Preclinical and Virtual Models of Mucosal Melanoma: Bridging Translational Gaps in a Rare and Lethal Cancer

ABSTRACT

Mucosal melanoma (MM) is a rare and lethal subtype of melanoma, disproportionately affecting Asian populations and exhibiting distinct clinicopathological and genetic features compared to cutaneous melanoma (CM). Often diagnosed at advanced stages, MM shows poor responses to conventional therapies, and no standardized treatment regimen currently exists. Progress in preclinical modeling, including cell lines, patient‐derived xenografts (PDXs), organoids (PDOs), and comparative animal models—has provided valuable tools for studying MM pathogenesis and therapeutic resistance. Yet these models remain limited in number, heterogeneity, and standardization, restricting their ability to capture MM's molecular diversity and immunosuppressive microenvironment. Beyond physical platforms, emerging virtual strategies—including computational simulations, artificial intelligence–driven multi‐omics integration, and in silico clinical trials—offer scalable, cost‐effective complements. By simulating tumor–immune–drug interactions using minimal biospecimens, these models offer a unique platform for hypothesis testing and patient stratification in this rare cancer type. This review summarizes MM's clinicopathological features and therapeutic challenges, evaluates current preclinical models, and highlights the synergistic integration of biological and virtual approaches. Future efforts should prioritize MM‐specific repositories, multi‐model integration, and incorporation of in silico pipelines to accelerate translational research and improve outcomes in this highly lethal malignancy.

Summary

This review highlights MM's uniqueness, preclinical models (PDOs, PDXs, and CCLs) progress, and virtual strategies' value. It addresses MM's poor therapy response, emphasizing model integration to bridge translational gaps, crucial for advancing precision medicine in this rare, lethal cancer amid pigment cell/melanoma biology challenges.

Integrated physical and virtual modeling provides a scalable framework to overcome sample scarcity, recapitulate mucosal melanoma biology, and accelerate translational discovery toward precision medicine.

Abbreviations

AI

artificial intelligence

AM

acral melanoma

BRAF

B‐raf proto‐oncogene, serine/threonine kinase

CCL

cancer cell lines

CDK4

cyclin‐dependent kinase

CM

cutaneous melanoma

CNVs

copy‐number alterations

CTLA‐4

cytotoxic T‐lymphocyte‐associated protein 4

CXCL3

C‐X‐C motif chemokine ligand 3

EGF

epidermal growth factor

EZH2

enhancer of zeste homolog 2

FOXP3

forkhead box P3

GEMM

genetically engineered mouse models

GRB2

growth factor receptor‐bound protein 2

HLA

human leukocyte antigen

HOX

homeobox protein hox

huPDX

immune‐humanized PDX

IGF‐IR

insulin‐like growth factor I receptor

MAPK

mitogen‐activated protein kinase

MHC‐1

major histocompatibility complex class I

MM

mucosal melanoma

NGFR

nerve growth factor receptor

OMM

oral mucosal melanoma

ORR

overall response rate

PBMCs

peripheral blood mononuclear cells

PDC

patient‐derived cells

PD‐L1

programmed death‐ligand 1

PDO

patient‐derived organoids

PDX

patient‐derived xenografts

PI16

peptidase inhibitor 16

PI3K

phosphoinositide 3‐kinase

RB

retinoblastoma protein

RTK

receptor tyrosine kinase

SVs

structural variants

TAMs

tumor‐associated macrophages

Teffs

effector T cells

TIGIT

T cell immunoreceptor with immunoglobulin and ITIM domain

TILs

tumor‐infiltrating lymphocytes

TMB

tumor mutational burden

Tregs

regulatory T cells

TWT

triple wild‐type

UVR

ultraviolet radiation

WGS

Whole‐Genome Sequencing

1. Introduction

MM represents a highly aggressive subtype of melanoma arising from mucosal surfaces, including the oral cavity, nasopharynx, urogenital tract, and gastrointestinal system. Although accounting for less than 2% of all melanomas, MM exhibits significantly worse clinical outcomes, with 5‐year overall survival rates consistently below 20% (Ma et al. 2021; Spencer and Mehnert 2016). This dismal prognosis stems from its insidious anatomical locations, nonspecific early symptoms, and propensity for submucosal infiltration, resulting in delayed diagnosis at advanced (T3–T4) or metastatic stages in most cases (Lerner et al. 2017).

