The current state of molecular profiling in gastrointestinal malignancies

This is a review of the current state of molecular profiling in gastrointestinal (GI) cancers and what to expect from this evolving field in the future. Individualized medicine is moving from broad panel testing of numerous genes or gene products in tumor biopsy samples, identifying biomarkers of prognosis and treatment response, to relatively noninvasive liquid biopsy assays, building on what we have learned in our tumor analysis and growing into its own evolving predictive and prognostic subspecialty. Hence, the field of GI precision oncology is exploding, and this review endeavors to summarize where we are now in preparation for the journey ahead.


Introduction
Comprehensive molecular profiling has evolved over the last decade. The evolution of molecular profiling changed the face of oncology from standard chemotherapy based on histology to personalized therapy. Using immunohistochemistry (IHC), fluorescence in situ hybridization (FISH), and whole-genome sequencing, oncologists are able to recognize the genomic drivers of tumorigenesis and provide patients with prognostic biomarkers and targeted therapy options ( Fig. 1; Table 1).
Molecular profiling using tissue next-generation sequencing (NGS) has become a standard of care practice, and recently, circulating tumor DNA (ctDNA) has emerged as a tool for molecular profiling, a predictor of response to systemic treatment, and a powerful way to measure minimal residual disease (MRD) using less invasive approaches [1][2][3].
A decade ago, the treatment of colorectal cancer (CRC) was the first of all GI malignancies to be influenced by molecular profiling. It was initially observed that patients with KRAS mutant CRC do not respond to epidermal growth factor rector (EGFR)-targeted agents such as the monoclonal antibodies cetuximab and panitumumab [4,5]. Later, it was recognized that BRAF, NRAS, and PIK3CA mutations and HER2 mutations and amplifications also confer non-responsiveness to EGFR-targeted agents and carry a generally poorer patient prognosis than wild-type disease [6,7]. Subsequently, patients with BRAF V600E mutant CRC were shown to benefit from treatment with vemurafenib, a small molecule tyrosine kinase inhibitor [8], and later, encorafenib combined with cetuximab emerged as a standard of care for this subset of patients following chemotherapy [9] (Table 1). With the advent of immune checkpoint inhibitors (ICIs) that target PD-1/PD-L1, microsatellite instability (MSI) was found to be the most significant predictor of CRC treatment response. MSI can be sporadic or driven by germline mutations in one of the MMR genes (MLH1, MSH2, MSH6, or PMS2), as found in hereditary Lynch syndrome [10,11].
MSI testing is essential for all GI malignancies because localized MSI-high (MSI-H) tumors will have a good prognosis, and advanced disease will likely respond to PD-1/PD-L1 inhibitors [12,13] (Table 1). Until very recently, advanced gastric and gastroesophageal cancers were treated solely with conventional chemotherapy. Now, advances in molecular profiling and signaling pathway knowledge have provided new treatment options for patients with PD-L1 or HER-2 overexpressing tumors, including PD-1/PD-L1 inhibitor therapy, HER2 targeted treatment, and anti-vascular endothelial growth factor (VEGF) antibody therapy [14][15][16][17].
In pancreatic cancer, understanding the role of germline testing for BRCA1, BRCA2, and PALB2 in homologous recombination repair has allowed the emergence of poly(adenosine diphosphate-ribose) polymerase inhibitors (PARPi) as a treatment option [22].

PD-L1
PD-1 is an inhibitory receptor expressed on several immune cells, particularly cytotoxic T cells. It interacts with 2 ligands: PD-L1 and PD-L2. PD-L1 is expressed on tumor cells and immune cells, whereas PD-L2 is expressed on macrophages and dendritic cells. The interaction of PD-1 with PD-L1 inhibits T-cell activation and cytokine production, which is vital to maintaining homeostasis of the immune response and preventing autoimmunity [23,24]. However, PD-1/ PD-L1 interactions within the tumor microenvironment provide an immune escape pathway for tumor cells by turning off cytotoxic T cells [25]. Tumor cells upregulate the PD-1 receptor or ligand to evade destruction by the host immune system. Thus, in blocking the PD-1 pathway with antibodies to PD-1 and PD-L1/PD-L2, the adaptive immune response is activated against tumor cells resulting in an anticancer response. Tumor cell PD-L1 expression is associated with response to anti-PD-1/anti-PD-L1 therapy [26].
PD-L1 protein expression in many cancer types, assessed via immunohistochemistry (IHC), is one of the FDA-approved predictive biomarkers for anti-PD-1 and anti-PD-L1 ICI monotherapy [26]. However, PD-L1 expression within tumors and between tumor sites may be heterogeneous [27][28][29], and assays may give variable results. To the latter point, there are multiple qualitative PD-L1 assays involving different antibodies to assess the expression of PD-L1 by IHC using chromogenic methods [30], and different antibody assays may give different results.
In esophageal/GEJ/gastric cancer, the PD-L1 combined positive score (CPS) has been tested as a predictive biomarker for immunotherapy. CPS is the number of cells staining for PD-L1 cells (tumor cells, lymphocytes, and macrophages) divided by the total number of evaluated tumor cells, multiplied by 100 [31]. Tumors are considered PD-L1 positive if they have a CPS > 1. A positive CPS is associated with improved GI cancer patient outcomes upon ICI therapy. In KEYNOTE 062, KEYNOTE 061, and KEYNOTE 059, the PD-1 inhibitor pembrolizumab demonstrated efficacy against gastric and GEJ cancer as first, second, or third-line treatment based on a CPS of > 1 [32]. On the other hand, the PD-1 inhibitor nivolumab is FDA approved in esophageal/GEJ/gastric cancers regardless of CPS, based on CHECKMATE 648 and CHECKMATE 649 studies [14,33]. How to use CPS in the selection of upper GI cancer patients for frontline ICI therapy remains a point of debate in the oncology community [34].

