Precision Treatment and Prevention of Colorectal Cancer—Hope or Hype?

Introduction

Precision medicine is defined as “an emerging approach for disease treatment and prevention that takes into account individual variability in genes, environment, and lifestyle for each person.”¹ The goal is to define and identify meaningful subgroups and to apply tailored approaches for screening and/or treatments. While this initiative has gained traction since it was introduced in 2015 by the Precision Medicine Initiative, questions have been raised about how precision medicine compares to more “traditional” approaches, based on broadly targeting a disease or a few risk factors. Further, there are concerns about how precision medicine might be misleading or even fleecing the public.² Precision-based therapeutic approaches for the treatment of colorectal cancer (CRC) appear to hold great promise and have already begun to transform clinical practice. For CRC prevention, precision strategies for identification of high-risk populations and tailored approaches have not yet fulfilled their hype and, furthermore, direct-to-consumer (DTC) genetic testing could be misleading regarding cancer risk. In this commentary, we provide perspective on the status of precision medicine for CRC treatment and screening by highlighting both the hope and hype of its clinical application.

Hope in CRC precision medicine

CRC treatment and prognostication.

The most immediately promising application of precision medicine is in cancer therapy and prognostication.³ In CRC, three key molecular phenomena – high-level microsatellite instability (MSI-H) and somatic mutations in the BRAF and RAS oncogenes – have long been the mainstay of precision medicine in CRC therapeutics as prognostic and/or predictive biomarkers (Table 1). Somatic genetic testing of tumors for these molecular markers is currently standard practice in metastatic CRC and endorsed by professional organizations.^{4, 5} MSI-H has been recognized to be a favorable prognostic marker in CRC, stage for stage, yet also predicts for a lack of benefit from fluoropyrimidine chemotherapy as monotherapy in the adjuvant setting.⁶ Activating somatic mutations in the BRAF oncogene have long been recognized to confer particularly poor prognoses in CRC.⁷ The prognostic impact of activating somatic mutations in KRAS or NRAS, on the other hand, remains a matter of debate, yet they strongly predict for lack of benefit from anti-EGFR therapy with cetuximab or panitumumab.⁸

Table 1:

Colorectal cancer somatic alterations and implications for prognosis and treatment

Somatic marker (frequency %)	Prognostic implications	Treatment implications	Guideline recommendation
KRAS/NRAS activating mutation (40–45%)	Inconsistent data	Lack of benefit from anti-EGFR therapies in metastatic disease	NCCN: Test in metastatic CRC ASCP/CAP/AMP/ASCO: Test if considering anti-EGFR therapy
BRAF V600E mutation (8–12%)	Poor prognosis	Emerging evidence of favorable response to combination targeted therapies (e.g. BRAF/MEK/EGFR inhibitors)	NCCN: Test in metastatic CRC ASCP/CAP/AMP/ASCO: Test in CRC for prognostication
MSI-H (15%)	Favorable prognosis	Favorable response to anti-PD-1 (+/− anti-CTLA-4) immune checkpoint inhibitors in metastatic disease Unfavorable response to single-agent fluoropyrimidine adjuvant therapy	NCCN: Test in all patients ASCP/CAP/AMP/ASCO: Test in all patients
ERBB2 amplification (3–5%)	Unknown	Favorable response to anti-HER2 therapies in metastatic setting Lack of benefit from anti-EGFR therapies	NCCN: Recommends anti-HER2 therapies if HER2 overexpression present (no specific recommendation for routine testing)
NTRK fusion (<1%)	Unknown	Favorable response to TRK inhibitors (e.g. larotrectinib)	NCCN: Test in metastatic CRC
Tumor “sidedness” (40–45% proximal)	Proximal tumors associated with poor prognosis	Right-sided RAS-WT tumors have unfavorable response to anti-EGFR therapies	NCCN: Only metastatic CRC originating from left-sided tumors should be offered anti-EGFR therapies as first line

