepatic and pulmonary metastases, and this has been attributed to their different molecular biological patterns causing distinct biological behaviours (18). Furthermore, mutations in KRAS and BRAF, which are more prevalent in CRCs from the right/transverse colon (19,20), appear to be associated with particular sites of metastases. KRAS mutations have been associated with lung (21,22), brain (21) but not liver metastases, and BRAF mutations have been associated with peritoneal (20,23) and distant lymph node metastases (20). Further studies in large independent series are necessary to validate these observations and to help unravel the underlying mechanisms.
Here, we used two mutation detection platforms, Pyrosequencing and Sequenom, for high-throughput somatic profiling of the EGFR pathway in 1,976 tumours from patients with aCRC from the MRC COIN trial who received oxaliplatin and fluoropyrimidine chemotherapy with and without cetuximab. We studied the inter-relationships between the somatic mutations, together with their correlations to the sites of the primary and the metastases, and response to cetuximab.
MATERIALS AND METHODS
Patient samples
COIN (ISRCTN27286448) is a MRC sponsored, CRUK funded, fully accrued 2,445 patient UK national trial, in which patients were randomised in a 1:1:1 ratio to receive continuous oxaliplatin-based chemotherapy (Arm A), continuous chemotherapy plus cetuximab (Arm B), or intermittent chemotherapy (Arm C), in first line treatment of aCRC. All patients chose between oral capecitabine, a 5FU prodrug, (two thirds of patients) or infusional 5FU (one third) as the partner for oxaliplatin prior to randomisation. All patients had measurable metastatic or locally advanced colorectal adenocarcinoma and received no previous chemotherapy for advanced disease. All patients had unresectable disease, whose only treatment option was palliative chemotherapy. Thus, nodal-only metastases included those with distant nodal metastases, such as para-aortic, mediastinal or supraclavicular fossa nodes as well as unresectable nodal recurrence in the pelvis. Similarly, patients with peritoneal disease had extensive peritoneal metastases, which was beyond surgical resectability. All patients gave fully informed consent for their samples to be used for bowel cancer research (approved by REC [04/MRE06/60]). Tumour samples were collected as formalin-fixed, paraffin embedded (FFPE) tissues.
Processing FFPE CRCs, DNA extraction and microsatellite instability (MSI) analyses
5μm sections were cut from FFPE CRCs. One section was stained with Hematoxylin and Eosin and visualised with a Mirax scanner. Samples containing concentrated pockets of tumour material were macrodissected using an unstained section. For samples containing limited regions of tumour, laser capture microdissection (LCM) was performed using 10μm sections cut onto PALM membrane slides (Carl Zeiss, Cambridge, UK) (see Supplementary Information). DNA was extracted using QIAamp DNA Microkits (Qiagen, Hilden, Germany) and eluted in 50μl water. MSI-status was determined using the markers BAT-25 and BAT-26.
Identification of somatic mutation ‘hot spots’ and mutant cell lines
We queried the Catalogue Of Somatic Mutations In Cancer (COSMIC) database (http://www.sanger.ac.uk/genetics/CGP/cosmic) for known common mutations in KRAS, BRAF, NRAS and PIK3CA in CRCs. Cell lines known to carry variants within these genes were identified from the Sanger Cancer Cell Line Project (http://www.sanger.ac.uk/genetics/CGP/CellLines/). We tested the sensitivity of Pyrosequencing and Sequenom to detect low levels of mutant alleles using the cell lines listed in the Supplementary Information. DNA extracted from these lines was quantified using a nanodrop spectrophotometer (Thermo Fisher Scientific, Waltham, MA) and serially diluted with wild-type DNA to generate known levels of mutant alleles. All dilutions were prepared and analysed in triplicate.
Pyrosequencing
For codons 12 and 13 of KRAS, we initially used the amplification primers 5′-GGCCTGCTGAAAATGACTGA-3′ and 5′-AGAATGGTCCTGCACCAGTAATA-3′ together with extension primer 5′-CTTGTGGTAGTTGGAGC-3′; however, this assay was subsequently modified by using the extension primers 5′-TGTGGTAGTTGGAGCTG-3′, 5′-TGTGGTAGTTGGAGCT-3′ and 5′-TGGTAGTTGGAGCTGGT-3′, as previously described (24). For codon 61 of KRAS, we used the amplification primers 5′-CTTTGGAGCAGGAACAATGTC-3′ and 5′-CTCATGTACTGGTCCCTCATTG-3′ together with the extension primer 5′-ATTCTCGACACAGCAGGT-3′, and for codon 600 of BRAF we used the amplification primers 5′-TGCTTGCTCTGATAGGAAAATGA-3′ and 5′-CAGGGCCAAAAATTTAATCAGTG-3′ together with the extension primer 5′-ATTTTGGTCTAGCTACA-3′. Reverse primers were biotinylated. PCR was performed in 50μl reaction volumes containing 25μl Megamix Gold (Microzone, Haywards Heath, UK), 10-20ng DNA and 10μM of primers. Thermocycling was performed at 95°C for 10min, followed by 38 cycles of 95°C for 30s, 57°C for 30s, and 72°C for 1min, followed by a final extension of 72°C for 10min. 40μl of PCR product was used for Pyrosequencing (Biotage AB, Uppsala, Sweden) (see Supplementary Information). Pyrograms were analysed by two independent observers.