TMB: tumor mutation burden
TMB, defined as the total number of exon mutations in a tissue sample [35], has emerged as an important biomarker associated with immunotherapy response in multiple tumor types [26]. TMB is a critical driver in the generation of immunogenic neopeptides presented on major histocompatibility complexes on the tumor cell surface [36]. These immunogenic components influence response to ICIs, meaning that TMB impacts ICI efficacy. This TMB effect on ICI efficacy is reflected in many retrospective studies, including the Phase II Keynote-158 trial [37], which led to the FDA approval of pembrolizumab in patients with unresectable or metastatic high TMB (≥ 10 mutations/megabase) solid tumors [38].
Detecting ctDNA in the blood is a noninvasive test called liquid biopsy (see the section on Molecular profiling in the blood). With an increasing interest in ctDNA, studies have been carried out to develop methods, including NGS, that can estimate the tumor fraction in a patient's plasma and measure the TMB from their blood with high accuracy [39]. Chalmers et al. [40] demonstrated that TMB can be accurately measured in blood by sequencing targeted gene panels, but accuracy is compromised when the sequenced genome region is less than 0.5 MB. The Guardant Health Omni panel (500 genes, 2.1 MB) and Foundation Medicine bTMB panel (394 genes, 1.14 MB) are plasma-based NGS assays containing sufficiently large genome region sizes to measure TMB across a broad range of TMB values. In a study by Qiu et al. [41], Guardant Health and Foundation Medicine tests were evaluated and compared in their ability to evaluate TMB from ctDNA. The investigators ascertained that tissue and plasma TMB correlated well using both assays as long as analyzed samples contained a high TMB; the correlation was compromised if samples contained only low to medium TMB [41].
TMB has been used as a predictor for response to ICI therapy, and early measurements of ctDNA were shown to help detect treatment failure [42]. MRD is another important biomarker of treatment failure. It refers to residual tumor cells present after cancer treatment and is associated with disease recurrence. MRD can be detected in blood using techniques like quantitative PCR and NGS. However, most recently, ctDNA has been used to detect MRD in the blood, serving as a powerful diagnostic and predictive tool (Fig. 2) [43].