NCCN (National Comprehensive Cancer Network), ASCP (American Society for Clinical Pathology), AMP (Association for Molecular Pathology), ASCO (American Society for Clinical Oncologists), CRC (colorectal cancer), MSI-H (high level microsatellite instability), TRK (tropomyosin receptor kinase), RAS-WT (RAS wild-type)

Perhaps the most notable advance in precision therapeutics for advanced CRC has been the use of immune checkpoint inhibitor therapy for treatment of refractory MSI-H CRC, biologic mechanisms of which have been reviewed.^{9, 10} Single-agent PD-1 inhibitor therapy (e.g. pembrolizumab or nivolumab) has demonstrated 69–77% disease control rates in such patients, often with durable responses persisting even after discontinuation of therapy.^{11, 12} Combination immune checkpoint inhibitor therapy for MSI-H CRC with nivolumab and ipilimumab may provide even higher likelihood of benefit with minimal added toxicity.¹³ Ongoing clinical trials are examining the role of immune checkpoint inhibitor therapy for the first-line palliative treatment of MSI-H CRC (; ), adjuvant therapy for resected stage III MSI-H colon cancer (), and even as potential chemoprevention in individuals at risk for MSI-H neoplasia due to Lynch syndrome ().

For RAS/BRAF wildtype CRC, a growing array of potential therapeutic targets in other oncogenic pathways are being identified. For example, CRC with ERBB2 (HER2/neu) amplification demonstrates resistance to anti-EGFR therapy,¹⁴ yet may greatly benefit from anti-HER2 therapies (e.g. trastuzumab, pertuzumab, lapatinib) classically used to treat ERBB2-amplified breast cancers.^{15, 16} Somatic oncogenic fusions in the NTRK genes are rare in CRC (<1%), yet confer marked susceptibility (75% response rate) to TRK inhibitors such as larotrectinib, for which tumor site-agnostic FDA approval has been granted.¹⁷ Interestingly, recent data have found that NTRK fusions and other potentially targetable oncogenic fusions in genes such as ROS1, ALK, and FGFR are enriched in MSI-H, RAS/BRAF-wildtype CRC with MLH1 promoter methylation, and that such cancers have particularly poor prognoses in the absence of targeted therapies.^{18, 19} Even for chemotherapy-refractory BRAF-mutated CRC (a subgroup with notoriously poor outcomes), targeted therapies are finally yielding beneficial results with combined RAF/MEK/EGFR inhibitor therapy yielding responses in 48% of patients with median overall survival greater than 15 months.²⁰

Cancer risk assessment and screening.

Identification of individuals at highest risk of CRC due to inherited genetic mutations is widely recommended using clinical criteria and genetic testing.^{21, 22} Multi-gene panel testing, in which multiple genes are tested simultaneously, is currently a standard of care approach in the evaluation of CRC hereditary syndromes. Multi-gene panels also detect pathogenic germline variants that might have gone undiscovered using conventional criteria as highlighted by Yurgelun and colleagues who found that 10% of unselected CRC patients carried a pathogenic variant by panel testing.²³ Yet, despite knowledge of these syndromes and tools to identify them, individuals and their family members remain undiagnosed, and improved efforts to identify high risk patients are needed.²⁴ Moreover, there are questions about the role of identifying mutations for which CRC risk estimates are not well-established, such as single mutations in MutYH, especially using direct-to-consumer (DTC) tests as will be discussed below.

Hype in CRC precision medicine

CRC treatment and prognostication.