Sequenom
Two hundred base pairs of sequence upstream and downstream of each mutation were downloaded from Ensembl to design the genotyping assays using the Sequenom MassARRAY Assay Design 3.1 software. In total, three multiplex assays were designed (Supplementary Table 1). PCR was performed in 5μl reaction volumes containing 0.5U of Taq polymerase, 5-10ng of genomic DNA, 100nM of PCR primers and 500μM of dNTPs. Thermocycling was performed at 95°C for 15 min, followed by 45 cycles of 94°C for 20s, 56°C for 30s and 72°C for 60s, followed by a final extension of 72°C for 3min. Unincorporated dNTPs were deactivated using 0.3U of shrimp alkaline phosphatase at 37°C for 40min, and primer extension was carried out using 7-14μM of each extension primer, 1U of iPLEX termination mix and 1U of iPLEX enzyme. Reactions were cycled at 94°C for 30s, followed by 40 cycles of 94°C for 5s, 52°C for 5s and 80°C for 5s, followed by a final extension at 72°C for 3 min. Extension products were spotted onto a matrix pad of a SpectroCHIP (Sequenom, San Diego, CA). After analysing the SpectroCHIPs using a Bruker MALDI-TOF mass spectrometer, spectra were processed by the SpectroREADER software and transferred to the MassARRAY Typer 4 Analyser. Genotyping was performed using the MassARRAY RTTM software (Sequenom). Automated calls were validated by manual review of the raw mass spectra.
Sanger Sequencing
Sanger sequencing of codons 12, 13 and 61 of KRAS and codon 600 of BRAF was carried out as described in the Supplementary Information.
Statistical analyses
Correlations between somatic mutations and the sites of metastases were carried out using the chi-square test (or Fisher’s exact test for n<5). Time-to-event curves for analysis of OS and PFS were estimated using the Kaplan-Meier method. Hazard ratios, confidence intervals and P-values were estimated using the log-rank method.
RESULTS
Sensitivities of the assays
For Pyrosequencing, we initially designed an assay with a single extension primer to detect mutations at codons 12 and 13 of KRAS. However, we found that G12C was not robustly detected in samples with 25% mutant alleles, so the assay was re-designed to a three extension primer system (24). This modified assay robustly detected G12A and G12D in samples with 12.5% mutant alleles, G12C and G13D in samples with 6% mutant alleles and G12V in samples with 2% mutant alleles. Sequenom robustly detected V600E in BRAF and Q61R in NRAS in samples with 10% mutant alleles, G12C and Q61L in KRAS and E542K, E545K and H1047R in PIK3CA in samples with 6% mutant alleles and G12V in KRAS in samples with 4% mutant alleles (Supplementary Table 2).
Tumour samples
We collected FFPE tumour blocks from 2,161/2,445 (88.4%) patients. One-hundred and eighty-five (8.6%) blocks contained insufficient tumour material for processing. For the remaining samples, 1,893 (95.8%) were from the primary CRC and 66 (3.3%) and 17 (0.9%) were from the liver and lymph node metastases, respectively.
Pyrosequencing and Sequenom genotyping
We screened for somatic mutations in KRAS (codons 12, 13 and 61) and BRAF (codon 600) using both Pyrosequencing and Sequenom, and in BRAF (codon 594), NRAS (codons 12 and 61) and PIK3CA (codons 542, 545, 546 and 1047) using only Sequenom. In total, 1,612 samples were successfully analysed for KRAS mutations using both technologies and 8,642/8,719 (99.1%) genotype calls were concordant (ranging from 97.8% to 99.7% depending on which mutation was analysed). For BRAF V600E, 1,344 samples were successfully analysed using both technologies and 1,319/1,344 (98.1%) genotype calls were concordant. Forty-three out of 77 samples with discordant KRAS calls and 12/25 samples with discordant BRAF calls were successfully Sanger sequenced to infer genotype. For the remaining calls where Sanger sequencing failed, the mutant genotype was selected (since there was an obvious mutant trace via one technology). Both technologies had high genotype success rates; 41,944/43,340 (96.8%) for Sequenom and 21,016/25,200 (83.4%) for Pyrosequencing.
Frequency and distribution of somatic mutations
In total, for KRAS we successfully genotyped 1,949/1,976 samples (98.6%), for BRAF 1,946/1,963 samples (99.1%), for NRAS 1,939/1,963 samples (98.8%) and for PIK3CA 1,907/1,963 samples (97.1%) (Table 1). Fifteen samples (0.8%) consistently failed genotyping. Of the remaining 1961 samples, 96.3% (n=1889) had genotypes for all of these genes and 98.8% (n=1938) had genotypes for at least three genes. In total, we detected thirteen KRAS mutations (G12A, G12D, G12V, G12C, G12R, G12S, G13C, G13D, G13S, G13V, Q61H, Q61L and Q61R, and five remained uncharacterised), two BRAF mutations (D594G and V600E), four NRAS mutations (G12C, Q61K, Q61L and Q61R, and one remained uncharacterised), and five PIK3CA mutations (E542K, E545K, Q546K, H1047L and H1047R). Overall, KRAS mutations were found in 824/1,949 aCRCs (42.3%), BRAF mutations in 175/1,946 aCRCs (9.0%), NRAS mutations in 69/1,939 aCRCs (3.6%), and PIK3CA mutations in 243/1,907 aCRCs (12.7%) (Supplementary Fig. 1). Of the 243 aCRCs with PIK3CA mutations, 162 had a mutation in exon 9 and 82 had a mutation in exon 20 (one sample carried mutations in both exons). MSI was observed in 66/1,565 (4.2%) of the successfully analysed aCRCs.
Table 1.
