Serum tumour markers combined with clinicopathological characteristics for predicting MMR and KRAS status in 2279 Chinese colorectal cancer patients: a retrospective analysis


 Background

Although serum tumour markers (STMs), clinicopathological characteristics and the status of KRAS and MMR play an important role in optimizing the treatment and improving the prognosis of colorectal cancer, their interrelationships remain largely unknown.
Methods

A retrospective analysis of 2279 patients who underwent KRAS or MMR status testing and STM measurements prior to treatment over the past four years was conducted. Univariate and multivariate logistic regression were performed to identify independent predictive factors of KRAS and MMR status. The area under receiver operating characteristic (ROC) curve (AUC) was used to evaluate the predictive value of individual and combined factors.
Results

Of the 784 patients tested for KRAS and 2279 patients tested for MMR status, KRAS mutations and dMMR were identified in 276 patients (35.20%) and 177 patients (7.77%), respectively. Logistic regression analysis demonstrated that right colon, well and moderate differentiation and negative CA19-9 were independent predictors for KRAS mutations. The ROC curve yielded an AUC of 0.609 for the combination of the three factors. Age < 65 was an independent predictive factor for dMMR, along with tumour size > 4.6 cm, right colon, poor differentiation, harvested lymph nodes ≥ 22, no lymph node metastasis, no perineural invasion, negative CEA and positive CA72-4. When the nine criteria were used together, the AUC was 0.849.
Conclusion

Both STMs and clinicopathological characteristics were found to be significantly associated with the status of KRAS and MMR. The combination of these two factors possessed a strong predictive power for targeted genes among CRC patients.


Introduction
Colorectal cancer (CRC) is the third most common type of cancer and the second leading cause of cancer-related death worldwide [1]. CRC imposes a substantial burden on the healthcare system, with the direct costs of CRC accounting for close to 10% and 12% of all direct cancer-related costs across the European Union [2] and the United States [3], respectively. It has been estimated that more than 20% of patients present with metastatic CRC, and approximately half of patients with localized CRC will develop metastases [4]. In the majority of mCRCs, tumour lesions tend to be unresectable, and chemotherapy is recommended to prolong survival and improve symptoms. Fluoropyrimidine-based chemotherapy regimens and monoclonal antibodies directed against epidermal growth factor receptor are approved for rst-line treatment of the disease. Molecular testing for KRAS and mismatch repair (MMR) status are mandatory to optimize the choice and sequencing of therapy [5].
Kirsten rat sarcoma viral oncogene (KRAS) is located downstream of EGFR signals, and KRAS mutation leads to its constitutive activation [6], which makes advanced colon cancer less responsive to anti-EGFR monoclonal antibodies such as cetuximab and panitumumab [7,8]. Mismatch repair (MMR) proteins are responsible for length alterations in microsatellites as they correct strand alignment and base matching errors during DNA replication [9]. CRC patients with mismatch repair de ciency (dMMR) are not only highly likely to have a better prognosis [10] and higher incidence of Lynch syndrome [11] but are less likely to bene t from 5-FU-based chemotherapy [12,13] and immune checkpoint blockade [14]. However, translating genetic testing into routine clinical practice is frequently hindered by many barrier factors, such as the high cost of testing, the lack of preoperative tumour tissue and the need for specialized clinical laboratories [15][16][17]. These di culties are particularly obvious in developing countries, especially in county-level hospitals. Therefore, there is an urgent need to develop a convenient, noninvasive and cost-effective modality to identify appropriate candidates for genetic testing.
Previous studies have demonstrated that serum tumour markers (STMs) and clinicopathological characteristics are both important prognostic factors as well as indicators of the therapeutic effect and recurrence risk in patients with CRC [18][19][20], while their association with KRAS and MMR status is largely unknown. In this study, we explored the predictive value of STMs in combination with clinicopathological indicators for KRAS and MMR status across East Asian CRC patients.

