Prognostic implications of mucinous differentiation in metastatic colorectal carcinoma may be explained by distinct molecular and clinico-pathological characteristics

Abstract

Background:

Mucinous histology accounts for 5-20% of colorectal cancer (CRC), but remains poorly characterized. This study characterized baseline characteristics, mutational profile, and clinical outcomes of patients diagnosed with mucinous CRC.

Methods:

We identified 1877 metastatic CRC patients with available histology and molecular profiling and summarized baseline clinical and pathologic characteristics and overall survival (OS) by histology. Separate cohorts with Consensus Molecular Subtype (CMS) and CpG Island Methylator information are also summarized.

Results:

Mucinous histology occurred in 277/1877 (14.8%) patients and was associated with increased prevalence of microsatellite instability (p<0.001) and right-sided primary (p<0.001). Increased frequency of CMS1 and lower rates of CMS2 were identified with mucinous compared to non-mucinous adenocarcinoma (p<0.0001). Mutations in SMAD4 (p<0.001), GNAS (p<0.001), ERBB2 (p=0.02), BRAF (p<0.001) and KRAS (p < 0.001) occurred at higher frequencies in mucinous CRC, while TP53 (p<0.001), APC (p<0.001) and NRAS mutations (p=0.03) were less common. Univariate (HR=1.38, 95% CI 1.17-1.63, p<0.001) and multivariate analysis (HR=1.36, 95% CI: 1.12-1.64, p=0.002) demonstrated that mucinous histology was associated with worse OS. Features associated with mucinous histology were independent predictors for shorter OS, including BRAF (HR=1.74, 95% CI: 1.35, 2.25, p<0.001) and KRAS (HR=1.42, 95% CI: 1.22, 1.65, p<0.001) mutations, right-sided location (HR = 1.20, 95% CI: 1.04, 1.39, p=0.01) and synchronous metastases (HR=2.92, 95% CI: 2.49, 3.42, p<0.001).

Conclusion:

Compared with non-mucinous adenocarcinoma, mucinous histology is associated with worse prognosis, even when controlling for known prognostic features. This unique biological behavior should be considered in the management and prognostic assessment of patients.

Micro Abstract:

Mucinous colorectal cancer (CRC) is a unique histology that is poorly defined. We aimed to better characterize this subtype of CRC and identified an increased frequency of mutations in GNAS, ERBB2, BRAF, KRAS and SMAD4 with mucinous histology, in addition to increased CMS1 and decreased CMS2 prevalence. Adverse clinical features included microsatellite instability, and synchronous metastatic disease at presentation. Even after controlling for clinical and mutation differences, mucinous histology was associated with worse overall survival.

Introduction:

Colorectal carcinomas (CRCs) are a heterogeneous group of neoplasms with varying genetic abnormalities and histology. The most common CRC histologic subtypes include adenocarcinoma, mucinous, signet ring cell, medullary, and micropapillary, however there are other rare histopathological variants [1]. The mucinous subtype accounts for 5% to 20% of all CRCs [2], and has been defined by the presence of abundant extracellular mucous which contributes to more than 50% of tumor volume [3].

The carcinogenesis of mucinous carcinoma is not well understood; however, several unique characteristics suggest a pathogenesis distinct from non-mucinous colorectal adenocarcinoma [4]. Mucinous CRCs tend to over-express MUC2 and MUC5AC genes, which are oligomeric mucus/gel-forming genes [5]. They are also more likely to exhibit microsatellite instability (MSI) than non-mucinous adenocarcinomas. In addition, mucinous CRCs are also shown to be associated with increased prevalence of CIMP-H phenotype [6]. Clinically, mucinous CRC has been shown to be associated with advanced tumor stage at the time of presentation and inferior outcomes compared with non-mucinous adenocarcinomas [7]. In early stage disease, rates of angioinvasion are lower in the mucinous histology, however it is associated with poor differentiation and an increased risk of local recurrence [7].

From a mutational perspective, mutations in PIK3CA [8], SMAD4 [9] and BRAF have previously been reported to be more common in mucinous CRCs [6,10,11], whereas there have been mixed reports regarding the mutational frequency of KRAS [6,12]. TP53 and APC are less frequently mutated in mucinous carcinomas [6], but their mutations have been shown to have a clinical impact on prognosis when present [13].