Mutation frequencies, together with MSI-status, according to trial arm (Arm A - continuous oxaliplatin-based chemotherapy, Arm B - continuous chemotherapy plus cetuximab, Arm C - intermittent chemotherapy). Total numbers per locus to do not exactly match individual numbers since: (i) for KRAS, four samples contained two independent mutations and five other samples contained uncharacterised mutations, (ii) for NRAS, one sample contained an uncharacterised mutation and, (iii) for PIK3CA, six samples contained two independent PIK3CA mutations. In those samples where genotypes were missing for rare mutations (with cumulative frequencies <1%), but where all other mutations were successfully tested as wild type, then an overall call of wild type was made at that locus.
| Arm A | Arm B | Arm C | Total | ||
|---|---|---|---|---|---|
| Gene | Mutation | no. of samples with mutations/no. successfully analysed |
no. of samples with mutations/no. successfully analysed |
no. of samples with mutations/no. successfully analysed |
no. of samples with mutations/no. successfully analysed |
| KRAS | G12A | 23/635 (3.6%) | 11/659 (1.7%) | 19/655 (2.9%) | 53/1949 (2.7%) |
| KRAS | G12D | 74/635 (11.7%) | 94/659 (14.3%) | 73/655 (11.1%) | 241/1949 (12.4%) |
| KRAS | G12V | 59/635 (9.3%) | 82/659 (12.4%) | 67/655 (10.2%) | 208/1949 (10.7%) |
| KRAS | G12C | 23/635 (3.6%) | 14/659 (2.1%) | 31/655 (4.7%) | 68/1949 (3.5%) |
| KRAS | G12R | 6/635 (0.9%) | 5/659 (0.8%) | 3/655 (0.5%) | 14/1949 (0.7%) |
| KRAS | G12S | 14/635 (2.2%) | 20/659 (3.0%) | 9/655 (1.4%) | 43/1949 (2.2%) |
| KRAS | G13C | 3/635 (0.5%) | 2/657 (0.9%) | 3/652 (0.5%) | 8/1944 (0.4%) |
| KRAS | G13S | 0/635 (0%) | 0/657 (0%) | 1/652 (0.2%) | 1/1944 (0.1%) |
| KRAS | G13D | 56/635 (8.8%) | 54/659 (8.2%) | 41/650 (6.3%) | 151/1944 (7.8%) |
| KRAS | G13V | 0/635 (0%) | 1/659 (0.2%) | 0/650 (0%) | 1/1944 (0.1%) |
| KRAS | Q61H | 5/518 (1.0%) | 8/541 (1.5%) | 6/539 (1.1%) | 19/1598 (1.2%) |
| KRAS | Q61L | 2/633 (0.3%) | 3/656 (0.5%) | 3/648 (0.5%) | 8/1937 (0.4%) |
| KRAS | Q61R | 3/633 (0.5%) | 3/656 (0.5%) | 2/648 (0.3%) | 8/1937 (0.4%) |
|
KRAS
total |
268/635 (42.2%) | 297/659 (45.1%) | 259/655 (39.5%) | 824/1949 (42.3%) | |
| BRAF | V600E | 50/632 (7.9%) | 40/661 (6.1%) | 64/651 (9.8%) | 154/1944 (7.9%) |
| BRAF | D594G | 7/622 (1.1%) | 5/655 (0.8%) | 9/649 (1.4%) | 21/1926 (1.1%) |
|
BRAF
total |
57/632 (9.0%) | 45/662 (6.8%) | 73/652 (11.2%) | 175/1946 (9.0%) | |
| NRAS | G12C | 0/621 (0%) | 11/653 (1.7%) | 6/634 (1.0%) | 17/1908 (0.9%) |
| NRAS | Q61K | 10/612 (1.6%) | 12/634 (1.9%) | 6/624 (1.0%) | 28/1870 (1.5%) |
| NRAS | Q61L | 2/633 (0.3%) | 5/652 (0.8%) | 2/646 (0.3%) | 9/1931 (0.5%) |
| NRAS | Q61R | 6/633 (1.0%) | 3/652 (0.5%) | 5/646 (0.8%) | 14/1931 (0.7%) |
|
NRAS
total |
18/631 (2.9%) | 32/659 (4.9%) | 19/649 (2.9%) | 69/1939 (3.6%) | |
| PIK3CA | E542K | 20/614 (3.3%) | 22/638 (3.4%) | 19/640 (3.0%) | 61/1892 (3.2%) |
| PIK3CA | E545K | 27/614 (4.4%) | 29/637 (4.6%) | 31/636 (4.9%) | 87/1887 (4.6%) |
| PIK3CA | Q546K | 3/612 (0.5%) | 10/630 (1.6%) | 6/623 (1.0%) | 19/1865 (1.0%) |
| PIK3CA | H1047L | 10/616 (1.6%) | 13/633 (2.1%) | 13/639 (2.0%) | 36/1888 (1.9%) |
| PIK3CA | H1047R | 13/616 (2.1%) | 14/633 (2.2%) | 19/639 (3.0%) | 46/1888 (2.4%) |
|
PIK3CA
total |
71/620 (11.5%) | 85/643 (13.2%) | 87/644 (13.5%) | 243/1907 (12.7%) | |
| MSI | 19/502 (3.8%) | 26/520 (5.0%) | 21/543 (3.9%) | 66/1565 (4.2%) |
KRAS and BRAF mutations were, in general, mutually exclusive (only four tumours carried mutations in both oncogenes, 0.2%). CRCs with mutations in both KRAS and NRAS were seen at a low level (n=14, 0.7%) (Supplementary Fig.1). Mutations in PIK3CA were more frequently observed in those CRCs with KRAS mutations (127/797, 15.9%) than in those without KRAS mutations (114/1,101, 10.4%; P=3.1×10−4). This association was most striking in CRCs harbouring exon 9 PIK3CA mutations (88/797 co-occurred, 11.0% versus 74/1,101, 6.7% without KRAS mutation; P=8.9×10−4), and was not observed in those with exon 20 PIK3CA mutations (40/797 co-occurred, 5.0%, versus 40/1,101, 3.6% without KRAS mutation P=0.14). MSI was noted within all somatic mutation sub-groups, although there was a significant correlation with BRAF mutations (20/139, 14.4% of BRAF mutant tumours had MSI versus 46/1,414, 3.3% BRAF wild-type tumours, P=5.3×10−10). All of these associations remained significant after correction for multiple testing.
