The Current Evidence and Future Direction of Adjuvant Treatment for Gastric Cancer in the Era of Precision Medicine

1. Intrinsic molecular subtypes

The first classification of GC based on gene expression patterns was published by a Singapore group [28]. They initially analyzed gene expression patterns in 37 GC cell lines and subsequently validated their findings in independent patient tumor samples. Their analysis identified two distinct groups: genomic-intestinal (G-INT) and genomic-diffuse (G-DIF). G-INT tumors demonstrated better survival and showed sensitivity to 5-fluorouracil (5-FU) and oxaliplatin while exhibiting resistance to cisplatin. Conversely, G-DIF tumors were associated with poor prognosis but demonstrated sensitivity to cisplatin.

Building on these findings, they conducted a proof-of-concept phase II ‘3G’ trial, where G-INT and G-DIF were classified as ‘G1’ and ‘G2’, respectively, with an indeterminate group designated as ‘G3’ [29]. The trial demonstrated that gene profiling could be completed within a clinically practical timeframe, suggesting its potential for clinical implementation. This pioneering work established the feasibility of bench-to-bedside translation and laid the foundation for molecular-based treatment stratification.

2. The Cancer Genome Atlas classification

The most widely appreciated and representative study for the molecular characteristics of GC was reported by The Cancer Genome Atlas (TCGA) group [30]. Through comprehensive bulk-sequencing–based multi-omic analysis, TCGA established four distinct molecular subtypes of GC, microsatellite instability (MSI), Epstein-Barr virus (EBV) associated, chromosomal instability (CIN), and genomically stable (GS) types. MSI type is associated with CpG island methylator phenotype and MLH1 silencing, one of the major gene of mismatch repair (MMR) system, causing hypermutation of tumor. EBV type is associated frequent PIK3CA mutations and up-regulated immune cell signaling including programmed death-ligand 1/2 (PD-L1/2) overexpression. Therefore, both MSI and EBV-positive GCs are recently considered as promising candidates for immune checkpoint inhibitor (ICI) therapy [31-33]. The remaining tumors are classified based on their copy number burden: those harboring extensive somatic copy number alterations are designated as CIN subtype, while the rest are categorized as GS subtype. The CIN subtype is strongly associated with intestinal-type histology and frequently harbors TP53 mutations and receptor tyrosine kinase activation. The GS subtype is characterized by diffuse-type histology and features CDH1/RHOA mutations and CLDN18-ARHGAP fusion. While the TCGA classification effectively captured the molecular heterogeneity of GC, initial prognostic correlations were limited, potentially due to the heterogeneous treatment protocols across contributing institutions.

3. Asian Cancer Research Group classification

Asian Cancer Research Group reported the results from microarray data (mRNA expression) of 300 GC tumors and they classified GC into four distinctive subtypes: MSI, microsatellite stable (MSS) with epithelial-mesenchymal transition (EMT), and MSS with and without TP53 activation [34]. The characteristics of MSI type is like the result of TCGA and showing better prognosis compared to other subtypes. EMT subtype is predominantly diffuse-type histology with worst prognosis, resembling GS type of TCGA. MSS/TP53-activated subtype encompasses most EBV-positive tumors; however, EBV-positive cases represent only a small proportion of the overall MSS/TP53-activated subtype, and MSS/TP53-inactivated is associated with TP53 mutation and is comparable to CIN type of TCGA. The prognosis of MSS/TP53-activated and -inactivated subtypes shows intermediate prognosis compared to MSI and EMT subtype. This classification demonstrated significant prognostic stratification, providing additional clinical utility.

4. Single patient classifier (nProfiler I)

An effort to translate molecular characteristics into adjuvant treatment strategies for GC was conducted by the Yonsei group [35]. Through comprehensive analysis of gene expression profiles from more than 1,000 tumor specimens, they classified GC into three distinct phenotypes: immune (IM), epithelial (EP), and stem-like (ST) types. The IM type was characterized by high GZMB/WARS expression, demonstrating favorable prognosis but limited benefit from adjuvant chemotherapy. The EP type, marked by high CDX1 expression, showed good response to adjuvant chemotherapy, while the ST type, characterized by high SFRP4 expression, was associated with poor prognosis and limited chemotherapy response.