Despite attempts to characterize the molecular features of mucinous CRC, the inconsistency of current data and small cohort sizes have made it difficult to obtain concrete information. Since determination of molecular alterations can assist in the development of targeted therapies [10], this study aimed to provide a clearer understanding of the relationship between genomic aberrations and mucinous histology in CRC, as well as to assess their possible impact on clinical outcome.

Materials and Methods:

Patients

After obtaining approval from the institutional review board at The University of Texas MD Anderson Cancer Center (MDACC), we reviewed clinical and pathology records of all patients with biopsy proven mCRC who had a Clinical Laboratory Improvement Amendments (CLIA) 46 gene next generation sequencing panel performed between January 1, 2012 and September 1, 2016. Patients with non-mucinous, non-adenocarcinoma histologies were excluded and the remaining 1825 patients were grouped by histology (mucinous vs non-mucinous adenocarcinoma). The inclusion of patients in the mucinous category required the documentation of >50% extracellular mucin on pathology reports, while tumors with a mixed signet-ring cell component were excluded from the analysis. Consensus molecular subtypes (CMS) were evaluated in a separate cohort of 196 patients from MDACC and 412 patients from the TCGA with previously reported expression data [14,15]. A previously published classifier was used to determine CMS subtype [15].

A third cohort of 459 mCRC patients was reviewed to provide information about CpG island methylation phenotype status. These patients are from the ATTACC umbrella protocol ( NCT01196130) at MDACC. PCR-based pyrosequencing of DNA was performed to examine the methylation status of multiple CpG island loci for each of the tested genes (hMLH1, Mint1, Mint2, Mint31, p14, and p16) and tumors were classified as CIMP-H tumors if ≥40% of CpG Island loci were methylated [16].

Genetic Analysis

Genomic sequencing was performed in the CLIA environment at MDACC and utilized archival formalin-fixed paraffin embedded (FFPE) samples from either surgical resection or tissue biopsies. Only samples with tumor content >20% after macro dissection were used for sequencing and matched normal were not sequenced. The assay has been previously described and uses the Ion Torrent AmpliSeq Cancer Panel Primers. Adequately covered amplicons were defined as those with a coverage depth of >250 reads. The lower limit of detection is estimated at 5-10% [17].

Statistical Analysis

Baseline comparisons of demographic, clinical and mutational characteristics were performed across histology groups. These were performed using Chi-squared, Fisher’s exact or Mann–Whitney tests. The multiple testing adjustment method by Benjamini and Hochberg (1995) was used to control the false discovery rate [18]. Overall survival (OS) was defined as the time interval between dates of diagnosis with metastatic disease and death, and was censored at the last follow-up date for patients who were alive. Survival curves were estimated using the Kaplan-Meier method [19]. Univariate Cox proportional hazard (PH) models were used to evaluate the risk factors associated with OS [20]. A multivariate Cox PH model was obtained by first including the variables with p-value < 0.20 in the univariate analysis. A backward elimination was then performed until all remaining factors had a p-value of less than 0.05.

Multivariate Cox PH models were constructed to assess the relationships between patient characteristics, mutations, and OS. A further subgroup analysis was performed only among patients with synchronous or metachronous metastatic disease. The same survival analysis methods were used to assess the relationships between patient characteristics, mutations, and OS within this subgroup. All statistical analyses were performed in R 3.3.1.

Results:

Association of Mucinous CRC with Clinical and Molecular Characteristics

Table 1 summarizes demographic and clinical characteristics by histology. Of 1877 CRCs, 277 (14.8%) were characterized as mucinous based on histological diagnosis. A total of 391 non-mucinous and 47 mucinous CRC cases had confident CMS calls and confirmed histology. Mucinous histology in CRC patients was associated with MSI-H (p<0.001) and right sided location (p<0.001). Furthermore, comparison of CMS subtypes identified a significantly larger proportion of CMS1,CMS3 and CMS4 with mucinous CRC, whereas CMS2 was more common with non-mucinous adenocarcinoma (p<0.0001; Figure 1). We additionally surveyed the prevalence of CIMP in a separate cohort of 457 patients with known CIMP status, and identified higher prevalence of CIMP-H tumors in mucinous CRC as compared to non-mucinous histology (p=0.003). Compounding this, we also confirmed association of CIMP-H tumors with mucinous histology (95% CI: 1.71-3.94, I²=81.90%) in a pooled analysis of multiple studies [21-32].