Somatic mutation status and site of the primary tumour
KRAS mutations were more common in CRCs from the right colon (52.7%) as compared to those from the left colon (38.2%, P=5.2×10−5; Table 2) and BRAF mutations were more common in CRCs from the transverse (28.6%) and right (17.4%) colon as compared to those from the left colon (7.8%, P=9.8×10−6 and 7.2×10−5, respectively; Table 2). PIK3CA mutations were more common in CRCs from the transverse colon (27.1%) as compared to the left colon (13.3%), which although significant (P=0.01), did not withstand rigorous correction for multiple testing (Table 2).
Table 2.
Correlations between the site of the primary and somatic mutation status. Numbers of tumours with a mutation in the respective oncogene or with MSI against the total number of specimens tested, with percentages in parentheses.
| Mutation | Right colon | Transverse colon | Left colon | Rectosigmoid junction | Rectum | ||||
|---|---|---|---|---|---|---|---|---|---|
| n | n | P | n | P | n | P | n | P | |
| KRAS | 256/486 (52.7) |
15/50 (30.0) |
2.3×10−3 *4.5×10−2 |
128/334 (38.2) |
5.2×10−5 *1.0×10−3 |
160/470 (34.0) |
6.3×10−9 *1.3×10−7 |
263/599 (43.9) |
4.0×10−3 *8.0×10−2 |
| BRAF | 85/488 (17.4) |
14/49 (28.6) |
5.5×10−2 *1.0 |
26/334 (7.8) |
7.2×10−5 *1.4×10−3 |
24/470 (5.1) |
2.0×10−9 *4.0×10−8 |
25/595 (4.2) |
7.8×10−13 *1.6×10−11 |
| NRAS | 14/486 (2.9) |
1/49 (2.0) |
1.0 *1.0 |
11/335 (3.3) |
0.74 *1.0 |
23/468 (4.9) |
0.10 *1.0 |
18/591 (3.1) |
0.87 *1.0 |
| PIK3CA | 69/473 (14.6) |
13/48 (27.1) |
2.4×10−2 *0.47 |
44/332 (13.3) |
0.57 *1.0 |
54/461 (11.7) |
0.19 *1.0 |
62/583 (10.6) |
5.3×10−2 *1.0 |
| †MSI (BRAFWT) | 22/319 (6.9) |
1/30 (3.3) |
0.71 *1.0 |
8/240 (3.3) |
6.4×10−2 *1.0 |
6/357 (1.7) |
6.8×10−4 *1.4×10−2 |
9/460 (2.0) |
5.2×10−4 *1.0×10−2 |
Since MSI and BRAF mutations significantly co-occurred, data shown for MSI BRAFWT.
P-values as compared to the frequency of the respective mutation in patients with a primary tumour in the right colon (*after correction for multiple testing, n=20).
Somatic mutation status and site of the metastases
We tested whether the somatic mutation status in the primary CRCs correlated with the site of the metastases and initially analysed 815 patients from COIN that had a single organ metastatic site. Of these, 547 had liver-only, 128 had lung-only, 93 had nodal-only and 47 had peritoneal-only metastases. We generated somatic mutation data on the primary CRCs from 79.3%, 75.0%, 87.1% and 74.5% of these patients, respectively. Significantly, 57.3% of patients with lung-only metatases had KRAS mutations as compared to 36.9% of patients with liver-only metastases (P=2.3×10−4, Table 3, Fig.1). This association was primarily driven by the mutations G12V and G12C (Supplementary Table 3). BRAF mutations were significantly more common in patients with peritoneal-only (22.2%) and nodal-only (21.0%) metastases, as compared to those with liver-only metastases (6.7%; P=9.2×10−4 and 3.7×10−5, respectively). MSI was also more common in patients with nodal (21.2%) and peritoneal-only (12.9%) metastases as compared to those with liver-only metastases (2.0%; P=4.0×10−11 and 7.5×10−3, respectively), although the association with peritoneal metastases did not withstand correction for multiple testing. The association with nodal metastases was more frequent in those that also carried a BRAF mutation (66.7% as compared to 4.2% of patients with liver-only metastases with MSI and BRAF mutations; P=1.3×10−4), but was also observed in those with BRAF wild type tumours (11.1% as compared to 1.8% patients with liver-only tumours; P=2.9×10−4) (Table 3).
Table 3.
Correlations between the site of the metastases and somatic mutation status (in patients that had a single organ metastatic site). Numbers of tumours with a mutation in the respective oncogene or with MSI against the total number of specimens tested, with percentages in parentheses.
| Mutation | Liver only | Lung only | Nodes only | Peritoneum only | |||
|---|---|---|---|---|---|---|---|
| n | n | P | n | P | n | P | |
| KRAS | 160/434 (36.9) |
55/96 (57.3) |
2.3×10−4 *3.5×10−3 |
27/81 (33.3) |
0.54 *1.0 |
14/35 (40.0) |
0.71 *1.0 |
| BRAF | 29/432 (6.7) |
5/97 (5.2) |
0.38 *1.0 |
17/81 (21.0) |
3.7×10−5 *5.6×10−4 |
8/36 (22.2) |
9.2×10−4 *1.4×10−2 |
| NRAS | 11/432 (2.5) |
2/95 (2.1) |
0.58 *1.0 |
3/81 (3.7) |
0.38 *1.0 |
0/36 (0) |
0.41 *1.0 |
| PIK3CA | 58/429 (13.5) |
10/93 (10.8) |
0.47 *1.0 |
10/81 (12.3) |
0.78 *1.0 |
2/35 (5.7) |
0.14 *1.0 |
| †MSI (BRAFWT) | 6/328 (1.8) |
1/77 (1.3) |
0.60 *1.0 |
6/54 (11.1) |
2.9×10−4 *4.4×10−3 |
2/24 (8.3) |
9.7×10−2 *1.0 |
Since MSI and BRAF mutations significantly co-occurred, data shown for MSI BRAFWT.