MSI
Microsatellite instability (MSI) is a molecular fingerprint for defects in the mismatch repair system (dMMR), which is associated with an increased risk of cancer [44]. The MMR system is composed of heterodimers (MSH2/MSH6 and MSH2/MSH3 complexes) that ensure the specific recognition of mispaired nucleotides generated due to DNA damage [45]. In humans, these complexes initiate DNA repair and recruit MLH1/PMS1, MLH1/PMS2, and MLH1/ MLH3 heterodimers to catalyze the excision of the mispaired nucleotides as well as error-free DNA resynthesis. Genetic and epigenetic inactivation of MMR genes cause MMR defects (dMMR) and give rise to spontaneous, genome-wide mutations [45]. This mainly affects the short tandem repeat DNA sequences termed microsatellites, which occur at specific foci throughout the genome.
MSI-H tumors contain many mutation-associated neoantigens, which, it is believed, are recognized as foreign by the immune system. The benefit of ICIs for patients with MSI-H/dMMR tumors was first documented in a Phase II trial, in which patients with metastatic cancer (78% colorectal) with and without dMMR received pembrolizumab. Only the patients with MSI-H/dMMR tumors benefited from the ICI therapy [46]. Results from this trial were confirmed in the larger Phase II KEYNOTE 158 study evaluating pembrolizumab in dMMR metastatic colorectal patients [37]. Later, results from the KEYNOTE 177 trial of pembrolizumab as first-line treatment for patients with MSI-H/dMMR metastatic colorectal cancer showed longer progression-free survival (PFS) compared to standard chemotherapy [47]. Similarly, the CHECK-MATE 142 trial suggested durable benefit from combined nivolumab plus ipilimumab (anti-CTLA-4) in patients with MSI-H/dMMR tumors [48].
Conventional methods used for MSI testing include immunohistochemistry (IHC) and PCR-based assays performed on tumor tissue samples. Tumor tissuebased NGS can also determine MSI status.
Several strategies have been developed to target HER2, including extracellular antibodies like trastuzumab, which targets domain IV of the receptor, and pertuzumab which binds to domain II and inhibits the heterodimerization of HER2 with other ErbB receptors [58]. Additionally, small tyrosine kinase inhibitors like lapatinib, tucatinib, or neratinib inhibit HER2 activity, while antibody-drug conjugates (ADCs), such as trastuzumab emtansine (T-DM1) and trastuzumab deruxtecan (T-Dxd), bind HER2 and introduce a potent cytotoxic agent into cells overexpressing the receptor [16].
The Phase III ToGA (Trastuzumab for Gastric Cancer) trial for patients with gastric or GEJ cancer with overexpression of HER2 or gene amplification was the first study to demonstrate the therapeutic benefit of targeting HER2 in GI cancers [15]. The US FDA subsequently approved trastuzumab for the first-line treatment of patients with metastatic, HER2-positive gastric or gastroesophageal cancer.
Subsequent trials in patients with GI cancers include the Phase 2 HERACLES trial [59] and the ongoing MyPathway basket trial [60]. In the HERACLES trial [59], standard treatment-refractory patients with KRAS wildtype (wt) CRC harboring a HER2 amplification received trastuzumab and lapatinib. In a subset of the ongoing MyPathway basket trial [60], patients with HER2amplified metastatic CRC received pertuzumab plus trastuzumab. The objective response rates were around 30% in both studies, and several other patients had stable disease, demonstrating that HER2 amplification is an actionable target. More recently, a trial of T-Dxd in previously treated patients with gastric or GEJ cancer found improved overall survival (OS; 12.5 months vs. 8.4 months; P = 0.0097) and ORR (40.5% vs. 11.3%) compared to standard chemotherapy. These results led to the US FDA approval of T-Dxd in the third or later lines of therapy [16].
The development of tumor resistance to HER2 inhibitors is a problem for which there are multiple possible mechanisms, including loss of HER2 expression and HER3 ligand-dependent HER2-HER3 interactions leading to evasion of apoptosis [61]. A TAF/FGF5/ FGFR2/c-Src/HER2 axis might act as a HER2-targeted therapy escape pathway, which seems to be reversed by FGFR inhibition [62].
Prior attempts to demonstrate an OS benefit from second-line HER2-targeted vs. standard cytotoxic therapy have failed, possibly due to loss of HER2 expression following trastuzumab-based first-line treatment. MOUN-TAINEER-02 trial investigators hope that dual targeting of HER2 with tucatinib and trastuzumab will overcome this resistance. Tucatinib is a small molecule tyrosine kinase inhibitor, which was shown to have "very potent, " selective activity against HER2, with minimal off-target effects [63]. The ongoing Phase II/III MOUNTAIN-EER-02 trial [NCT04499924] [64] is enrolling patients with advanced or metastatic HER2-positive (overexpression or amplification) gastric or GEJ cancer with disease progression (PD) after frontline therapy, including a HER2-directed antibody. Patients receive second-line treatment with paclitaxel plus ramucirumab, either with tucatinib plus trastuzumab, tucatinib plus trastuzumabplacebo, tucatinib-placebo plus trastuzumab, or two placebos.
Future directions include using liquid biopsy genotyping assays as a viable, real-time alternative to tissue-based genotyping in the identification of HER2 alterations in the metastatic setting. HER2 copy number is typically assessed using surgically-obtained tissue, but necessary information can now be obtained conveniently and noninvasively using ctDNA (See the "Molecular profiling in blood" section).
FGFR has become a molecular target of increasing interest in CCA. There are several FGFR-targeted therapies of interest, mostly in the form of tyrosine kinase inhibitors (TKIs). Pemigatinib is a selective oral FGFR1-3 inhibitor investigated in the open-label singlearm FIGHT-202 trial for previously treated advanced CCA [77]. Among patients with FGFR2 alterations, pemigatinib displayed an overall response rate (ORR) of 36%, disease control rate (DCR) of 80%, and median duration of response (DOR) of 7.5 months. Pemigatininb is now FDA approved for previously treated, unresectable, locally advanced, or metastatic CCA with an FGFR2 alteration. Pemigatininb combined with gemcitabine and cisplatin is currently being studied in the first-line, phase III FIGHT-302 trial (NCT03656536). Infigratinib, another selective FGFR 1-3 TKI, obtained accelerated FDA approval in subsequent-line settings for FGFR2altered CCA. This drug demonstrated an ORR of 23% and mDOR of 5 months [78]. In CCA, other drugs currently under study include derazantinib (FGFR1-3 inhibitor) and erdafinitib (FGFR1-4 inhibitor). Toxicities of FGFR inhibitors are predictable and similar across this class of therapeutics and include hyperphosphatemia (50-80%), nail toxicity (35%), and ophthalmologic toxicity (4-9%) [77,[79][80][81].