Based on these findings, they developed a single patient classifier (SPC) that could predict both prognosis and potential benefit from adjuvant chemotherapy at patient-level. The clinical utility of this classifier was validated in a subset of the CLASSIC trial cohort, a randomized controlled study comparing surgery alone versus surgery with adjuvant chemotherapy, providing robust clinical evidence for its application [35]. The unique property of SPC is that, unlike other molecular subtyping schemes which are population-level classifications, it provides pivotal information on prognosis and responsiveness to adjuvant chemotherapy for individual patients.

The SPC was subsequently commercialized as the nProfiler I Stomach Cancer Assay and was designated as the first innovative medical technology in Korea (2019-243). Several consecutive studies have evaluated its clinical performance, including feasibility studies [23,36-38]. Currently, additional retrospective and prospective multicenter validation studies have completed enrollment and are undergoing analysis (SPRINT trial, NCT04600518; PREDICT trial, NCT04487717).

5. Convergent findings across molecular classifications of GC

The above-mentioned molecular classification systems for GC have revealed several consistent patterns in terms of prognosis and treatment response (Table 2). From a prognostic perspective, microsatellite instability–high (MSI-H) tumors, EBV-positive cases, and the IM type identified by SPC consistently demonstrate favorable outcomes compared to other subtypes within the same stage. The favorable prognosis of MSI-H tumors can be attributed to their hypermutation status resulting from MMR deficiency, which generates numerous neoantigens recognizable by tumor-infiltrating T cells. This enhanced host immune response enables effective tumor control, often resulting in presentation at earlier stages with better outcomes [39]. Similarly, EBV-positive tumors and IM types are characterized by enriched immune cell infiltration and elevated expression of immune-related genes, reflecting effective host immune surveillance system. This pattern mirrors observations in other solid tumors, where high tumor-infiltrating lymphocyte density correlates with improved outcomes [40-43].

Conversely, subtypes associated with diffuse histology, including GS, EMT, and ST types, consistently demonstrate poor prognosis. These subtypes show elevated EMT-related signaling and significant influence of cancer-associated fibroblasts (CAFs) in their tumor microenvironment [44].

Regarding chemotherapy response, tumors exhibiting intestinal-like characteristics, including the G-INT and EP type (CDX1-positive group in SPC), generally demonstrate good response to conventional chemotherapy. This pattern parallels observations in colorectal cancer (CRC), where CDX2-positive tumors show better chemotherapy response [45].

Two distinct patterns of poor chemotherapy response have emerged. First, EMT-like subtypes, characterized by elevated EMT signaling (including Wnt and transforming growth factor β pathways) and well-developed CAF infrastructure, typically show poor response. These tumors, often manifesting as diffuse type of histology or Borrmann type IV, exhibit extensive desmoplasia, which may impair drug penetration and delivery [46,47]. Second, despite their favorable prognosis, immune-enriched subtypes often show limited benefit from chemotherapy. This paradox might be explained by the low probability of microscopic residual disease due to effective immune control, and the potential negative impact of chemotherapy on tumor-controlling immune cells, possibly neutralizing their prognostic advantage [48].

This understanding of molecular subtypes and their relationship to the prognosis and treatment response provides crucial insights for treatment stratification and the development of personalized therapeutic approaches in GC management. The consistent patterns observed across different classification systems strengthen the validity of these findings and their potential clinical applications.

1. Clinical implementation of molecular characteristics in GC treatment

The rationale for omitting adjuvant chemotherapy in stage I GC provides valuable insights for treatment optimization in more advanced stages. This approach in stage I is based on excellent prognosis with surgery alone and minimal potential benefit from chemotherapy, coupled with the risk of treatment-related toxicities.

Extending this concept to stage II/III GC raises an important question: Should all patients with stage II/III disease receive adjuvant chemotherapy after D2 gastrectomy? Historical clinical trials such as CLASSIC and ACTS-GC, adopted a “one-size-fits-all” approach, conducted before our current understanding of molecular heterogeneity in GC. These trials demonstrated an overall survival benefit from adjuvant chemotherapy, but the benefit was limited to a subset of patients. Notably, approximately 50% of patients could be cured by surgery alone [18,19], suggesting that these patients may not require additional adjuvant chemotherapy. This highlights the critical need for better patient selection strategies, as many patients may be unnecessarily exposed to chemotherapy toxicity without deriving actual benefit.

