Comprehensive analysis of mutational and clinicopathologic characteristics of poorly differentiated colorectal neuroendocrine carcinomas

Poorly differentiated neuroendocrine carcinoma (NEC) is a rare subtype of colorectal cancer (CRC). This study aimed to investigate clinicopathologic characteristics of colorectal NECs and elucidate genomic differences and similarities between colorectal NECs and colorectal adenocarcinomas (ACs). A total of 30 colorectal NECs were screened for frequently identified CRC oncogenic driver genes by targeted next-generation sequencing of 382 genes. The median age of the patients was 67 years (range, 44 to 88 years). NECs occurred predominantly in the rectum (47%) and exhibited multiple adverse prognostic pathologic factors, including frequent lymphatic and vascular invasions, high rates of lymph node metastasis and distant metastasis and advanced TNM stage. The 1-, 3-, and 5-year overall survival rates of NEC patients were 46.7%, 36.4%, and 32.7%, respectively, with a median overall survival period of 11.5 months. In a molecular analysis, NECs showed high rates of BRAF mutation (23%), predominantly p.V600E (71%), and alterations in RB1 (47%), particularly deletion (57%). The frequencies and distributions of other genes, such as KRAS, APC, SMAD4, and PIK3CA, and microsatellite instability status were similar to those of ACs. These findings provide beneficial information for selecting therapeutic options, including targeted therapy, and a better understanding of the histogenesis of this tumour.

www.nature.com/scientificreports/ from 4 to 59%, as revealed by Sanger sequencing or targeted next-generation sequencing. In addition, KRAS mutations occur in 8% to 70% of cases, and the frequency of TP53 mutation reportedly ranges from 21 to 80%. In a recent series of 25 colorectal NECs, Shamir et al. found RB1 alterations in 14 tumours (56%), with TP53 being mutated in 12 (48%), similar to the frequencies of small-cell carcinoma of the lung 11 . In contrast, other frequently identified somatic mutations in colorectal ACs, such as mutations in BRAF, KRAS, and PIK3CA, but not in APC, were less common than in other studies [7][8][9][10] . Due to the rarity of this disease entity, the frequencies of identified mutations in colorectal NECs have been inconsistently described. Therefore, our study adds an additional 30 pure NEC cases to the previous literature, which helps to validate the prior molecular data of colorectal NECs. It is known that different frequencies and distributions of mutations in key oncogenes and tumour suppressors exist between right-and left-sided CRCs. In comparison analysis of frequently identified somatic mutations in right-and left-sided CRCs from the Cancer Genome Atlas dataset, BRAF mutations, particularly p.V600E, are significantly more common in right-sided CRC (24.2% vs. 2.1%) 12 . Right-sided CRCs had higher rates of microsatellite instability and PIK3CA mutations and increased mutational burden, whereas mutations in APC and TP53 were enriched in left-sided CRC (81.9% vs. 63.6% and 64.6% vs. 34.8%). Considering that the different mutational rates of key oncogenic driver genes of CRCs depend on the tumour site, a specific analysis would involve comparison of identified mutations of colorectal NECs with those of colorectal ACs after tumour site matching.
The main purpose of this study was to better elucidate the clinicopathologic features and clinical outcome of colorectal NECs and genomic distinctions as well as similarities between these tumours and conventional ACs. We also investigated potential therapeutically targetable molecular alterations in colorectal NECs to optimize patient selection for new drugs and their combinations.