IDH1/2
Isocitrate dehydrogenase (IDH) is a key enzyme in the tricarboxylic acid cycle and comprises 2 subtypes: IDH1, located in the peroxisomes and cytosol, and IDH2, located in the mitochondria [82,83]. In CCA, IDH1 mutations are found in 15-25% of cases, particularly in intrahepatic CCA. IDH2 mutations are less frequent, found in up to 3% of CCAs [84,85]. IDH mutations generally lead to a gain-of-function that disrupts normal catalytic activity. The net effect is increased conversion of α-ketoglutarate to D-2-hydroxyglutarate, which leads to downstream cellular proliferation through pathways including DNA methylation and VEGFR [82,83,86].
Multiple IDH-selective inhibitors are being investigated in vitro and in clinical trials. Ivosidenib, an oral small-molecule inhibitor, was among the first to be studied clinically: A phase I study confirmed tolerability and demonstrated a median PFS (mPFS) of 3.8 months in previously treated patients with IDH1-mutated CCA [87]. The recent ClarIDHy Phase III placebo-controlled trial demonstrated an OS benefit from ivosidenib that trended towards statistical significance (mOS 10.3 vs. 7.5 months, HR 0.79, 95% CI 0.56-1.12; p = 0.093), becoming statistically significant once a mathematical model adjusting for treatment crossover effects was employed (mOS 10.3 vs. 5.1 months, HR = 0.49; 95% CI 0.34-0.70; p < 0.0001) [88]. Ivosidenib was granted FDA approval in August 2021 [89]. Additional promising IDH1-targeting drugs are under early investigation, including another smallmolecule inhibitor, olutasidenib (NCT03684811).

BRCA/PALB2
Germline mutations in BRCA1 and BRCA2 are well studied and associated with a high risk of cancer, particularly breast and ovarian cancers, with high hereditary penetrance in an autosomal dominant pattern [90][91][92]. This associated risk has been established across other cancers, including pancreatic cancer, where BRCA2 mutations pose a relative risk of 3.5-10 compared to non-carriers.
BRCA1 and BRCA2 are suppressor genes of the same family, located on long arms of chromosome 17 and 13, respectively [93,94]. They play instrumental roles in DNA damage response, particularly in maintaining chromosomal stability in the process of homologous recombination repair [95,96]. One of the early successes of whole-genome sequencing was the identification of the BRCA2 and its partner and localizer gene PALB2 in 1-4% of familial PDAC [90,97,98]. PALB2 colocalizes with BRCA2 at the site of DNA damage to enable DNA repair [99].
Deficiencies in homologous recombination and DNA repair pathways predict sensitivity to platinum-based chemotherapy regimens as well as poly(ADP-ribose) polymerase inhibitors (PARPi). Patients with PDAC and homologous recombination gene mutations had improved PFS and OS when treated with frontline platinum-based therapy compared to patients without such mutations (HR 0.44, 95% CI 0.29-0.67; P < 0.01) [100,101]. Similarly, PARPi appear to be active in PDAC; the phase III POLO trial demonstrated sensitivity to olaparib of patient tumors with homologous recombination gene mutations. Maintenance olaparib after platinum-based induction therapy showed superior mPFS to placebo (7.4 vs. 3.8 months, HR 0.53, 95% CI 0.35-0.82) [102]. Olaparib is FDA-approved in the maintenance setting for PDAC.
In the NCI-MATCH EAY131-H trial, a combination of dabrafenib (BRAF inhibitor) and trametinib (MEK inhibitor) produced favorable response rates in a total of 35 pretreated patients with a range of solid tumors, all harboring a BRAF V600E mutation [104]. A confirmed objective response rate (ORR) of 37.9% (90% CI 22.9-54.9%; P < 0.0001 against a null rate of 5%) was reported. The median duration of response was 25.1 months. Four of the 35 patients enrolled on trial had CCA, and 3 of these 4 experienced a partial response (PR) [104].
The same drug combination was tested in patients with bile duct cancer harboring BRAFV600E mutations as part of the ROAR Basket Trial [21]. The ROAR Basket Trial enrolled 9 different cohorts of 178 patients with rare malignancies, all harboring BRAF V600E mutations. The bile-duct cancer cohort included 35 patients treated with the combination of the BRAF inhibitor dabrafenib and the MEK inhibitor trametinib. Most patients (74%) had stage IV disease at enrollment, and all 35 patients with biliary tract cancer had received prior chemotherapy (80% had received at least 2 prior lines). The median duration of treatment exposure was 6 months, with 86% of the patients being treated for more than 3 months. The median follow-up was 8 months. A PR was reported in 42% of the cohort by investigator assessment and 36% by independent review. Stable disease was achieved in 45% and 39%, respectively, and 12% had progressive disease as their best response (by either assessment method). The mPFS by investigator assessment was 9.2 months, and the median OS was 11.7 months. With regard to safety, adverse effects (AEs) were found to be comparable to those previously reported with each agent alone, with no new toxicities observed. Potential study treatmentrelated toxicities included fatigue, neutropenia, hyponatremia, and hypophosphatemia. No grade 5 events were observed [21].