Table 1.

Baseline Characteristics by Histology

Variable	Non Mucinous Adenocarcinoma	Mucinous	P-value
	N=1548 (%)	N=277 (%)
Age at dx
Median	55.18	55.69	0.39
Range	19.07-91.77	19.22-82.25
Gender
Female	680 (43.9)	110 (39.7)	0.19
Male	868 (56.1)	167 (60.3)
Initial Stage
1-3 or local	543 (35.1)	103 (37.2)	0.50
4	1005 (64.9)	174 (62.8)
Site
Right-sided	537 (34.7)	143 (51.6)	<0.001
Left-sided	1011 (65.3)	134 (48.4)
MSI-H
Yes	35 (2.3)	18 (6.5)	<0.001
No	1224 (79.1)	206 (74.4)
Missing	289 (18.7)	53 (19.1)
No. of mutations
Median	2	2	0.36
Range	0-20	0-23
Consensus Molecular Subtype Cohort (N=608)
CMS subtypes
Number	391	47
CMS1	51 (13.0)	16 (34.0)	<0.0001
CMS2	195 (49.9)	3 (6.4)
CMS3	48 (12.3)	14 (29.8)
CMS4	97 (24.8)	14 (29.8)
CpG Island Methylator Phenotype (CIMP) Cohort (n=457)
Number	358	91	0.003
CIMP-H	70 (20)	31 (34)

P-value was calculated by excluding missing values.

Figure 1.

The distribution of Consensus Molecular Subtypes among patients with mucinous CRC compared to patients with non-mucinous adenocarcinoma.

Association of Mucinous CRC with Mutations

Analysis of the mutational profiles of patients with non-mucinous adenocarcinoma and mucinous CRC yielded strong associations between histology and the frequency of particular mutations as shown in Table 2 and Figure 2. The higher frequency of GNAS (p < 0.001), ERBB2 (p = 0.02), KRAS (p < 0.001), BRAF (p < 0.001), and SMAD4 mutations (p < 0.001) in mucinous CRC was a particularly notable finding. Conversely, the prevalence of TP53 (p < 0.001), APC (p < 0.001) and NRAS (p=0.03) mutations was significantly lower among patients with mucinous histology. Association of these mutations with each other in co-occurrence is further expanded in Figure 3.

Table 2.

Mutational Profile by Histology

Mutations	Non Mucinous Adenocarcinoma	Mucinous	P value	P value adjusted for multiple comparison
	N=1548 (%)	N=277 (%)
TP53	1074 (69.4)	117 (42.2)	<0.001	<0.001
APC	735 (47.5)	101 (36.5)	<0.001	0.002
FBXW7	119 (7.7)	21 (7.6)	0.95	1.00
SMAD4	161 (10.4)	65 (23.5)	<0.001	<0.001
PIK3CA	245 (15.8)	39 (14.1)	0.46	0.53
ATM	32 (2.1)	8 (2.9)	0.39	0.53
CTNNB1	23 (1.5)	6 (2.2)	0.43	0.53
ERBB2	13 (0.8)	7 (2.5)	0.02	0.049
ERBB4	17 (1.1)	5 (1.8)	0.36	0.53
GNAS	9 (0.6)	19 (6.9)	<0.001	<0.001
KDR	16 (1.0)	3 (1.1)	1.00	1.00
PTEN	35 (2.3)	10 (3.6)	0.18	0.30
KRAS	737 (47.6)	165 (59.6)	<0.001	<0.001
NRAS	72 (4.7)	5 (1.8)	0.03	0.06
BRAF	92 (5.9)	42 (15.2)	<0.001	<0.001

Figure 2.

Prevalence of mutations according to histology. *** stands for p-value <0.001 and ** stands for p value <0.02

Figure 3.

Association of co-occurring mutations. The thickness of the connecting lines indicates the frequency with which the two mutations co-occur in mucinous CRC (A) and non-mucinous adenocarcinoma (B). Abbreviations = CT1, CTNNB1; ER2, ERBB2; ER4, ERBB4; RAS, KRAS; SM, SMAD4; B600, BRAFV600; Bm, BRAF-mutated; Bn, BRAF-nonV600.