P-values as compared to the frequency of the respective mutation in patients with liver-only metastases (*after correction for multiple testing, n=15).
Figure 1.
Histogram showing the distribution of somatic mutations according to the site of the metastases (in patients that had a single organ metastatic site). KRAS mutations were associated with lung-only metastases, BRAF mutations with peritoneal and nodal-only metastases and MSI (BRAFWT) with nodal-only metastases.
We determined whether these correlations were maintained in any patients with lung (n=784), nodal (n=906) or peritoneal (n=283) metastases (Supplementary Table 4). KRAS mutations were associated with lung and peritoneal metastases (48.2% and 46.5% of patients, respectively, as compared to 36.9% with liver-only metastases, P=1.3×10−4 and 1.1×10−2, respectively), BRAF mutations with peritoneal metastases (12.7% of patients as compared to 6.7% with liver-only metastases, P=6.3×10−3), NRAS mutations with lung metastases (5.5% of patients as compared to 2.5% with liver-only metastases, P=1.6×10−2), and MSI (BRAF wild type) with peritoneal metastases (5.1% of patients as compared to 1.8% with liver-only metastases P=3.4×10−2). However, only the correlation between KRAS mutations and lung metastases was maintained after rigorous correction for multiple testing (P=2.0×10−3).
We also determined whether the correlations between somatic mutation status and site of metastases were maintained when we did not use the ‘liver-only metastatic group’ as the reference cohort. We simply considered the frequency of the somatic mutations in any patients with, versus those without, a specific metastatic site. We found that KRAS mutations were associated with more patients with lung metastases (P=1.3×10−5), BRAF mutations with fewer patients with liver (P=1.6×10−3) and lung (P=4.0×10−5) metastases, NRAS mutations with more patients with lung metastases (P=1.2×10−4) and MSI (BRAF wild-type) with fewer patients with liver metastases (P=1.6×10−3) (Supplementary Table 5). All of these associations remained significant after correction for multiple testing.
Mutation status, survival and response to cetuximab
We have previously shown that the addition of cetuximab to standard chemotherapy did not improve OS or PFS in patients from COIN with KRAS, BRAF and NRAS wild-type tumours (25). However, exploratory analyses revealed that cetuximab had a borderline improvement in PFS among such patients when used with oxaliplatin and infusional 5FU (OxFU) (HR 0.72, 95% CI 0.53–0.98, P=3.7×10−2), but not oral 5FU (HR 1.02, 0.82–1.26, P=0.88). Here, we tested whether PIK3CA mutation status, MSI-status, individual somatic KRAS, BRAF, NRAS and PIK3CA mutations or somatic mutations grouped by exon, were associated with response to cetuximab in either the full cohort or those treated with OxFU.
Irrespective of the treatment arm (to search for potential prognostic effects), PIK3CA mutation status did not affect OS (HR=0.91, 95% CI 0.75-1.11, P=0.37) or PFS (HR=1.06, 95% CI 0.89-1.26, P=0.49). This was regardless of whether PIK3CA mutations were split between those in exon 9 (OS HR=0.88, 95% CI 0.70-1.12, P=0.31; PFS HR=1.07, 95% CI 0.87-1.32, P=0.51) and those in exon 20 (OS HR=0.92, 95% CI 0.65-1.28, P=0.61; PFS HR=1.00, 95% CI 0.74-1.36, P=0.99). MSI was associated with worse survival independent of somatic mutation status, treatment arm and chemotherapy regimen (adjusted OS HR=1.60, 95% CI 1.14-2.24, P=6.6×10−3; adjusted PFS HR=1.66, 95% CI 1.21-2.27, P=1.6×10−3). Significantly, MSI was still associated with worse survival in the BRAF wild type subgroup (adjusted OS HR=1.89, 95% CI 1.30-2.76, P=8.5×10−4; adjusted PFS HR=1.85, 95% CI 1.31-2.61, P=5.1×10−4) (Fig. 2).
Figure 2.
Prognostic effect of MSI (BRAFWT) in the advanced disease setting. MSI (BRAFWT) was associated with worse OS (A) and PFS (B), independent of somatic mutation status, treatment arm and chemotherapy regimen.
In terms of predicting response to cetuximab, patients with PIK3CA wild type tumours did not show improved OS or PFS, regardless of chemotherapy regimen (OS HR 1.01, 95% CI 0.88-1.16, P=0.84 [any chemotherapy] and HR 0.96, 95% CI 0.76-1.21, P=0.73 [OxFU]; PFS HR 0.99, 95% CI 0.88-1.12, P=0.90 [any chemotherapy] and HR 0.88, 95% CI 0.71-1.09, P=0.25 [OxFU]) (Supplementary Fig.2) or somatic KRAS status (OS HR 1.03, 95% CI 0.86-1.24, P=0.74 and PFS HR 0.92, 95% CI 0.78-1.08, P=0.30 [KRAS wild type]) (Supplementary Fig.3).
Patients with MSI appeared to have worse survival after treatment with cetuximab, but this was not statistically significant due to the small numbers of patients (OS HR 1.27, 95% CI 0.65-2.45, P=0.49; PFS HR 1.26, 95% CI 0.68-2.34, P=0.47, n=45 [any chemotherapy]). We did not observe any individual somatic KRAS, BRAF, NRAS or PIK3CA mutations, or mutations grouped by exons, that were associated with response to cetuximab within those patients treated with any chemotherapy or OxFU after correction for multiple testing (Fig.3).
Figure 3.
Forest plots depicting PFS hazard ratios for individual mutations in KRAS, BRAF, NRAS and PIK3CA, and MSI-status for patients that received (A) any chemotherapy, and, (B) OxFU. P-values are uncorrected for multiple testing (none were significant after Bonferroni correction). Note - For some mutations there was insufficient data to estimate a HR.