2. Microsatellite instability–high

MSI-H phenotype, associated with Lynch syndrome, has been extensively studied in CRC. Multiple retrospective studies demonstrated that MSI-H CRC shows favorable prognosis and lacks benefit from 5-FU–based chemotherapy [49]. These findings were validated through retrospective analyses of samples from various randomized controlled trials (RCTs) comparing surgery alone versus surgery plus chemotherapy, and subsequent meta-analyses pooling these results [50]. Consequently, NCCN guidelines have recommended considering MSI status when making decisions about adjuvant chemotherapy in CRC [51].

In GCs, although initial retrospective studies showed conflicting results, a meta-analysis [52] pooling these studies revealed that MSI-H GC similarly demonstrates better prognosis within the same stage and limited benefit from adjuvant chemotherapy. These findings were validated in subset analyses of major RCTs including MAGIC and CLASSIC trials [53,54], comparing surgery alone versus surgery plus chemotherapy. Furthermore, an individual patient data meta-analysis of four RCT cohorts provided further confirmation of these results [55].

Given the nature of biomarker research, conducting additional RCTs specifically examining surgery alone versus surgery plus chemotherapy in MSI-H patients would be challenging. However, despite the retrospective approach, the level of evidence is considerable, as the hypothesis was validated using RCT samples [56]. Moreover, considering the well-established biology of MSI-H in solid tumors, these findings appear to have accumulated sufficient evidence for clinical implementation in GC, as is the case in CRC.

Based on these results, the recent ESMO guideline recommends that “Adjuvant (postoperative) chemotherapy should be avoided in resected MSI-H GC” [57]. Additionally, the updated CSCO guideline suggests that “Taking adverse reactions related to chemotherapy and patients’ financial implications into account, it is suggested that for dMMR/MSI-H patients, (neo)adjuvant treatments such as immunotherapy in clinical trial settings could be first considered, unless unwillingness from the patient’s side, after detailed discussion with the patient and families about the risk and benefits of different treatment strategies, postoperative observation or chemotherapy can be considered” [21]. The KGCA guidelines are expected to reflect similar considerations in their upcoming update [20,58,59]. While challenging within Korea’s healthcare environment, there is a growing need to provide patients with comprehensive information and involve them in treatment decisions [60].

3. SPC and EBV subtype

The IM type of SPC and EBV-positive GC share similar biological characteristics, particularly in terms of activated immune-related signaling pathway genes. Both subtypes demonstrate better prognosis compared to the same-stage tumors and limited benefit from chemotherapy. The performance of SPC has been validated in RCT samples (CLASSIC trial) [35], and it is currently being utilized in Korean clinical practice as a non-reimbursed test. Based on additional large-scale retrospective cohort results [23,36,37], SPC is now applicable to stage II/III GC according to TNM 8th edition criteria (It should be noted that the CLASSIC trial was conducted based on TNM 6th edition, and thus SPC was previously limited to stage II/III GC as defined by the 6th edition.). If the ongoing multicenter prospective validation study (PREDICT trial, NCT04487717) confirms these findings, SPC, like MSI status, is expected to become an additional biomarker that should be considered in guideline recommendations for determining the need for adjuvant chemotherapy after surgery.

Regarding EBV-positive GC, while its characteristics were initially confirmed in CLASSIC trial samples [38], subsequent studies have shown conflicting results [61,62]. Although the biological characteristics suggest similar outcomes, additional research and evidence will be necessary before implementing EBV status as a determinant for adjuvant chemotherapy decisions in clinical practice.

4. Clinicians’ point of views: dilemma to treat or not to treat

From a clinician’s perspective, the decision to omit adjuvant chemotherapy presents significant challenges. In the current practice environment where adjuvant chemotherapy has become standard for all stage II/III GC patients, the decision to withhold chemotherapy creates understandable anxiety about potentially exposing patients to increased risk of recurrence. Even with evidence suggesting that certain subtypes (e.g., MSI-H) show excellent prognosis without adjuvant therapy, translating population-based evidence to individual patient care creates considerable psychological burden. Clinicians may experience guilt if their patient, despite having favorable molecular features, experiences recurrence after omitting adjuvant therapy.

From the patient’s perspective, receiving ineffective chemotherapy represents a significant burden. Patients endure treatment-related toxicities without therapeutic benefit, and in cases where the disease recurs despite chemotherapy, they suffer both the adverse effects and disease progression. Additionally, this approach has broader socioeconomic implications, creating unnecessary healthcare costs and insurance burden.