Association of Mucinous CRC with Clinical Outcome

At the end of follow up, 1026 patients (56%) died of CRC or other causes. The median follow-up was 4.3 years, while the 5-year OS rate was 43% (95% CI: 41% - 46%) (Figure 4). Mucinous CRC was found to confer a worse prognosis (HR = 1.36, 95% CI: 1.12, 1.64; p<0.001) compared to non-mucinous adenocarcinoma (Figure 4).

Figure 4.

Kaplan-Meier estimates for overall survival of mucinous CRC patients as compared with non-mucinous adenocarcinoma patients.

Univariate analysis also identified microsatellite instability, right-sided lesions, synchronous metastatic presentation, and mutations in KRAS, BRAF, FBXW7, NRAS and PIK3CA as predictors of poor prognosis. Additionally, GNAS (p=0.19), ERBB2 (p=0.10) and SMAD4 (p=0.004) mutations were initially found to be poor prognosticators by univariate analysis but not multivariate analysis. In the multivariate model controlling for each of these independent poor prognostic markers, mucinous histology was still associated with a worse OS (HR 1.36, 95% CI 1.12-1.64, p=0.002). KRAS (HR = 1.42, 95% CI: 1.22, 1.65, p < 0.001) and BRAF (HR = 1.74, 95% CI: 1.35, 2.25, p < 0.001) mutations, right-sided lesions (HR = 1.20, 95% CI: 1.04, 1.39, p = 0.01) and synchronous metastatic presentation (HR = 2.92, 95% CI: 2.49, 3.42, p < 0.001) were also shown to be prognostic factors for shorter OS in this model (Table 3).

Table 3.

Univariate and Multivariate Cox Proportional Hazards Regression Models to evaluate the significance of prognostic variables in overall survival

Characteristic/Mutation	Levels	Death	Total	HR	95% CI	P-value
Univariate
Age at dx		1026	1825	1.00	(0.99, 1.00)	0.48
Gender	Female vs. Male	1026	1825	1.10	(0.97, 1.24)	0.15
Initial Stage	4 vs. 1-3 or local	1026	1825	2.70	(2.35, 3.10)	<0.001
Site	Left-sided vs. Right-sided	1026	1825	0.72	(0.63, 0.81)	<0.001
MSI-H	Yes vs. No	804	1483	1.52	(1.04, 2.21)	0.03
Number of Mutations		1026	1825	1.02	(0.99, 1.05)	0.13
Histology	mucinous vs. non mucinous adenocarcinoma	1026	1825	1.38	(1.17, 1.63)	<0.001
Mutations
TP53	Yes vs. No	1026	1825	0.88	(0.78, 1.00)	0.06
APC	Yes vs. No	1026	1825	0.89	(0.79, 1.01)	0.07
FBXW7	Yes vs. No	1026	1825	1.28	(1.02, 1.60)	0.03
SMAD4	Yes vs. No	1026	1825	1.31	(1.09, 1.57)	0.004
PIK3CA	Yes vs. No	1026	1825	1.27	(1.08, 1.50)	0.004
ATM	Yes vs. No	1026	1825	0.92	(0.60, 1.43)	0.72
CTNNB1	Yes vs. No	1026	1825	0.79	(0.44, 1.44)	0.44
ERBB2	Yes vs. No	1026	1825	1.59	(0.92, 2.75)	0.10
ERBB4	Yes vs. No	1026	1825	1.10	(0.66, 1.84)	0.71
GNAS	Yes vs. No	1026	1825	1.39	(0.85, 2.28)	0.19
KDR	Yes vs. No	1026	1825	0.88	(0.49, 1.60)	0.68
PTEN	Yes vs. No	1026	1825	1.29	(0.88, 1.89)	0.20
KRAS	Yes vs. No	1026	1825	1.27	(1.12, 1.44)	<0.001
NRAS	Yes vs. No	1026	1825	1.31	(1.00, 1.73)	0.05
BRAF	Yes vs. No	1026	1825	1.91	(1.55, 2.35)	<0.001
Multivariate
Histology	mucinous vs. non mucinous adenocarcinoma	804	1483	1.36	(1.12, 1.64)	0.002
Initial Stage	4 vs. 1-3 or local			2.92	(2.49, 3.42)	<0.001
Site	Left-sided vs. Right-sided			0.83	(0.72, 0.96)	0.01
KRAS	Yes vs. No			1.42	(1.22, 1.65)	<0.001
BRAF	Yes vs. No			1.74	(1.35, 2.25)	<0.001