DISCUSSION
Here, we used Pyrosequencing and Sequenom for high-throughput somatic mutation detection in the EGFR pathway. Both assays robustly detected low levels of mutant alleles (provided that the Pyrosequencing extension primers were designed to generate de novo peaks) with all mutations being detected in samples with 12.5% mutant alleles. Previous studies have also suggested that these platforms have equal mutation detection sensitivities, with superior detection limits as compared to traditional Sanger sequencing (26). Our analyses based on ~10,000 somatic genotypes, showed that these platforms had 99% genotype concordance and high genotype success rates. However, the main advantage of Sequenom was the ability to multiplex the reactions and herein we describe a three-assay system to screen for 33 somatic mutations within the EGFR pathway.
Our comprehensive somatic profiling of the RAS-RAF-MAP and PI3K-PTEN-AKT sub-pathways in 1,976 aCRCs has allowed us to interrogate any functional co-operations between these pathways in colorectal tumourigenesis. We found that exon 9, but not exon 20, mutations in PIK3CA were associated with KRAS mutations. The reason for this association is likely to be due to the independent biological roles played by different mutations in PIK3CA. Exon 9 mutations lie in the helical domain of the protein while exon 20 mutations lie in the kinase domain (27). Although mutations of both exons result in activated AKT signalling, they have different requirements for interaction with the PI3K regulatory subunit p85 and with GTP bound RAS. The gain of function coinciding with exon 9 mutations requires interaction with GTP bound RAS, but is independent of binding to p85. Conversely, mutations of exon 20 require p85 binding but are independent of GTP bound RAS (27). Somatic profiling of other oncogenes and tumour suppressor genes within our collection of aCRCs are likely to provide further insights into the different mechanisms of colorectal tumourigenesis.
We have previously shown that KRAS and BRAF mutations confer a poor prognosis in patients with aCRC irrespective of treatment (OS - KRAS mutant, 14.4 months; BRAF mutant, 8.8 months; all wild-type, 20.1 months) (25). Here, we show that MSI is also associated with poor prognosis in patients with aCRC (OS - 9.3 months). Consistent with this, a study of 524 aCRC patients showed that those with tumours with MSI had significantly worse survival as compared to those with tumours that were microsatellite-stable (11.1 months versus 22.1 months, P<0.001) and this association was maintained in the subset of BRAF wild-type tumours (20). It is noteworthy that in the earlier stages of colorectal tumourigenesis, MSI is more frequently observed and is associated with good prognosis (19). These differences warrant further investigation and may be influenced by an interaction with the fluoropyrimidine-based treatment used in COIN.
In agreement with others (20-23), we found that KRAS mutations were highly associated with lung-only metastases and BRAF mutations with peritoneal and nodal-only metastases. In addition, we noted that MSI was associated with nodal-only metastases and all of these associations were maintained after correction for multiple testing. We also found that the associations between KRAS mutations and lung metastases, and BRAF mutations with peritoneal metastases were maintained in the larger and more heterogeneous cohort of ‘any’ patients with those metastases, although only the KRAS association then remained significant after correction for multiple testing. Even when we performed these analyses without using a liver-only cohort as a reference group, we still found that the association between KRAS mutations and lung metastases was maintained. Our data therefore show that different somatic profiles are associated with different metastatic sites, suggesting that these profiles influence the tumour’s biological behaviour (20). It is important to note that in terms of clinical utility these profiles may be of limited use. No marker could be used to reliably predict a metastatic site, and patients are routinely monitored with computed tomography scans which include thorax, abdomen and pelvis anyway.
Although it is clear from existing data that there is a paucity of KRAS and BRAF mutations in tumours from the left colon, it is not clear whether there is a demarcation between KRAS and BRAF mutant tumours in the ascending and transverse colons. In the COIN trial, we collected data on the location of the primary tumour using right and left colon, rectosigmoid junction, rectum and other (in which transverse colon was often reported). Although we noted an excess of KRAS and BRAF mutations in tumours from the right colon, we also noted an excess of BRAF mutations in CRCs from the transverse colon (28.6%) which was not seen for KRAS mutations. This observation should be treated with caution since the junction between mid and hind gut (and hence the embryological separation of the right and left colon) occurs at a variable location within the transverse colon; however, these data suggest that future studies should categorise CRCs into ascending, transverse and descending colon to help identify potential differences between CRCs harbouring mutations in these oncogenes. Such differences may help explain the different patterns of metastatic spread.
We have previously reported that cetuximab had a borderline improvement in PFS amongst KRAS, BRAF and NRAS wild type patients treated with OxFU. In the more comprehensive analysis reported herein, this association failed to reach statistical significance (P=0.06). Interestingly, others have reported that patients with PIK3CA exon 20 mutations had worse response to cetuximab (10) and that different somatic mutations in KRAS have differential effects on response (8,9). However, our data do not support either of these observations which may be a consequence of the complicating gastrointestinal toxic effects of the drugs used in the COIN trial, a mechanistic interaction between cetuximab and oxaliplatin, or the use of small cohorts of patients harbouring individual somatic mutations.
Supplementary Material
TRANSLATIONAL RELEVANCE.
Cetuximab, a monoclonal antibody against the epidermal growth factor receptor (EGFR), improves overall survival in patients with advanced CRC (aCRC) in whom other treatments have failed. Efficacy is thought to be dependent upon an absence of somatic mutations in members of the EGFR signalling cascade. We profiled this pathway in 1,976 tumours from patients with aCRC that had been treated with oxaliplatin and fluoropyrimidine chemotherapy ±cetuximab (from the MRC COIN trial). KRAS and PIK3CA exon 9 mutations significantly co-occurred as did MSI and BRAF mutations, supporting their functional co-operation in colorectal tumourigenesis. KRAS mutations were associated with lung-only metastases, BRAF mutations with peritoneal and nodal-only metastases and MSI (BRAFWT) with nodal-only metastases, suggesting that different biological patterns influence tumour behaviour. MSI was associated with worse survival, independent of somatic mutation status. No individual mutation, subsets of mutations, or MSI-status were associated with response to cetuximab in the COIN trial.