Molecular profiling in the blood
A liquid biopsy identifies components of a tumor in the blood, such as ctDNA, circulating tumor cells (CTC), circulating tumor RNAs, and circulating tumor exosomes [105]. Of all these entities, ctDNA is the most studied to date and comprises fragments of cell-free DNA (cfDNA) that retain tumor-specific mutations and epigenetic characteristics [106][107][108]. These fragments are released into the circulation spontaneously or after apoptosis or necrosis. ctDNA makes up a highly variable fraction of total cfDNA in peripheral blood, and this fraction (reportedly ranging from less than 0.1% to almost 90%) is impacted by disease stage, tumor type, and analysis technique, to name a few. The half-life of ctDNA is 16 min to 2 h, allowing for indirect real-time tumor characterization [109][110][111]. Although the liquid biopsy was described as early as the 1940s, it is only recently, with enhancements in genomic sequencing techniques and identification of novel biomarkers, that we have seen more commercialized applications of liquid biopsy platforms [112][113][114].

Overview
The advantages of blood-based molecular profiling include the ability to study intra-and inter-tumor genomic heterogeneity, the ability to perform tumor profiling in the absence of available tissue, avoidance of invasive biopsy procedures, the feasibility of frequent longitudinal testing, and a quick turnaround time to inform treatment plans [115]. To date, numerous commercial liquid biopsy profiling assays are available to providers. Methodologies overlap with tissue profiling and involve comprehensive genomic, single-gene, or hotspot gene testing. In hotspot testing, commonly altered regions within select genes are evaluated. Of the comprehensive genomic tests, only two are FDA approved today. Guardant Health's Guardant360 CDx, targeting 55-74 genes using DNA NGS, was first FDA approved on August 7, 2020. Roche's FoundationOne Liquid CDx, targeting over 300 genes using DNA NGS, was first approved on August 26, 2020 [116][117][118]. Although approved as companion diagnostic tests for therapeutics in lung, prostate, ovarian, and breast cancer, these assays are increasingly studied and adopted in GI cancers to identify prognostic and predictive biomarkers, actionable mutations with FDA-approved therapies, responses to treatment, and mechanisms of resistance. These and other comprehensive gene-based assays utilize NGS and can identify single nucleotide variants (SNVs), insertion and deletions, gene rearrangements, copy number variants (CNVs), TMB scoring, and microsatellite status [112,116,119,120].
Numerous studies have recently evaluated ctDNA genomic profiling in GI cancers to identify actionable mutations, monitor disease response, and understand resistance mechanisms ( Fig. 2; Table 2). Concordance rates of 85-98% have been reported for ctDNA and tissue genomic profiling, and promising sensitivities and specificities in NGS/PCR-based ctDNA assays, for example, 50-93% and 97-100% respectively for RAS mutations, 63-100%, and 98-100% for BRAF mutations, and 33-98% and 98% for ERBB2 amplifications, have been observed [109,[121][122][123]. In addition, studies have suggested that ctDNA captures genomic heterogeneity between primary and metastatic sites, which should act to enhance therapy selection [124]. As a noninvasive and convenient test, blood-based genomic profiling has the potential to replace or complement tissue testing, particularly when considering targeted therapies.
Taking anti-EGFR therapy as an example [125][126][127]131], ctDNA studies demonstrate that RAS/EGFR mutant clones emerge during treatment, which might regress upon the withdrawal of anti-EGFR therapy, thereby allowing for rechallenge with the targeted therapy [128]. This regression could not be reasonably assessed using tumor tissue because it would mean the risk of repeated biopsying. Being much less invasive, liquid biopsy and ctDNA analysis may allow for uncomplicated identification of patients suitable for rechallenge based on realtime genomic analysis. This idea was evaluated in the CHRONOS phase II trial, the first prospective interventional study to use liquid biopsy to guide anti-EGFR rechallenge therapy in CRC. Hence, liquid biopsy genotyping differentiated between patients with RAS/BRAF/ EGFR mutated versus wild-type tumors, and one-third of wild-type patients had an objective response on rechallenge with an anti-EGFR antibody [129]. Other studies have identified specific EGFR and RAS mutations in the plasma after disease progression on treatment and highlight ctDNA as a tool in clinical practice to inform therapeutic development and tailor treatments based on emerging resistance mutations [128]. The feasibility of BRAF plasma testing and similar potential applications in this targetable gene have also been demonstrated [130,131,135].
Another good example of ctDNA application is in HER2 amplified disease. Recent prospective trials have suggested that plasma HER2 amplifications predict response to HER2-directed therapies such as lapatinib, trastuzumab, pertuzumab, and fam-trastuzumab deruxtecan-nxki (T-DXd) [136][137][138][139]. For example, better responses to T-DXd in gastric cancers were seen when plasma HER2 and higher copy number amplifications were detected [138]. Also, changes in plasma copy number during HER2-directed therapy were associated with therapeutic response and survival in upper GI cancers and CRCs [122,136,137,139]. Baseline and emerging resistance mutations detected in the plasma at the time of progression, such as MYC, EGFR, FGFR2, and MET amplification, have also been reported along with promising therapeutic strategies to overcome resistance, including combining anti-HER2 and other targeted or immune therapies [136,139].
ctDNA can detect plasma FGFR2 alterations, sometimes at a higher frequency than tissue testing, identify patients who may benefit from infigratinib, and identify emerging resistance point mutations [140][141][142]. Recently, plasma-tissue accuracy and survival data have also been described in CCA patients with IDH1 plasma mutations treated with ivosidenib [144]. PDAC is highly KRASmutated and harbors rarer targetable mutations. Many studies have evaluated KRAS and TP53 ctDNA detection and kinetics as prognostic and predictive biomarkers in PDAC cases [145,147,149,150].
Early data suggest that plasma MSS and TMB have potential roles as prognostic and predictive biomarkers in GI cancers, although further study is needed to validate these assays in the guidance of immunotherapy [151,152]. Collectively, these data highlight the evolving potential of blood-based molecular profiling to precisely guide patient therapy and overcome tumor resistance using only minimally invasive procedures. Consequently, clinical trials across GI cancers are increasingly focused     [146] Multiple genes PDAC Prospective