Methods
Patient selection and histopathology. The institutional review board of Asan Medical Center (AMC), Seoul, South Korea, approved this study. All methods were performed in accordance with relevant guidelines and regulations. We reviewed the records of patients diagnosed with poorly differentiated NECs and neuroendocrine tumours of WHO grade 3 in the colorectum from the Pathology Department at AMC. A total of thirty-five surgically resected cases were initially searched and reviewed. Neuroendocrine markers including synaptophysin and chromogranin and Ki-67 staining were performed for all thirty-five cases. Among them, well-differentiated neuroendocrine tumors of grade 3 (Ki-67 > 20% or mitoses > 20/2 mm 2 ) and posttherapy specimens of colorectal NECs were excluded. Mixed adenoneuroendocrine carcinoma (MANEC) is a mixed malignant neoplasm with a neuroendocrine component combined with a glandular component. Each component should account for at least 30% of the tumor cell population 6 . To maximize DNA extraction of the pure component of NEC for molecular analysis, cases of MANEC were also excluded. Poorly differentiated NEC cases with more than 70% neuroendocrine components were selected for molecular and immunohistochemical analyses. Tumours were morphologically classified into small-cell or large-cell types according to the WHO classification criteria 6 . Tumours consisting of small to medium-sized cells with a high nuclear-cytoplasmic ratio, scant cytoplasm, and fusiform nuclei containing fine granular chromatin without prominent nucleoli were considered small-cell NEC. Large-cell NECs were considered those consisting of large polygonal cells with round nuclei, moderate amounts of cytoplasm, and sometimes prominent nucleoli. The mitotic index was obtained by evaluating the most mitotically active 2 mm 2 area of the tumour. For a control group, initially, 120 surgical cases of colorectal AC after matching the tumour site were searched from the pathology report profile and reviewed. In cases of suspicious neuroendocrine differentiation within a tumour, immunohistochemistry for neuroendocrine markers was performed to confirm neuroendocrine differentiation. Ultimately, 100 surgical cases of colorectal ACs without neuroendocrine differentiation were selected as a control group. Histologic findings including mitotic count, depth of invasion, lymphatic invasion, vascular invasion, perineural invasion, regional lymph node metastasis, distant metastasis, and pathologic TNM staging were also evaluated. Tumours were staged according to the 8 th edition of the American Joint Committee on Cancer (AJCC) TNM staging system for CRCs after the clinicopathologic features and radiologic findings had been re-reviewed 13 . Demographic and clinical information, including age, sex, underlying disease, type of surgery, tumour site, date of diagnosis, postoperative treatment, last follow-up status, date of death or last follow-up, were collected from a patient medical record review.
Immunohistochemistry for p53, Rb1, p16, and mismatch repair proteins. Immunohistochemistry was performed on representative whole tissue sections using the avidin-biotin method. The primary antibodies used were against MLH1 (mouse monoclonal antibody clone G168-728 at a dilution of 1:300; Cell Marque, CA, USA), MSH2 (mouse monoclonal antibody clone FE11 at a dilution of 1:100; Calbiochem, CA, USA), MSH6 (mouse monoclonal antibody 44 at a dilution of 1:300; BD Transduction Laboratories, CA, USA), PMS2 (mouse monoclonal antibody clone A16-4 at a dilution of 1:125; BD Transduction Laboratories, CA, USA), p53 (mouse monoclonal antibody clone DO-7 at a dilution of 1:1,500; DAKO, Glostrup, Denmark), Rb1 (clone 3C8 at a dilution of 1:10,000; QED Bioscience, CA, USA), p16 (clone E6H4; prediluted; Ventana, AZ, USA), anti-Human papillomavirus (HPV, mouse monoclonal antibody clone K1H8 at a dilution of 1: 400; DAKO, Glostrup, Denmark). Anti-HPV is immunoreactive with paraffin sections of formalin-fixed tissues infected with HPV type 6,11,16,18,31,33,42, 51, 52, 56 and 58. An automated stainer (Ventana Medical Systems, AZ, USA) was used according to the manufacturer's protocol. Immunohistochemical expression of each MMR protein was considered intact if nuclear staining of neoplastic cells was detected with internal control positivity in the nonneoplastic crypt epithelium. Expression of p53 was considered aberrant if there was diffuse and strong nuclear staining (more than 2/3 of cells) or complete loss of expression. Rb1, p16, and HPV were classified as immunoreactive if nuclear staining (Rb1 and HPV) or