Methods
Study design and patient cohort From January 6, 2016, through December 10, 2019, a total of 5457 patients were diagnosed with CRC in our centre. All patients were subjected to thorough history taking, and their information was collected from "Biological big data platform for individualized diagnosis and treatment of gastrointestinal cancer" (national software copyright 2019SR1267841). This study protocol was approved by the ethics committee of our college.
A ow diagram for the screening of eligible CRC participants is presented in Figure 1. A total of 2521 patients with MMR or KRAS testing were identi ed. A total of 242 patients with the following conditions were excluded from the study: (1) 211 patients underwent neo-chemoradiotherapy before KRAS and MMR status detection; and (2) 31 patients did not have data for STMs. Tumour stage was classi ed according to the 8th edition of the American Joint Committee on Cancer Staging System.

KRAS mutation analysis
The primers for the ampli cation and Sanger dideoxy chain termination sequencing of KRAS gene exon 2 were forward 5′-GTCCTGCACCAGTAATATGC-3′ and reverse 5′-ATGTTCTAATATAGTCACATTTTC-3′ for exons 3 and 4. Polymerase chain reaction (PCR) was performed using 100 ng of genomic DNA as a template. Each mixture contained 10 pmol of each primer. The reactions were performed in a total volume of 31.5 μL. The ampli cation reaction was as follows : an initial denaturing cycle of 95 °C for 5  min; 45 cycles of 94 °C for 25 s, 58 °C for 25 s, 72 °C for 25 s; and a nal extension cycle at 72 °C for 10 min. The PCR products were then puri ed and subjected to direct sequencing using an automatic sequencer (ABI-3730 DNA Sequencer; Life Technologies, CA). Tumours with any KRAS mutations were classi ed as mutant KRAS, whereas the rest were classi ed as wild-type KRAS. Representative histological images of two patients with KRAS mutant or wild-type CRC are shown in Figure 2.
Immunohistochemical analysis of MMR status Immunohistochemical staining was performed by the streptavidin-biotin-peroxidase detection method. First, CRC tissues xed with 4% formaldehyde and embedded with para n were cut into 5 µm thick slices that were xed onto glass slides. After rehydration with ethanol and microwave antigen retrieval, tissue sections were labelled with anti-MSH2 antibody (clone G219, 1:100, Cellmark, Rocklin, CA, USA), anti-MSH6 antibody (clone 44, prediluted; Ventana Medical Systems), anti-MLH1 antibody (clone M1, prediluted; Ventana Medical Systems) or anti-PMS2 antibody (clone mrq-28, 1:200; Cellmark) overnight at 4 °C. After washing with PBS, slides were incubated with the speci c HRP-conjugate antibody at 37 °C for 10 min, cleaned with cold PBS and treated with peroxidase-conjugated biotin streptavidin complex for 10 min. Finally, staining was performed with DAB and counterstaining was performed with Mayer's haematoxylin. Binary interpretation was used to determine whether MMR was de cient or pro cient as follows: tumours displaying loss of expression of one or more MMR proteins were considered to be dMMR, whereas tumours with intact MMR proteins were classi ed as pMMR.

Statistical analysis
Statistical analysis was performed using SPSS 23.0 (SPSS Inc., Chicago, IL, USA). Data are presented as numbers and percentages for categorical variables, and continuous data are expressed as the mean ± standard deviation, unless otherwise speci ed. Patient characteristics were compared using t tests for continuous variables and X 2 or Fisher exact tests for categorical variables. All candidate predictors with a P < 0.05 in univariate analysis were included in a multivariate logistic regression model. The discrimination ability of individual and combined factors was measured by the area under the ROC (receiver operating characteristic) curve (AUC). A value of P < 0.05 was considered signi cant.