Current therapeutic strategies—encompassing surgery, local ablation, chemotherapy, targeted therapy, and immunotherapy—yield suboptimal outcomes. Surgical resection remains the primary approach for resectable cases, yet anatomical constraints often compromise margin clearance and functional preservation. Targeted therapies, primarily for KIT‐mutated subsets, provide transient responses with tyrosine kinase inhibitors like imatinib, but resistance inevitably develops (Carvajal et al. 2011; Guo et al. 2011; Hodi et al. 2013). Immunotherapy with PD‐1 blockade demonstrates lower objective response rates in MM compared to CM (Heppt et al. 2017), and while combination strategies (e.g., dual immune checkpoint inhibition or immune‐antiangiogenic therapy) show promise (Shoushtari et al. 2020), toxicity and patient stratification remain unresolved challenges. Collectively, MM lags behind in precision medicine and combinatorial treatment development.

Advancing MM research critically requires clinically relevant preclinical models that accurately reflect its molecular drivers, immunosuppressive microenvironment, and tissue‐specific growth patterns. Conventional CM models (e.g., BRAF‐mutant cell lines or murine models) fail to mirror MM biology. Recent efforts have employed patient‐derived cell lines, PDX, organoids, genetically engineered mouse models (GEMMs), and humanized mice, offering platforms for mechanistic and therapeutic exploration. However, each model has inherent limitations, and the scarcity of standardized, biologically representative MM models persists as a critical barrier.

Beyond these experimental systems, emerging virtual strategies—including machine and deep learning, multi‐omics network analysis, mechanistic modeling, and patient‐specific in silico cell simulations—are reshaping the landscape of rare cancer research. By enabling scalable simulations of tumor–immune–drug interactions with minimal sample requirements, these computational approaches complement biological models and hold particular promise for overcoming data scarcity in MM.

Integrating physical models with computational simulations may establish a multi‐tiered preclinical framework to accelerate translational progress and optimize therapeutic discovery.

In this review, we systematically examine the clinicopathological and genomic landscape of MM, delineating the hallmarks of its immune microenvironment alongside the persistent therapeutic hurdles and bottlenecks that impede translational progress. We further evaluate existing preclinical models, highlight the potential of virtual modeling, and outline future directions to bridge biological experimentation with computational innovations, ultimately advancing precision medicine in MM.

2. Clinicopathological and Genetic Features of Mucosal Melanoma

2.1. Epidemiology

MM is a rare melanoma subtype originating from melanocytes in mucosal tissues. It constitutes only 2% of all newly diagnosed melanomas in the United States (2023 data) but accounts for 20%–30% of cases in China and other East‐Asian populations (Yang et al. 2025). Primary sites are distributed unevenly: head‐and‐neck mucosa (55%), anorectal or vaginal mucosa (24%), and urogenital mucosa (21%), with head‐and‐neck lesions being the most common (Hicks and Flaitz 2000). Marked ethnic differences exist. In Caucasian populations (e.g., North America/Europe), CM predominates, with MM constituting a minor fraction. Conversely, among East Asian populations, MM accounts for 20%–30% of all melanomas, ranking as the second most common subtype after acral lentiginous melanoma (ALM, ~40%). Disease onset spans a wide age range, with 83% of patients aged 31–70 years with a modest female preponderance (Yde et al. 2018). Unlike CM, MM incidence has remained stable over recent decades; any apparent rise is likely attributable to improved diagnostic accuracy rather than a true increase in incidence. These epidemiological patterns highlight MM's distinct biological identity compared with CM and set the stage for its unique molecular and immunological hallmarks.