cytoplasmic staining (p16) in all neoplastic cells was observed. A DNA library was prepared as described in our previous report using the S1 method 14 . Each library was constructed with sample-specific barcodes six-bp in size and quantified using Qubit Kit. Eight libraries were pooled for hybrid capture using the Agilent SureSelect XT custom kit (OP-AMCv3RNA bait, 1.2 Mb; Agilent Technologies, CA, USA). The enriched target concentration was measured by quantitative polymerase chain reaction (qPCR; Kapa Biosystem, Inc., MA, USA). DNS libraries that passed quality checks were sequenced using MiSeq. Sequenced reads were mapped to the human reference genome (NCBI build 37) using Burrows-Wheeler Aligner (version 0.5.9) with default options. PCR duplicates were removed using the Picard tool. Then, de-duplicated reads were realigned at known indel positions using GATK IndelRealigner, and base quality was recalibrated with GATK Table Recalibration 15 . Somatic mutations of single-nucleotide variants and short indels were called in tumour tissue with matched normal tissue using MuTect (1.1.7) and SomaticIndelocator in GATK [15][16][17] . Germline variants from somatic variant candidates were filtered out using the common dbsnp database (build 141; found in ≥ 1% of samples), the Korean Reference Genome database 18 and an in-house panel of normal variants 19,20 . Filtered somatic variants were annotated with Variant Effect Predictor (v79) and then converted to MAF files using vcf2maf (v1.612) 21 . False-positive variants were manually curated using Integrative Genomic Viewer 22 .

Statistical analysis.
Pearson's chi-square test or Fisher's exact test was applied to evaluate correlations between clinicopathologic variables and frequently mutated genes. Continuous variables were analysed using Student's t-test or the Mann-Whitney U-test. Overall survival curves were constructed using the Kaplan-Meier method and compared with the log-rank test. Cox proportional hazard models were employed to estimate the combined influence of clinicopathologic variables on survival. A p-value < 0.05 was considered statistically significant.
Ethics approval and consent to participate. Informed consent was obtained from all study participants. This study was approved by the institutional review boards at Asan Medical Center, Seoul, South Korea.
The study was conducted in accordance with the Declaration of Helsinki.

Results
Clinicopathologic features of colorectal NECs and ACs. A comparison analysis of the clinicopathologic features of colorectal NECs and ACs is summarized in Table 1. For colorectal NECs, the patients consisted of twenty men and ten women (male to female ratio: 2:1), with a mean age of 67 years (range: 44 to 83 years). One patient each had a history of Lynch syndrome and familial adenomatous polyposis. The most common primary tumour sites were the rectum (47%), right-sided colon (30%), and left-sided colon (23%). The median tumour size was 5.65 cm (range, 1.6-16 cm). Large-cell morphology was more commonly observed than smallcell morphology (60% vs. 40%). The median mitotic count per 2mm 2 was 62.5, ranging from 12 to 141/2mm 2 . The proliferation index assessed by Ki-67 was high, with a median percentage of 75% (range, 50%-95%  Fig. 3. Compared with the 100 AC group samples, the NEC samples frequently harboured BRAF mutations (23% vs. 6%; p = 0.0112), particularly p.V600E (71%). The frequencies of identified KRAS mutations were similar in both groups (53% (NEC) vs. 53% (AC); p = 1.0). Although the frequencies of TP53 and APC mutations in the NEC group were slightly higher than those in the AC group (43% vs. 35%; p = 0.0546, 37% vs. 31%; p = 0.6569, respectively), PIK3CA mutations were less frequently identified in the NEC group (10% vs. 15%; p = 0.7633). Overall, there was no statistically significant difference in the frequencies of mutations in TP53, APC, or other genes, including SMAD4, PIK3CA, PTEN, and FBXW7, which have been reported to be mutated in colorectal ACs by next-generation sequencing.  (Fig. 4). In addition, patients with NEC in the left colon showed an inferior OS than patients with NEC in the rectum or the right colon (p = 0.014). In multivariate analysis, distant metastasis at diagnosis (M category) was the only independent prognostic factor related to OS (Table 3). Regarding other potential molecular alterations  www.nature.com/scientificreports/ related to survival, there were no significant differences in OS rates between groups with mutated and wild-type BRAF, groups with mutated and wild-type RB1, or groups with mutated and wild-type TP53.