ACKNOWLEDGEMENTS
We thank the patients and their families who participated in COIN and gave their consent for this research, and the investigators and pathologists throughout the UK who submitted samples for assessment. We thank Hood Mugalassi and Amanda Gilkes for helpful advice and Tony Lai for use of laboratory space in the Wales Heart Research Institute, Cardiff University. COIN was conducted with the support of the National Institute of Health Research Cancer Research Network.
GRANT SUPPORT This work was supported by the Bobby Moore Fund from Cancer Research UK, Cancer Research Wales and the NISCHR Cancer Genetics Biomedical Research Unit (all to J.P.C.). The COIN trial was funded by Cancer Research UK and the MRC. An unrestricted educational grant from Merck Serono provided additional support to this work and the trial (to T.S.M. and J.P.C.). B.C. was supported by a Ph.D. grant of the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen).
Conflict of interest: This work was partly supported by an unrestricted educational grant from Merck Serono (to T.S.M. and J.P.C.). D. Lambrechts is consultant/advisory board member and has commercial research grant from Roche and Sanofi and T.S. Maughan has honoraria from speakers’ bureau from Merck serono.
REFERENCES
- 1.Karim-Kos HE, de Vries E, Soerjomataram I, Lemmens V, Siesling S, Coebergh JW. Recent trends of cancer in Europe: a combined approach of incidence, survival and mortality for 17 cancer sites since the 1990s. Eur J Cancer. 2008;44:1345–89. doi: 10.1016/j.ejca.2007.12.015. [DOI] [PubMed] [Google Scholar]
- 2.André T, Boni C, Navarro M, Tabernero J, Hickish T, Topham C, et al. Improved overall survival with oxaliplatin, fluorouracil, and leucovorin as adjuvant treatment in stage II and III colon cancer in the MOSAIC trial. J Clin Oncol. 2009;27:3109–16. doi: 10.1200/JCO.2008.20.6771. [DOI] [PubMed] [Google Scholar]
- 3.Saltz LB, Cox JV, Blanke C, Rosen LS, Fehrenbacher L, Moore MJ, et al. Irinotecan plus fluorouracil and leucovorin for metastatic colorectal cancer. Irinotecan Study Group. N Engl J Med. 2000;343:905–14. doi: 10.1056/NEJM200009283431302. [DOI] [PubMed] [Google Scholar]
- 4.Seymour MT, Maughan TS, Ledermann JA, Topham C, James R, Gwyther SJ, et al. FOCUS Trial Investigators; National Cancer Research Institute Colorectal Clinical Studies Group. Different strategies of sequential and combination chemotherapy for patients with poor prognosis advanced colorectal cancer (MRC FOCUS): a randomised controlled trial. Lancet. 2007;370:143–52. doi: 10.1016/S0140-6736(07)61087-3. [DOI] [PubMed] [Google Scholar]
- 5.Karapetis CS, Khambata-Ford S, Jonker DJ, O’Callaghan CJ, Tu D, Tebbutt NC, Simes RJ, et al. K-ras mutations and benefit from cetuximab in advanced colorectal cancer. N Engl J Med. 2008;359:1757–65. doi: 10.1056/NEJMoa0804385. [DOI] [PubMed] [Google Scholar]
- 6.Woodburn JR. The Epidermal Growth Factor Receptor and Its Inhibition in Cancer Therapy. Pharmacology & Therapeutics. 1999;82:241–50. doi: 10.1016/s0163-7258(98)00045-x. [DOI] [PubMed] [Google Scholar]
- 7.Allegra CJ, Jessup JM, Somerfield MR, Hamilton SR, Hammond EH, Hayes DF, et al. American Society of Clinical Oncology provisional clinical opinion: testing for KRAS gene mutations in patients with metastatic colorectal carcinoma to predict response to anti-epidermal growth factor receptor monoclonal antibody therapy. J Clin Oncol. 2009;27:2091–6. doi: 10.1200/JCO.2009.21.9170. [DOI] [PubMed] [Google Scholar]
- 8.De Roock W, Jonker DJ, Di Nicolantonio F, Sartore-Bianchi A, Tu D, Siena S, et al. Association of KRAS p.G13D mutation with outcome in patients with chemotherapy-refractory metastatic colorectal cancer treated with cetuximab. JAMA. 2010;304:1812–20. doi: 10.1001/jama.2010.1535. [DOI] [PubMed] [Google Scholar]
- 9.Tejpar S, Celik I, Schlichting M, Sartorius U, Bokemeyer C, Van Cutsem E. Association of KRAS G13D tumor mutations with outcome in patients with metastatic colorectal cancer treated with first-line chemotherapy with or without cetuximab. J Clin Oncol. 2012;30:3570–7. doi: 10.1200/JCO.2012.42.2592. [DOI] [PubMed] [Google Scholar]
- 10.De Roock W, Claes B, Bernasconi D, De Schutter J, Biesmans B, Fountzilas G, et al. Effects of KRAS, BRAF, NRAS, and PIK3CA mutations on the efficacy of cetuximab plus chemotherapy in chemotherapy-refractory metastatic colorectal cancer: a retrospective consortium analysis. Lancet Oncol. 2010;11:753–62. doi: 10.1016/S1470-2045(10)70130-3. [DOI] [PubMed] [Google Scholar]
- 11.Nawa T, Kato J, Kawamoto H, Okada H, Yamamoto H, Kohno H, et al. Differences between right- and left-sided colon cancer in patient characteristics, cancer morphology and histology. J Gastroenterol Hepatol. 2008;23:418–23. doi: 10.1111/j.1440-1746.2007.04923.x. [DOI] [PubMed] [Google Scholar]