HR, Hazard Ratio; CI, Confidence Interval. P-value was calculated by excluding missing values. Reference Group for hazard ratio calculations is the category listed first in the second column

Given a hazard ratio of 2.92 in multivariate analysis, metastatic presentation was indicated to have a particularly strong prognostic effect. Therefore, a dedicated subgroup analysis to assess the clinical outcomes of 1179 patients who presented with synchronous metastatic disease was performed and a separate subgroup analysis was performed on those patients with metachronous metastatic disease. Supplemental Table 1 summarizes the prognostic variables significantly associated with a worse prognosis in the synchronous metastatic subgroup. Once again, mucinous histology was associated with a worse OS in univariate (HR 1.47, 95% CI 1.20-1.80; p<0.001) and multivariate (HR 1.28, 95% CI 1.04-1.58; p=0.02) analysis.

However, in patients who presented with early stage CRC (stages 1-3) we found inconsistent findings in comparison with the cohort as a whole, such that mucinous histology failed to reach statistical significance as a prognostic factor in this subgroup when controlling for other variables in the multivariate analysis (Supplemental Table 2). In addition, while multivariate analysis of the full cohort does not delineate MSI-H as a significant prognosticator, a subgroup analysis correlated MSI-H with poorer OS in patients with early-stage (stages 1-3) CRC at presentation (p<0.001).

Discussion:

Despite the recognition of mucinous histology as a well-established clinical entity among CRCs, it remains poorly characterized and the molecular underpinnings and potential vulnerabilities of this histology remain unknown. We report an overall incidence of 14.8% in a large cohort of almost 1900 patients. These results concur with previous estimates of 15% in a US-based report [33] and 10-20% in other western populations [34,35].

We found significant differences in baseline and molecular characteristics that differ between mucinous and non-mucinous adenocarcinoma histology. In addition to differences in common MAP-kinase alterations, our analysis identified significantly higher frequencies of GNAS, SMAD4 and ERBB2 mutations among the mucinous subgroup. The roles of ERBB2 and GNAS are vaguely understood in mucinous CRC — but ERBB2 has the potential to bind with MUC4 as a ligand [36] while GNAS and SMAD4 mutations have been noted in other neoplasms of mucinous origin [37,38], suggesting their roles in the pathogenesis of mucinous CRC may be important. Previous studies have reported frequent GNAS mutations in the settings of intraductal mucinous neoplasms of the pancreas and mucinous appendiceal carcinomas, while also implicating these mutations in the overexpression of gel-like mucins [37,38]. Since these mutations are characteristic of the mucinous subtype, our results may be indicating a potential function of mutant GNAS in the pathogenesis of mucinous CRC. Similarly, SMAD4 and ERBB2 mutations may also be associated with mucinous histology [9,39], however, their contribution to its clinical correlates is vaguely understood.

High rates of CMS1 were noted among mucinous CRCs in our study, which supports a common biology characterized by microsatellite instability, right-sided location, and CIMP. We also noted an association with CMS3 and CMS4, with a substantially lower rate of CMS2 than non-mucinous CRC (6.4% CMS2 in mucinous, vs 49.9% in non-mucinous, p<0.0001). This is an important finding, because CMS2 is the most common subtype in CRC and characterizes canonical alterations in the WNT pathway with alterations in copy number [15]. Clearly, given the differences in location, microsatellite status, and CMS subtypes, the biological pathways that drive mucinous CRC are distinct from non-mucinous adenocarcinoma.