Patient clinical characteristics
Among the 2279 recruited CRC patients, the number of participants tested for KRAS and MMR was 784 and 2279, respectively. The characteristics are summarized based on whether the patients were tested for KRAS or MMR (Table 1). Of the 784 patients tested for KRAS and 2279 patients tested for MMR status, KRAS mutations and dMMR were identi ed in 276 patients (35.20%) and 177 patients (7.77%), respectively.
MMR status is an important factor when deciding whether to use adjuvant chemotherapy for patients with stage II CRC [5] and is a signi cant prognostic indicator in stage III CRC patients with recurrence after adjuvant chemotherapy [23]. Therefore, we further analysed whether dMMR was associated with clinicopathological features and STMs in stage II/III CRC ( Table 2). The results were similar to those for the whole CRC population, except for T stage, which was not associated with dMMR in stage II/III CRC.

Predictive value of STMs in combination with clinicopathological features for KRAS mutation
For the whole CRC population, univariate logistic regression analysis demonstrated that histology type, tumour location, degree of differentiation, and CEA and CA 19-9 levels were signi cantly associated with KRAS mutations (Table 3). When these predictive factors were subsequently assessed in the multivariate logistic regression, all except for non-adenocarcinoma and CEA remained highly significant. Therefore, right colon was found to be an independent predictor of KRAS mutations (OR, 1.550; P = 0.012), along with well and moderate differentiation (OR, 2.203; P = 0.001) and negative CA19-9 (OR, 1.600; P = 0.022). The predictive potential of these factors using ROC curves is shown in Figure 2A. When the three indexes were used together, the AUC was 0.609.
For stage II/III CRC, all potential predictors were consistent with those for the whole CRC population, except for T stage, which was not associated with dMMR. In the multivariate logistic regression, nonadenocarcinoma (P = 0.585), lymphovascular invasion (P = 0.354) and CA72-4 (P = 0.058) were found to be unrelated to dMMR, and the remaining eight indicators were identi ed as independent predictors for dMMR. When the eight criteria were used together, the AUC was 0.849 ( Figure 2D).
A summary of the AUCs of the individual and combined assessments used to predict mutation status is presented in Table 7.

Discussion
CRC is the third most common cancer in men and the second most common cancer in women worldwide, accounting for approximately 10% of all cancer-related deaths. [24] To minimize the side effects of current treatments and achieve better results, tremendous progress has been achieved in targeted therapy for CRC over recent decades. The status of KRAS and MMR was reported to be signi cantly correlated with the clinical outcomes of target therapy [7,8,[12][13][14]. For example, KRAS mutations make CRC less responsive to anti-EGFR monoclonal antibodies [7,8], and dMMR makes CRC less likely to bene t from 5-FU-based chemotherapy [12,13] and immune checkpoint blockade [14]. However, the rate of KRAS and MMR detection was far below expected, mainly due to the following aspects: 1) a signi cant part of the population in developing counties cannot afford the high cost of gene testing; 2) for unresectable mCRC, endoscopic biopsy and cytology of peritoneal washing may not provide high-quality samples or su cient tumour cells for detection of KRAS mutation and MMR status after their initial use for histologic detection to rst con rm the cancer type [15,16]; 3) as a result of signi cantly uneven distribution of medical resources in China, the quali ed clinical laboratory and professional team required for gene testing are not available in county-level hospitals; and 4) the values of KRAS and MMR status are poorly understood at basic-level hospitals or in underdeveloped areas. In contrast, STMs and histopathological characteristics can be quickly and accurately detected in county-level or province-level hospitals at a low cost.
In this study, we explored the interrelationships among STMs, histopathological characteristics, and MMR and KRAS status using data from 2279 participants. Of the 784 patients tested for KRAS and 2279 patients tested for MMR status, KRAS mutations and dMMR were identi ed in 276 patients (35.20%) and 177 patients (7.77%), respectively. The discriminative ability of clinicopathological characteristics in combination with STMs was 0.609 for KRAS mutations and 0.849 for dMMR in the whole population. In addition, the combination of STMs and clinicopathological characteristics yielded an AUC of 0.622 for KRAS mutations and 0.849 for dMMR among TNM(II/III) participants.
Previous studies on the correlation between clinicopathological characteristics and KRAS mutations are controversial. Gao et al [18]. reported that no signi cant difference between KRAS mutations and tumour locations was observed, whereas Wilson et al [25]. and Julien et al [26]. supported that right-side CRC has a higher KRAS mutation rate. In our study, right colon (OR, 1.550; P = 0.012) and well and moderately differentiated tumours (OR, 2.203; P = 0.001) were independent predictive factors of KRAS mutations. STMs were reported to not be associated with KRAS mutations in several studies [21,27,28]. However, negative CA199 was found to be signi cantly correlated with KRAS mutations in our study. Furthermore, the AUC was 0.609 when clinicopathological characteristics were combined with CA 19 − 9. The combination exceeds the discriminative ability of individual relevant factors, including right colon (AUC = 0.547), well and moderate differentiation (AUC = 0.545) and CA 19 − 9 (AUC = 0.544).
Previous studies have shown that MMR status was signi cantly correlated with clinicopathological characteristics of CRC[29-31], including tumour location, degree of differentiation, perineural invasion, and number of harvested lymph nodes, which are consistent with our results. In our study, younger age (OR, 1.923; P = 0.006), larger tumour (OR, 2.646; P < 0.001), and fewer positive lymph nodes (OR, 2.924; P < 0.001) were also independent predictive factors for dMMR. However, the existing evidence of the correlation between MMR status and STMs is controversial. Fan et al [31]. reported that dMMR was not associated with CEA, CA72-4, CA 242 and CA 19 − 9, but Schiemann et al [32]. reported that patients with high microsatellite instability had lower preoperative CEA serum levels than those with microsatellite stability. In our study, dMMR was signi cantly associated with negative CEA and positive CA72-4. When STMs were combined with clinicopathological characteristics, the AUC increased from 0.837 to 0.849. This study's limitations deserve commentary. First, this was a nonrandomized retrospective analysis from a single centre, and as such, there were potential biases for comparison, such as patient inclusion and sample selection biases. Second, there is a lack of a validation group to further validate our results. Third, we did not evaluate the treatment response or perform a survival analysis according to clinical characteristics or serum tumour marker levels. However, our results demonstrated that clinicopathological characteristics in combination with STMs possessed a strong predictive power for KRAS and MMR status among CRC patients.