We also compared the OS rates of patients with colorectal NEC and those with 100 colorectal adenocarcinomas (Fig. 4). Among 100 patients with colorectal adenocarcinoma, 54 (54%) patients died of the disease. The median overall survival was 164.5 months, with 1-, 3-, and 5-year survival rates of 96%, 83%, 78%, respectively. Overall, the OS rate of patients with colorectal NEC was poorer than that of patients with colorectal adenocarcinoma (p < 0.001).

Discussion
Poorly differentiated colorectal NEC is a rare malignant neoplasm accounted for only 0.6% of all CRCs 1 . The clinicopathologic features and prognosis of colorectal NECs have been described in a few relatively large series. With a median OS of 11.5 months, the findings demonstrate that poorly differentiated colorectal NEC has an aggressive biologic behaviour with a poor clinical outcome, in line with prior studies. A retrospective study of 100 colorectal NECs analysed at M.D. Anderson Cancer Center reported a median age at diagnosis of 55 years (range 33-88 years), with 51% of the patients being men 23 . The majority of poorly differentiated NECs had a small-cell morphology (89%) rather than a large-cell morphology (8%), and metastatic disease was noted in 64 (64%) patients at diagnosis. In another retrospective analysis of data from the National Cancer Database (2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015) consisting of 1208 poorly differentiated colorectal NECs, the median age at diagnosis was 65 years, and 50% of patients were male 3 . A small-cell morphology was slightly more frequent than a large-cell morphology (54% vs. 46%). Lymphovascular invasion was observed in 25% of NEC cases, and approximately 55% of patients had advanced disease (III/IV) at diagnosis. In addition, Takaziwa et al. reported that patients with colorectal NECs were diagnosed at older age (median, 68 years) and with a male predominance (2:1) 7 . Among 25 colorectal NECs, a large-cell morphology was more commonly observed than was a small-cell type (64% vs. 36%). NEC tumours showed high rates of lymph node metastasis (82%) and distant metastasis (44%) and were at a more advanced stage (stage III/IV, 92%). In the present series, the patients were diagnosed at an older age (mean age: 67 years), with a male predominance (male to female ratio: 2:1), and among the 30 NEC tumours, a large-cell morphology was more frequently observed than a small-cell morphology (60% vs. 40%). Compared with typical ACs, NECs revealed multiple adverse prognostic pathologic factors, including frequent lymphatic invasion (97%), vascular invasion (77%), lymph node metastasis (80%), and distant metastasis at diagnosis (70%), which are associated with advanced TNM stage (stage III/IV, 83%). Interestingly, based on the data from the Japanese and current series, patients with colorectal NECs in East Asian countries tend to have more adverse clinicopathologic factors related to advanced TNM staging than do patients in Western countries. However, due to the rarity of this disease entity, discrepancy in the analysed clinicopathologic findings of colorectal NECs depending on racial/ geographic differences or observer variations is unclear.