Limitations
The liquid biopsy has been rapidly integrated into clinical practice for many solid tumors, but physicians should exercise caution when interpreting results. Tumor samples provide an abundance of tumor DNA compared to a liquid biopsy; therefore, liquid-based NGS may be limited by apparent lower sensitivity due to significantly lower levels of ctDNA [119]. Often, performing NGS on several genes means reduced depth of sequencing and sensitivity, not to mention added cost and effort compared to targeted sequencing of one or a few genes [119,120,155]. Studies have demonstrated that ctDNA levels and concordance may vary based on tumor histology, anatomic location, and stage [110,156,157]. Limitations arise if the tumor is a poor ctDNA shedder [122,158]. Additionally, the presence of clonal hematopoiesis may lead to false positives in genotyping [159]. Unique molecular identifiers, dual barcode indexing, methylation assays, matched normal WBC DNA analysis, and cfDNA fragment length analysis can improve the accuracy of ctDNA analysis [109,160,161]; however, differences between laboratory testing platforms can contribute to discordant results across many mutations [162]. Despite the intrinsic limitations of assays, blood-based genomic profiling remains a promising tool for patient diagnosis, therapy guidance, and identification of patients for trial enrollment.

ctDNA use in screening and diagnosing cases with insufficient tissue
As molecular profiling technology has advanced, there is increasing interest in using ctDNA for cancer screening (Fig. 2), diagnosis of inaccessible tumors, and management of cancers of unknown primary (CUP). Massive genomic profiling efforts, including The Cancer Genome Atlas's Pan-Cancer initiative and other massive genomic profiling efforts, have identified tumor DNA, RNA, and protein patterns based on histology, anatomic location, and tissue types. These patterns can, in turn, be used to characterize undifferentiated tumors and identify tissues of origin [163]. For example, the highly-specific Can-cerSeek test [164] used ctDNA and protein biomarkers coupled with machine-learning to diagnose 8 early-stage cancers, including liver, gastric, pancreatic, esophageal, and colorectal cancer. The Circulating Cell-free Genome Atlas (CCGA) study [165] used a methylation-based assay from Grail to allow deeper sequencing and identified over 50 early-stage cancers, including liver/bile duct, gastric, pancreatic, esophageal, colorectal cancer, and anal cancer. The CancerSeek assay demonstrated sensitivities ranging from 69 to 98% in the detection of 5 cancer types, including liver, stomach, pancreatic, and esophageal [164]. The CCGA assay had 67.3% (CI 60.7-73.3%) sensitivity in 12 cancer types, including anal, colorectal, esophageal, liver/bile duct, pancreatic, and stomach [165]. Both assays had greater than 99% specificity, identifying the tissue of tumor origin with great accuracy. In another active study, Grail's cfDNA methylation assay predicted the tissue of tumor origin with 92.3% accuracy [166]. Although these platforms are promising, they are still limited by inadequate sensitivity in early-stage diseases. Moreover, questions about disease management after detection, feasibility, ethics of general population testing, and cost-benefit ratios remain [167].
As it currently stands, GI cancer-specific but not yet multicancer screening tests have been FDA approved. Epigenomics Epi proColon ® detects methylated SEPT9 DNA in the blood and was FDA approved on April 13, 2016, as the first blood-based CRC screening tool [168,169]. Guardant's LUNAR-2 ctDNA screening test for CRC is also currently under investigation, among others [170]. Blood-based hydroxymethylation and protein glycosylation signatures are promising biomarkers for the early detection of PDAC [171,172]. Distinct circulating miRNAs signatures in blood and bile might act as biomarkers to differentiate between biliary cancers and other benign hepatobiliary diseases [173,174]. However, miRNAs are nonspecific, and the best source of miRNA collection and the translation of miRNA assays into clinical practice are as yet undefined [175,176].
Finally, the feasibility and utility of analyzing ctDNA from blood to characterize CUPs and identify targetable mutations have been described in multiple studies over the past decade [177][178][179][180]. Historically, patients with CUP had limited treatment options and poor prognoses because many standard-of-care therapies are tumor-specific. As more biomarkers are identified, broad blood-based NGS can uncover targetable mutations and identify more previously non-indicated therapies for these patients.