While some CRC therapeutics and high-risk patient identification are realizations of personalized medicine, other “personalized” approaches risk over-hype pending further validation. Various efforts to systematically molecularly profile metastatic cancers, including CRCs, with sequencing of tumor tissue and/or circulating cell-free DNA have claimed high rates of therapeutic actionability in anywhere from 37–66% of all cases tested.^25–28 A consistent critique of such headline-worthy results, however, has been that “therapeutic actionability” has typically been defined as the availability of a clinical trial for which a given molecular alteration is an entry criteria, rather than the existence of a proven effective molecularly targeted therapy. For example, one such study²⁸ reported “actionable” findings in 66% of all tested CRCs, yet over half of these were activating KRAS mutations which were considered “actionable,” yet their main actionability is that they predict for lack of benefit from anti-EGFR based therapies (described above). Another study²⁵ reported 50% of metastatic CRC patients in their cohort had a “potentially actionable” mutation and/or amplification identified by testing, yet an appropriate clinical trial at the testing institution was identified for only 30% of subjects, half of whom were unable to enroll, primarily due to lack of available spots and/or ineligibility for the trial for other reasons. The misleading notion of potential therapeutic “actionability” thus greatly risks the notion of providing false hope and can result in psychological harm to patients eager to find their tumor’s “Achilles’ heel”.²⁹

CRC risk assessment and screening.

Another example of precision approaches that require further study and validation are risk models to guide individualized CRC screening. Beyond age and family history, no other risk factors are currently recommended to inform optimal CRC screening (Table 2). Even among these “established” factors, there is controversy about when to start screening given that the American Cancer Society conditional recommendation to start population-based CRC screening at age 45 was not endorsed by other organizations due to lack of evidence.³⁰

Table 2:

Colorectal cancer screening strategies and levels of evidence

Strategy		Guidelines	Strength of recommendation	Evidence
Age	Begin age 50	USMSTF, USPSTF, NCCN, ACP	Strong	RCTs, simulation models
	Begin age 45	ACS	Qualified	Simulation models
Race	African Americans begin age 45	USMSTF, ACP	Weak	Simulation models
Family History*	First degree relative at age 40**	USMSTF	Weak	Retrospective observational studies
Risk Scores	Environmental risk	Not in guidelines	n/a	Prospective observational studies, simulation models
	Genetic risk	Not in guidelines	n/a	Prospective observational studies, simulation models (primarily Caucasian population)
	Combined genetic and environmental risk	Not in guidelines	n/a	Prospective observational studies, simulation models (primarily Caucasian population)

USMSTF (US Multi-Society Task Force on Colorectal Cancer), USPSTF (US Preventative Services Task Force), NCCN (National Comprehensive Cancer Network), ACP (American College of Physicians), ACS (American Cancer Society) FDR (first-degree relative)

^*Individuals who meet criteria for consideration of an inherited syndrome should be referred to genetics and then managed accordingly

^**Individuals with first-degree relative with CRC are recommended to undergo screening starting at age 40 or 10 years prior to age at diagnosis of relative

Predictive models that combine multiple risk factors and/or biomarkers are potential applications of precision medicine for CRC risk stratification.³¹ Polygenic risk scores (also called genetic risk scores) utilize genetic variants known as single nucleotide polypmorphisms (SNPs), that have been associated with CRC in genome-wide association studies (GWAS).³² Individual SNPs have a small effect on disease risk, but, when taken in aggregate, can predict a significant amount of variation in CRC risk and can therefore impact screening recommendations. To date, more than 50 loci associated with CRC are currently estimated to explain about 12% of the genetic contribution to CRC risk.³³ However, the discriminatory performance of polygenic risk scores for CRC thus far has been modest,³⁴ especially when compared to scores developed for breast cancer.³⁵

Models combining polygenic risk scores and environmental risk factors have the potential to increase performance compared to either model alone. Jeon et al reported a model combining 19 environmental and lifestyle factors (E-score) with genetic risk factors based on 63 CRC-associated SNPs (G-score) with improved discriminatory ability than with either model alone.³⁶ Using this model, patients without a family history of CRC with a score in the bottom 10^th or top 90^th percentile could be stratified into low and high risk groups with a recommended initial CRC screening age that differs by 12–14 years.³⁶ Combination of risk scores with objective markers, such as FIT testing, has been also been shown to increase their discriminatory capacity for advanced neoplasia or CRC detection, albeit marginally.³¹ In addition to further discovery of CRC-associated variants to increase the discriminatory capacity of genetic risk models, these approaches will require further validation and extension to non-European populations.