- 12.Gervaz P, Bouzourene H, Cerottini JP, Chaubert P, Benhattar J, Secic M, et al. Dukes B colorectal cancer: distinct genetic categories and clinical outcome based on proximal or distal tumor location. Dis Colon Rectum. 2001;44:364–72. doi: 10.1007/BF02234734. [DOI] [PubMed] [Google Scholar]
- 13.Lanza G, Jr, Maestri I, Ballotta MR, Dubini A, Cavazzini L. Relationship of nuclear DNA content to clinicopathologic features in colorectal cancer. Mod Pathol. 1994;7:161–5. [PubMed] [Google Scholar]
- 14.Sugai T, Habano W, Jiao YF, Tsukahara M, Takeda Y, Otsuka K, et al. Analysis of molecular alterations in left- and right-sided colorectal carcinomas reveals distinct pathways of carcinogenesis: proposal for new molecular profile of colorectal carcinomas. J Mol Diagn. 2006;8:193–201. doi: 10.2353/jmoldx.2006.050052. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Bleeker WA, Hayes VM, Karrenbeld A, Hofstra RM, Hermans J, Buys CC, et al. Impact of KRAS and TP53 mutations on survival in patients with left- and right-sided Dukes’ C colon cancer. Am J Gastroenterol. 2000;95:2953–7. doi: 10.1111/j.1572-0241.2000.02327.x. [DOI] [PubMed] [Google Scholar]
- 16.Fric P, Sovová V, Sloncová E, Lojda Z, Jirásek A, Cermák J. Different expression of some molecular markers in sporadic cancer of the left and right colon. Eur J Cancer Prev. 2000;9:265–8. doi: 10.1097/00008469-200008000-00006. [DOI] [PubMed] [Google Scholar]
- 17.Soong R, Powell B, Elsaleh H, Gnanasampanthan G, Smith DR, Goh HS, et al. Prognostic significance of TP53 gene mutation in 995 cases of colorectal carcinoma. Influence of tumour site, stage, adjuvant chemotherapy and type of mutation. Eur J Cancer. 2000;36:2053–60. doi: 10.1016/s0959-8049(00)00285-9. [DOI] [PubMed] [Google Scholar]
- 18.Benedix F, Kube R, Meyer F, Schmidt U, Gastinger I, Lippert H, Colon/Rectum Carcinomas (Primary Tumor) Study Group Comparison of 17,641 patients with right- and left-sided colon cancer: differences in epidemiology, perioperative course, histology, and survival. Dis Colon Rectum. 2010;53:57–64. doi: 10.1007/DCR.0b013e3181c703a4. [DOI] [PubMed] [Google Scholar]
- 19.Hutchins G, Southward K, Handley K, Magill L, Beaumont C, Stahlschmidt J, et al. Value of mismatch repair, KRAS, and BRAF mutations in predicting recurrence and benefits from chemotherapy in colorectal cancer. J Clin Oncol. 2011;29:1261–70. doi: 10.1200/JCO.2010.30.1366. [DOI] [PubMed] [Google Scholar]
- 20.Tran B, Kopetz S, Tie J, Gibbs P, Jiang ZQ, Lieu CH, et al. Impact of BRAF mutation and microsatellite instability on the pattern of metastatic spread and prognosis in metastatic colorectal cancer. Cancer. 2011;117:4623–32. doi: 10.1002/cncr.26086. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Tie J, Lipton L, Desai J, Gibbs P, Jorissen RN, Christie M, et al. KRAS mutation is associated with lung metastasis in patients with curatively resected colorectal cancer. Clin Cancer Res. 2011;17:1122–30. doi: 10.1158/1078-0432.CCR-10-1720. [DOI] [PubMed] [Google Scholar]
- 22.Cejas P, López-Gómez M, Aguayo C, Madero R, de Castro Carpeño J, Belda-Iniesta C, et al. KRAS mutations in primary colorectal cancer tumors and related metastases: a potential role in prediction of lung metastasis. PLoS One. 2009;4:e8199. doi: 10.1371/journal.pone.0008199. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Yokota T, Ura T, Shibata N, Takahari D, Shitara K, Nomura M, et al. BRAF mutation is a powerful prognostic factor in advanced and recurrent colorectal cancer. Br J Cancer. 2011;104:856–62. doi: 10.1038/bjc.2011.19. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Ogino S, Kawasaki T, Brahmandam M, Yan L, Cantor M, Namgyal C, et al. Sensitive sequencing method for KRAS mutation detection by Pyrosequencing. J Mol Diagn. 2005;7:413–21. doi: 10.1016/S1525-1578(10)60571-5. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Maughan TS, Adams RA, Smith CG, Meade AM, Seymour MT, Wilson RH, et al. MRC COIN Trial Investigators. Addition of cetuximab to oxaliplatin-based first-line combination chemotherapy for treatment of advanced colorectal cancer: results of the randomised phase 3 MRC COIN trial. Lancet. 2011;377:2103–14. doi: 10.1016/S0140-6736(11)60613-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 26.Thomas RK, Baker AC, Debiasi RM, Winckler W, Laframboise T, Lin WM, et al. High-throughput oncogene mutation profiling in human cancer. Nat Genet. 2007;39:347–51. doi: 10.1038/ng1975. [DOI] [PubMed] [Google Scholar]
- 27.Zhao L, Vogt PK. Helical domain and kinase domain mutations in p110alpha of phosphatidylinositol 3-kinase induce gain of function by different mechanisms. Proc Natl Acad Sci USA. 2008;105:2652–2657. doi: 10.1073/pnas.0712169105. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.