Unlike colorectal ACs, surgery alone is rarely curative for poorly differentiated NECs, and the survival benefit from surgery is controversial in prior studies 24,25 . According to poorly differentiated colorectal NEC data from the National Cancer Database (2004-2015), the median OS for patients who underwent surgical resection was 10.5 months compared with 6.9 months for patients who did not undergo surgery (p < 0.001) 3 . In our series, the www.nature.com/scientificreports/ median OS was 11.5 months; the 3-year survival was 36.4%, and the 5-year survival was 32.7%. The patients in our series with surgically resected colorectal NECs had a better OS than those who did not undergo surgery in the data from the National Cancer Database (11.5 months vs. 6.9 months). Regardless, it is uncertain whether this correlation is due to selection bias or true therapeutic effects. Thus, further studies about the survival benefit of surgical resection in treating colorectal NECs are needed. Approximately 8% of CRCs harbour BRAF mutations, predominantly in exon 15, namely, p.V600E 26 . BRAF mutation is known to be prevalent in tumours of the proximal colon and tumours with poor differentiation (grade 3 or 4), mucinous histology, and high microsatellite instability 26 . However, the frequencies of BRAF mutations in poorly differentiated colorectal NECs are controversial (Table 4)   RB1 is a critical negative regulator of the Rb1/p16 pathway, which controls the G1 checkpoint of the cell cycle 29 . Inactivating mutations or homozygous deletions in RB1 have been reported in neuroendocrine neoplasms of the lung, gastrointestinal tract, and prostate 11,[30][31][32] . Furthermore, a comprehensive whole-genome sequencing study of 110 pulmonary small-cell carcinomas found near-universal biallelic inactivation of RB1 via mutations, deletions, and complex rearrangements 31 . Nonetheless, there are contradicting data on the frequency of molecular alterations in RB1 in colorectal NECs. Only one (3.7%) patient with heterozygous RB1 loss was reported in a previous molecular analysis of 27 colorectal MANECs and NECs by targeted next-generation sequencing 9 . In contrast, a recent molecular analysis of 24 colorectal NECs and MANECs revealed that fourteen (58%) tumours showed biallelic alterations in RB1 with next-generation sequencing using a 479-gene panel 11 . Similarly, in our series, fourteen (47%) pure NECs displayed genomic alterations in RB1, including both mutations and deep deletions, with the latter being more prevalent (57%) than the former (43%). The tumours with RB1 alterations also exhibited aberrant expression of Rb1 and p16 by immunohistochemistry. Genomic alterations in RB1 accompanied by loss of Rb1 and overexpression of p16, representing deregulation of the Rb1-p16 pathway, are commonly observed in colorectal NECs. Prior and our data support that deregulation of the Rb1-p16 pathway, as confirmed by immunohistochemistry and next-generation sequencing, plays a critical role in the histogenesis of colorectal NECs, as it does in small-cell lung cancer. Unlike a study result by Shamir et al., we failed to find an association between p16 positive colorectal NECs and HPV infection. The difference of HPV infection status in colorectal NEC tumours between two studies may be caused by selection bias, presence of NEC tumours coinfected with HPV, or others. Further studies would be needed for investigating the role of HPV infection in the histogenesis of a subset of colorectal NEC tumours.
The frequencies of other oncogenic driver genes commonly identified in CRC, such as TP53, KRAS, APC, PIK3CA, and PTEN, have been inconsistently reported in prior studies (Table 4) [7][8][9][10][11] . Two recent studies demonstrated relatively high mutation rates of TP53, KRAS, and PIK3CA, roughly similar to those of our molecular data 9,11 . However, two other studies showed relatively lower mutation rates for these genes 8,10 . These studies did not compare these mutated genes of colorectal NECs with those of colorectal ACs, and it remains unclear whether there are significant differences in these mutated genes in colorectal NECs. Thus, the present study compared these mutations of colorectal NECs with those of colorectal ACs after tumour site matching and detected no significant differences in the frequencies of TP53, KRAS, APC, PIK3CA, PTEN, GNAS, and SMAD4 mutations between 30 NEC tumours and 100 AC tumours. The almost parallel frequencies and distributions of the main oncogenic gene mutations, except for BRAF, identified between the NEC and AC groups indicate that colorectal NECs are genetically similar to colorectal ACs, suggesting that colorectal NECs arise from the same origin as colorectal ACs, with intestinal glands likely serving as the primary origin.
In summary, our study demonstrates that colorectal NECs display an aggressive biologic behaviour with multiple adverse clinicopathologic prognostic factors and a poor clinical outcome. In molecular analysis, BRAF mutations, predominantly p.V600E, were more frequently identified in poorly differentiated NECs than in conventional ACs. Additionally, almost exclusive deregulation of the Rb1/p16 pathway was revealed by immunohistochemistry and next-generation sequencing. Frequencies and distributions of the main mutated oncogenic driver genes, except BRAF, and the microsatellite instability status were similar in NECs and ACs. These findings www.nature.com/scientificreports/ provide beneficial information for the use of potential therapeutics such as BRAF inhibitors and a better understanding of the histogenesis of this tumour.