While risk models could help tailor CRC screening, application to the general population still needs to be proven. Current CRC screening strategies rely on opportunistic patient interactions and screening uptake. Any new precision prevention tools should be simple, accurate, convenient, non-invasive and possibly cost-effective compared to existing strategies to gain wide acceptance. Risk models, for example, could benefit from direct input from electronic medical records rather than patient questionnaires, which can decrease uptake³⁷ and are subject to recall bias. Ethical, privacy-related, financial and psychological concerns of obtaining genetic information may prove to be barriers. The impact of such barriers may be magnified in minority populations.³⁸ Importantly, lack of diversity of derivation cohorts for existing risk models is a significant vulnerability for widespread clinical application. This lack of generalizability is most pronounced for polygenic risk models, which are overwhelmingly derived from populations of European ancestry.³⁹ Indeed, external validation studies have demonstrated worse model performance when applied to diverse populations.³⁴ Even with identification of more CRC risk variants and optimization of performance characteristics, application of risk models to real world clinical practice does not appear to be on the immediate horizon and will require additional variant discovery and validation.

Finally, DTC testing for a variety of indications has gained momentum for personalized medicine in the general public, though concerns about utility, validity and privacy could outweigh benefits of greater awareness of genetic conditions, accessibility, and reduced cost. For example, in January 2019, the Food and Drug Administration (FDA) approved inclusion of the 2 most common MutYH variants in European populations on the 23&Me test.⁴⁰ When 2 mutated copies of MutYH are inherited (i.e., biallelic), an individual develops an attenuated adenomatous polyposis syndrome with markedly elevated CRC risk. However, data on CRC risk in carriers of a single mutated MutYH gene (i.e., monoallelic) are conflicting, and current guidelines recommend higher risk screening only in monoallelic carriers with a first degree relative with CRC.²² Given that a single mutation in MutYH is found in approximately 1 in 50 individuals of European descent, many individuals using this DTC test will be monoallelic carriers, but they might or might not require earlier CRC screening based on this result. Thus, for CRC risk assessment, current DTC panels could lead to several negative outcomes such as misinterpretation or false reassurance based on an individual’s race/ethnicity, personal and/or family cancer history. As recommended by the American College of Medical Genetics, DTC genetic test consumers should be informed about what testing can or can’t say about health conditions as well as understand the validity and utility of an individual test as well as the privacy of their genetic information.⁴¹

Conclusions

Precision medicine offers potential for tailored approaches to prevent and treat CRC. Targeted therapeutics have truly transformed outcomes for some CRC patients, while predictive models using genetic and environmental factors for CRC risk stratification and tailored screening are emerging but will require validation in diverse populations. Additional research is needed such as discovery of new targets and therapeutics, development and validation of more precise risk predictive models, inclusion of diverse populations, and implementation science. Application of precision approaches to CRC surveillance also requires further study. To fully realize precision medicine in the population, healthcare also will need to adapt to handle big data, disruptive technologies and artificial intelligence. At the same time, precision medicine initiatives should not detract from critical public health efforts to increase CRC prevention in the general population; rather, knowledge gained from precision-based practices can be used to inform future public health efforts. As we move toward realization of precision medicine, we must guard against overhype but should be optimistic about the hope and, reality for some, of individualized and equitable CRC treatment and prevention.

Abbreviations:

B2M

beta-2-microglobulin

CRC

colorectal cancer

DTC

direct-to-consumer

GWAS

genome-wide association study

MSI-H

microsatellite instability-high

SNP

single nucleotide polymorphism

TRK

tropomyosin receptor kinase