Conclusion
In conclusion, this is the largest retrospective study to investigate the interrelationship among KRAS mutations and dMMR, STMs and clinicopathological characteristics. KRAS mutations was signi cantly correlated with right colon, well and moderate differentiation and negative CA19-9. DMMR was signi cantly associated with younger age, larger tumours, right colon, poor differentiation, more harvested lymph nodes, fewer positive lymph nodes, no perineural invasion, negative CEA and positive CA72-4. The discriminative ability of clinicopathological characteristics combined with STMs reached 0.609 for KRAS mutations and 0.849 for dMMR. Radiomics signatures based on deep learning features have been used in previous studies to predict KRAS and MMR status and have resulted in remarkable accomplishments [31,[33][34][35]. Therefore, for those CRC patients who could not undergo genetic testing, our ndings will hopefully be integrated with radiomics or other markers to achieve a stronger discrimination ability of KRAS and MMR status.

Availability of data and materials
The data used or analysed during this study are included in this article and available from the corresponding author upon reasonable request.

Ethics approval and consent to participate
Ethics approval was obtained from the Huazhong University of Science and Technology, and written informed consent was obtained from study participants.

Consent for publication
Not applicable.  Values presented are n (%) unless otherwise noted. Values presented are n (%) unless otherwise noted. Abbreviations: OR, odds ratio; 95% CI, 95% con dence interval; * Items were included in the multivariate analysis only when P value <0.05 in univariate analysis. Abbreviations: OR, odds ratio; 95% CI, 95% con dence interval; * Items were included in the multivariate analysis only when P value <0.05 in univariate analysis. Abbreviations: OR, odds ratio; 95% CI, 95% con dence interval; * Items were included in the multivariate analysis only when P value <0.05 in univariate analysis. Abbreviations: OR, odds ratio; 95% CI, 95% con dence interval; * Items were included in the multivariate analysis only when P value <0.05 in univariate analysis.