Hence, this introduces the aspect of genetic variation in the presence or absence of a mucinous histology. The co-occurrence of BRAF mutations with the mucinous subtype in CRC is well-documented in the literature [6,12] and further supported by our findings. A similar relationship has been delineated between mucinous differentiation and KRAS mutations; which some have hypothesized may contribute to mucinous differentiation [12,40]. However, in our data, the association of KRAS and mucinous histology is modest. Some studies have reported a similarly low incidence of KRAS mutations in association with mucinous carcinoma [6], but this may be the result of a larger fraction of the patient population presenting with BRAF mutations, which are nearly mutually exclusive with KRAS [41], and the effect of this relationship would be heavily pronounced in a small cohort. Furthermore, the elevated rates of BRAF and KRAS mutations in mucinous tumors are both independent predictors of poorer clinical outcome in our studies and as previously reported [42-44].

Our evaluation of the baseline characteristics associated with mucinous CRC revealed significantly higher prevalence of microsatellite instability. This agrees with previous studies in confirming the correlation between mucinous histology and high microsatellite instability [6,45,46]. Previous literature has linked microsatellite instability with methylation of the promoter region of the MLH1 gene, which relates to the high percentages of CIMP detected with mucinous CRC among our cohort [6,47]. Interestingly, the relationship between microsatellite instability and clinical outcome in CRC has been largely favorable in the early stage colorectal cancer, while it appears less favourable in patients with metastatic disease [48,49]. Furthermore, our findings indicate a significant association between the mucinous subtype and right-sided location; although some studies show similar results [46,50], other analyses have not [6].

Given the number of adverse features associated with mucinous histology, it is important to consider whether its negative prognosis is due to a molecular subtype or if it is merely due to an association with other negative features. Due to the size of our cohort, we could control for negative prognostic features both in multivariate models and in the subgroup analysis of only those patients who had synchronous metastatic disease. Even when controlling these variables, mucinous histology appeared associated with a worse outcome [51]. Although these various interconnecting negative prognostic factors may still contribute to the negative prognosis of mucinous histology, our findings suggest that mucinous histology is a distinct subgroup and its negative prognosis is not merely due to association with these features. Prior literature is in agreement with the correlation demonstrated between mucinous CRC and worse clinical outcome [7,33,50,51].

Despite the size of our study and robust molecular characterization, it must be interpreted in the context of its limitations. Our analysis is subject to the disadvantages of a retrospective design, lack of adjustment for treatment variables and limitation to the 46 genes included on our NGS panel. Moreover, since this study was conducted at a single institute, there are likely population imbalances such as ethnicity and socioeconomic factors that may also introduce bias into our cohort. Irrespective of these limitations, this study provides a validation of several associations seen in smaller studies in this large cohort and also provides valuable new insights into the molecular underpinnings of the mucinous histology that we hope will help support the development of precision oncology for this population.

Conclusion

This study provides one of the largest and most comprehensive analyses of mucinous histology to date and highlights important differences from non-mucinous adenocarcinoma. Not only do these differences contribute to the pathogenesis of mucinous CRC, these differential pathways appear to impact prognosis. In view of its poor prognosis, it may be beneficial to approach this subtype as a discrete biological entity, such that its clinical and molecular distinctions be incorporated into therapeutic decision-making and prognostic evaluation.

Clinical Practice Points.

The carcinogenesis of mucinous CRC is poorly understood; however, unique clinico-pathologic characteristics documented in previous reports suggest a pathogenesis distinct from non-mucinous adenocarcinoma; this includes a higher likelihood of exhibiting microsatellite instability, the CIMP-H phenotype, or an advanced tumor stage at diagnosis. Overall, mucinous histology has been shown to predict inferior survival outcomes, either independently or in conjunction with adverse prognostic factors such as BRAF mutations. However, the genomic and pathologic characterization of mucinous CRC as a distinct entity requires further efforts.

We report novel findings with respect to increased prevalence of GNAS, ERBB2 and SMAD4mutations, which may indicate a potential function of these mutant genes in the pathogenesis of mucinous CRC. In addition, the association of mucinous histology with CMS1 and the lower frequency of CMS2, in combination with the inclination of mucinous CRC to present with a right-sided location, microsatellite instability, and mutated KRAS and BRAF in our population, indicates that the biological pathways driving mucinous CRC are distinct from those of non-mucinous adenocarcinoma.

The demarcation of this subtype as a discrete biological entity may have significant clinical implications. Given the independent role of mucinous histology as a poor prognostic factor in patients, early recognition of these molecular distinctions may improve effective risk stratification and prognostic evaluation, and may be incorporated into therapeutic decision-making in the future.