However, clinical practice already presents situations where universal application of adjuvant chemotherapy is challenging, such as in cases of advanced age (e.g., over 85 years), significant comorbidities, or when patients or families express preference against chemotherapy. The varying risk levels even within the same stage (II or III) further complicate treatment decisions.

Current guidelines provide uniform recommendations across stages II and III, primarily due to the design of previous landmark clinical trials [18,19]. However, emerging evidence shows that certain molecular subtypes, such as MSI-H or immune-enriched tumors, demonstrate excellent prognosis (> 90% 5-year survival) in stage II disease, comparable to stage I outcomes [37,38,54]. These findings suggest the potential for a more refined approach that integrates both anatomical staging and molecular features, potentially leading to more personalized treatment strategies.

Taken together, the evolution toward personalized treatment approaches requires several key elements: (1) enhanced risk stratification tools incorporating both clinical and molecular features, (2) comprehensive patient communication strategies, (3) implementation of shared decision-making processes, (4) institutional support systems for personalized treatment protocols. Despite the challenges of implementing personalized treatment approaches within the situation of the Healthcare system in Korea, we must move forward with careful consideration. The key is striking a delicate balance between de-escalation of therapy based on molecular evidence and ensuring appropriate patient selection with rigorous monitoring protocols. This paradigm shift demands careful consideration of both physician concerns regarding treatment optimization and patient perspectives regarding treatment decisions, while acknowledging the practical constraints of the healthcare environment.

1. Neoadjuvant chemotherapy

In Asia, where radical gastrectomy with D2 lymphadenectomy has been well-established, adjuvant chemotherapy following surgery has been the standard treatment approach. However, there has been a growing interest in neoadjuvant chemotherapy for locally advanced GC. The recently updated PRODIGY trial demonstrated improved survival outcomes, including both disease-free and overall survival, compared with postoperative adjuvant chemotherapy [63,64]. Given the well-documented advantage of neoadjuvant chemotherapy in increasing R0 resection rates, this approach is likely to become a standard treatment option for clinically diagnosed lymph node or serosa positive disease.

However, this evolution in treatment strategy raises important considerations regarding molecular subtypes. Since the chemotherapeutic agents used in neoadjuvant and adjuvant settings are largely similar, there is concern that tumor subtypes known to be unresponsive to adjuvant chemotherapy might be disadvantaged by neoadjuvant treatment. In these cases, neoadjuvant chemotherapy could potentially worsen outcomes by delaying definitive surgery [53].

The case of MSI-H tumors warrants thoughtful consideration. While subgroup analysis from the PRODIGY trial showed comparable outcomes between adjuvant and neoadjuvant approaches, not surgery alone for MSI-H tumors [65], these findings need to be interpreted within the broader context of MSI-H tumor biology. Previous studies have shown that MSI-H tumors often demonstrate favorable outcomes with surgery alone [53-55], and their response to conventional chemotherapy may be limited [66,67]. These observations suggest that careful consideration may be needed when planning neoadjuvant chemotherapy for MSI-H tumors, and further investigation could help clarify optimal treatment strategies for this specific molecular subtype [68].

However, the situation is nuanced by the fact that MSI-H tumors are prime candidates for immunotherapy. While moderate-to-far advanced MSI-H GCs might theoretically benefit from neoadjuvant chemotherapy, such cases are relatively rare given that MSI-H tumors typically present at earlier stages. This scenario underscores the need for additional clinical trials specifically addressing the role of neoadjuvant therapy in molecularly defined subtypes.

These considerations highlight the growing complexity of treatment decision-making in the era of precision medicine and emphasize the need for molecular subtype-guided treatment strategies in both the adjuvant and neoadjuvant settings. Future clinical trials should incorporate molecular stratification to better define optimal treatment approaches for specific tumor subtypes.

2. Immune checkpoint inhibitors

The advent of ICIs targeting programmed death-1/PD-L1 has revolutionized cancer treatment, demonstrating remarkable long-term survival benefits in recurrent and metastatic solid tumors [69,70]. In GC, ICIs have already been established as first-line treatment for PD-L1–positive cases [71-73]. This success prompted investigation into their potential role as adjuvant therapy, expanding their therapeutic applications.

However, contrary to expectations, ICIs as adjuvant therapy in an unselected “one-size-fits-all” approach failed to demonstrate superiority of neoadjuvant and adjuvant chemotherapy pembrolizumab over conventional chemotherapy in GC [74]. This unexpected outcome may be attributed to several factors. Notably, tumors likely to respond to ICI, such as MSI-H or immune-enriched types (e.g., EBV or IM type of SPC), might have been adversely affected by conventional chemotherapy. In these cases, standard chemotherapy may have not only shown limited anti-tumor efficacy but potentially suppressed beneficial host immune responses [48].