ctDNA in MRD
Efforts are underway to identify patients at high risk of early relapse and develop interventions to lengthen patient survival. The strategy of MRD monitoring (Fig. 2) and eradication is already established in hematologic malignancies and is stimulating interest in GI cancers [181][182][183].
MRD ctDNA assays are often characterized as tumoragnostic or tumor-informed. Tumor-agnostic approaches do not require pre-existing knowledge of a tumor's genomic profile and often employ broad-based NGS, narrower PCR, or methylation assays to identify common cancer markers circulating in the blood. Numerous studies in localized or oligometastatic CRC patients undergoing curative surgery have demonstrated that ctDNA detection post-surgery or post-adjuvant therapy is a strong independent prognostic marker for survival [184][185][186]. Others suggest that ctDNA levels correlate with tumor burden and response to treatment [187]. Promising results in PDAC (often targeting KRAS) and gastroesophageal cancer have also been reported [188][189][190][191]. While limited by decreased sensitivity, tumor-agnostic methods are advantageous in their quick turnaround time, low cost, and simultaneous broad-based genomic profiling and resistance mechanism identification potential.
Although slower and more expensive, a tumorinformed approach offers higher sensitivity and is particularly attractive in assessing MRD when the ctDNA level is low, as in early-stage disease. Here, a patient's tumor tissue is tested using genomic sequencing, and tumor-specific mutations are identified. These mutations are subsequently targeted in the blood using a personalized assay. Examples include SafeSeqS, CAPP-Seq, Archer DX, Radar, and Signatera. To date, most GI cancer studies have been in CRC and demonstrate sensitivities ranging from 48 to 100%, specificities ranging from 90 to 100%, positive predictive values (PPV) over 98%, and median lead times to radiographic relapse of about 8-9 months [192][193][194][195][196][197][198][199][200][201][202][203]. For example, using the Signatera test, a ctDNA positive status after adjuvant therapy and on postoperative longitudinal testing was found to confer 18 times and 30 to 40 times, respectively, higher risk of relapse compared to a ctDNA negative status. Moreover, ctDNA was found to outperform CEA in predicting relapse [192,193,198]. Studies characterizing ctDNA in the provision of early prognostic data are also emerging in gastroesophageal cancer, PDAC, and CCA [188,189,204,205]. Some studies in the adjuvant and metastatic setting also suggest that ctDNA clearance or kinetics may predict treatment response and survival [198,206].
We know that ctDNA positivity is highly prognostic but still lack knowledge on optimal disease management strategies for patients with MRD. While a positive test may theoretically hasten surveillance diagnostics, the role of local or systemic therapy escalation, in this case, is unclear, especially in the setting of serial, lowlevel ctDNA without evidence of radiographic relapse. Also, despite the achievable 0.01% level of ctDNA detection using tumor-informed assays, false negatives due to ctDNA levels being below the limit of detection should be considered when interpreting negative results. This is likely to be the case for low-shedding tumors, for example [207]. Therefore, in the absence of prospective, randomized data to support de-escalation strategies, a patient with a negative test should receive standard adjuvant therapy if not otherwise contraindicated. It is also important to realize that each company's tests are uniquely constructed, using different error-correcting techniques that are frequently updated.