In alignment with the principles of precision medicine, there is a clear need for additional clinical trials specifically targeting: PD-L1 positive tumors, MSI-H cases, and immune-enriched subtypes. Such molecularly guided trials would better evaluate the true potential of ICIs in the adjuvant setting for specific patient populations. This approach represents a more nuanced strategy that acknowledges the molecular heterogeneity of GC and the importance of patient selection in ICI response. Future studies should focus on identifying and validating predictive biomarkers to optimize patient selection for adjuvant ICI in GC.

3. ctDNA and minimally residual disease

The fundamental rationale for adjuvant chemotherapy lies in addressing potential minimally residual disease (MRD) following curative resection. A reliable method for detecting MRD would enable precise identification of patients who truly require adjuvant therapy, revolutionizing treatment decision-making.

This concept was elegantly demonstrated in CRC, where a landmark study showed that ctDNA status effectively stratified patients’ need for adjuvant therapy [75,76]. In stage II/III CRC, patients with negative ctDNA showed identical outcomes regardless of adjuvant chemotherapy administration, while the benefit of adjuvant therapy was confined to ctDNA-positive patients. This represents a paradigm shift in selecting patients for adjuvant therapy based on molecular evidence of residual disease.

Similar study designs could be applied to GC, and results from such investigations are anticipated in near future [77-79]. However, several technical challenges need to be addressed: while the potential application of MRD detection in GC holds great promise, several significant challenges must be addressed [80]. The technical limitations of ctDNA analysis in GC present a primary concern, particularly the consistently low yield of ctDNA obtained from patients. Unlike some other cancer types, GC lacks widely prevalent molecular alterations that could serve as universal targets for detection. Moreover, GC typically demonstrates lower tumor purity compared to other cancer types, particularly in diffuse-type tumors where cancer cell cellularity is notably low [81]. This inherent characteristic could result in lower ctDNA yields, further complicating detection. This molecular heterogeneity, combined with the challenges of low tumor purity, poses substantial challenges for developing standardized MRD detection methods. Overcoming these challenges may be possible through personalized approaches to MRD detection in GC. Rather than relying on universal mutation panels, analyzing individual tumor-specific somatic alterations and developing customized libraries for each patient could prove more successful for ctDNA detection. This patient-specific approach, while more resource-intensive, may better capture the molecular heterogeneity of GC. The successful implementation of such personalized ctDNA analysis could transform the current paradigm of adjuvant therapy in GC, moving from a stage-based approach to a more precise, molecularly guided treatment strategy. While this represents a significant step forward in the personalization of GC treatment, careful validation through well-designed clinical studies is still needed.

As medicine has evolved, evidence-based practice has become a fundamental and prerequisite condition of modern healthcare. Driven by advances in genomic analysis technologies, we are now moving toward personalized treatment based on individual tumor characteristics, and further advancing toward comprehensive care that considers each patient’s unique characteristics and environmental factors (Fig. 1).

Fig. 1.

Evolving decision-making in stage II/III gastric cancer: current practices and future direction. ctDNA, circulating tumor DNA; EBV, Epstein-Barr virus; IM, immune; MSI-H, microsatellite instability–high; XELOX, capecitabine plus oxaliplatin.

The field of genomic analysis has progressed so rapidly that next-generation sequencing is no longer “next generation.” We have advanced from bulk to single-cell sequencing, and now to spatial genomics and 3D spatial imaging, which capture the precise location of individual cancer cells and their interaction with neighboring cells and extracellular matrix [82]. These technological advances will deepen our understanding of cancer biology and its complexity.

GC remains a global health burden beyond Asian and Korean disease. This presents both a responsibility and an opportunity—this is our mission, and no one else will undertake it on our behalf. As Sir Isaac Newton once said, “If I have seen further, it is by standing on the shoulders of giants.” Our continued efforts will contribute to building these giants’ shoulders even higher, and in the near future, our successors will stand upon them to see even further, ultimately realizing precision cancer medicine beyond adjuvant therapy for GC. The journey toward precision medicine in GC is not just about adopting new technologies or implementing existing knowledge; it represents our commitment to advancing medical science to cure disease that significantly impacts our society.