Circulating tumor cells (CTCs)
It has been shown that CTCs are an intermediate stage of cancer metastasis. Like ctDNA, CTCs are obtained from peripheral blood; however, CTCs may have a greater clinical impact as they can be grown, propagated, and extensively studied in vitro and in vivo under optimal conditions [208,209]. However, it is unclear if a single cell assay accurately reflects entire tumor heterogeneity.
Data supporting CTC enumeration as a predictor of clinical outcome dates back as early as 2004, when it was shown that patients with metastatic breast cancer had shorter mPFS and mOS if they had higher CTC levels [210]. Since then, this finding has been further validated across a wide range of tumor types, including GI cancers [208]. CTC enumeration shows promise in clinical management guidance; for example, as shown in breast cancer cases, the discovery of discordant driver mutations between an individual's CTCs and their primary tumor may inform targeted treatment decisions [211][212][213].
Several commercial systems and clinical services (Epic, RareCyte ™ , CytoTrack, SRI FASTcell ™ ) exist [214]. Current methods of CTC detection rely on one of three basic principles. The CellSearch ® system was the first and is the only FDA-approved device for CTC enumeration [215]. The CellSearch ® platform relies on antibody detection of CTC markers [208,216]. Cohen et al. used the CellSearch ™ system to estimate CTCs in their prospective multicenter mCRC study [217]. Their results showed that patients with ≥ 3 CTCs/7.5 mL blood had shorter mOS than patients with < 3 CTCs (P < 0.0001), and these differences persisted at follow-up time points after therapy. It was concluded that the number of CTCs was an independent predictor of disease-free survival (DFS) and OS in metastatic cancer [217]. An alternative CTC enumeration approach relies on isolating CTCs according to prespecified cancer-specific gene products (RNAs and proteins) [208,218]. However, this method involves the lysing of captured CTCs, limiting their use in downstream analyses. A third technique isolates CTCs according to their physical characteristics; CTCs are generally much larger than blood cells (30 µm vs. 7-9 µm), allowing their isolation and enumeration [219].

CTCs and organoids
CTC study has already added to our understanding of cancer metastases. For example, it has been described that CTCs often carry genetic variations in driver mutations that are different from the primary tumor; these differences would likely help explain the propensity for primary tumors to metastasize and seed into other organs [220].
As the next frontier in precision medicine, the ability to grow and expand CTCs ex vivo is an invaluable, noninvasive tool in the study of cancer biology and metastasis [208,209,216]. One step further, the ability to create CTC-derived xenografts by injecting CTCs directly into immunocompromised mouse hosts holds vast implications in both research and clinical settings. These organoids maintain tumor heterogeneity and allow investigation of therapeutic elements on the xenograft that mirror patient response to the same treatment [221,222]. Such clinical applications have already been used to perform in vivo drug screens with high success and hold implications for new drug discovery [221]. For research purposes, organoids can be used for disease modeling to understand the process of carcinogenesis. They can be manipulated easily using retroviruses and inhibitors, for example, and can be used to identify key driver mutations, as shown already in some GI cancer organoid studies [223,224]. The clinical role of CTCs is currently limited but is expected to expand on the heels of technology improvements, including CTC-isolation and organoid-development techniques.
What can we learn from the blood, and how can we use biomarker testing in the future?
Tissue molecular profiling provides clinically significant subtyping of all GI cancers. The liquid biopsy promises dynamic tumor characterization through various platforms, and we believe these capabilities will be increasingly incorporated into clinical cancer management. The liquid biopsy is already integrated into the standard of care for gastric, esophageal, and GEJ cancers, for which NCCN guidelines recommend plasma ctDNA profiling by NGS to detect targetable alterations or clones with altered treatment sensitivity when patients are not candidates for tumor-tissue biopsy and NGS [207].
In the future, molecular profiling of the liquid biopsy will likely complement or replace the GI tumor-tissue biopsy in select scenarios. Future therapeutic studies should include ctDNA analyses to identify prognostic and predictive liquid biopsy biomarkers. Serial testing should also be further assessed as a way to quickly and noninvasively characterize disease response or mechanisms of resistance. Finally, "MRD with no evidence of radiographic disease" might become a theoretical "new stage, " warranting novel treatment strategies. As liquid biopsy techniques improve, blood-based testing will hopefully better identify MRD and screen earlystage GI tumors with hopes of curing more patients and improving outcomes.

Summary
Molecular profiling for patients with GI malignancies is clearly making an impact and has become the standard of care in many situations. In fact, in 2021, ASCO chose molecular profiling in GI cancers as its Advance[ment] of the Year [225]. An increasing number of actionable biomarkers are being identified, and drugs that act on these biomarkers are continually being developed, providing patients with better treatment options, improved quality of life, and increased survival compared to standard therapy alone. Likewise, analytical methods using tumor tissue and, more recently, blood are constantly being developed and improved, promoting the identification of biomarkers and gene signatures that help diagnose disease and predict therapy success in this oftentimes refractory group of malignancies. Together with machine learning, our evolving biomarker technology is promising to help us fight an even smarter war against GI cancers.