2.2. Molecular and Genomic Landscape

Genomic analyses reveal marked distinctions between MM and CM. Activating mutations in BRAF and NRAS—present in 30%–52% of CM (“Genomic Classification of Cutaneous Melanoma,” Cancer Genome Atlas Network 2015), respectively—occur in only 6% and 8% of MM (Zhou et al. 2019). Conversely, KIT alterations are enriched, with activating mutations identified in ~23% of mucosal melanoma; these lesions constitutively activate the MAPK/PI3K pathways and confer sensitivity to imatinib (Beadling et al. 2008). Notably, molecular studies across melanoma subtypes reveal that MM exhibits a significantly lower tumor mutational burden (TMB) compared to CM (2.3–2.7 vs. 36.3–49.2 mut/Mb, respectively) (Hayward et al. 2017; Newell et al. 2019). Due to its distinct anatomical sites, MM is less exposed to ultraviolet radiation (UVR) and consequently lacks the UVR‐associated mutational signatures predominant in CM (Hayward et al. 2017; Newell et al. 2022, 2019). This low TMB leads to a scarcity of tumor neoantigens, which may partly explain the inferior response rates of MM to immune‐based therapies. Despite this low burden of point mutations, MM is further characterized by a high burden of structural variants (SVs) and copy‐number alterations (CNVs) (Hayward et al. 2017). A recurrent, MM‐specific complex rearrangement between chromosomes 5 and 12 is observed in 44% of tumors and correlates with adverse prognosis (Shi et al. 2022). Recurrent amplification and mutation of the transmembrane nucleoporin gene POM121 (16.9% and 15.4%, respectively; combined frequency 30.8%) have also been documented (Zhou et al. 2019). Consistent across independent cohorts are amplifications of CDK4 and CCND1 together with biallelic loss of CDKN2A, implicating dysregulated CDK4 signaling as a potential driver of MM pathogenesis (Broit et al. 2021). These findings underscore that MM is shaped by structural genome instability and CDK4‐pathway dysregulation, which have direct implications for targeted drug development and model design.

2.3. Immune Microenvironment of MM

2.3.1. Immunologically Cold Phenotype

Both CD8⁺ TILs and CD4⁺ TILs are significantly reduced in MM compared to CM. Although the absolute number of tumor‐infiltrating regulatory T cells (Tregs) is comparable between MM and CM, the proportion of FOXP3⁺ Tregs within the CD4⁺ T cell compartment is significantly higher in MM (Nakamura et al. 2020). The abundance of Tregs, coupled with the expression of inhibitory receptors, limits the tumor‐clearing capacity of CD4⁺ and CD8⁺ effector T cells (Teffs) through mechanisms such as Teff suppression and the catalysis of immunosuppressive molecules, thereby fostering an immunosuppressive microenvironment (Imianowski et al. 2025). Li et al. observed higher infiltration of CXCL3⁺ tumor‐associated macrophages (TAMs) in MM compared to AM and CM (Li, Cui, et al. 2025). Consistent with the role of CXCL3⁺ TAMs (where CXCL3 is a major chemokine for neutrophil recruitment), MM also exhibits greater neutrophil infiltration than AM and CM. Beyond their intrinsic cytotoxic function, neutrophils demonstrate immunosuppressive roles within tumors (Yu et al. 2024). Furthermore, Li et al. demonstrated that neutrophils contribute to immunosuppression in melanoma by suppressing T cell‐mediated killing (Li, Cui, et al. 2025).

2.3.2. Impaired Antigen Presentation

In most advanced‐stage MM lesions, tumor cells markedly down‐regulate MHC‐I expression (Mengoni et al. 2025). Compared with cutaneous melanocytes, normal mucosal melanocytes already exhibit low basal MHC‐I levels and high expression of neural‐crest migration genes such as SOX10, while genes critical for antigen processing and presentation (e.g., HLA‐DR and TAP1) are transcriptionally silent (Babu et al. 2025). These features collectively disrupt effective antigen presentation, rendering MM intrinsically resistant to immune‐checkpoint inhibition.

2.3.3. Heterogeneity in Immune Checkpoint Expression

Li et al. observed significantly higher CTLA4 expression on Tregs in CM compared to AM and MM (Li, Cui, et al. 2025). This differential expression may explain why anti‐CTLA4 antibodies, used clinically for melanoma treatment, are generally more effective in CM patients than in MM patients (Klemen et al. 2020). The PD‐L1 positivity rate in MM (approximately 16.7%–44%) is substantially lower than that in CM (44.7%–62%) (Kaunitz et al. 2017; Nakamura et al. 2019). Furthermore, even in PD‐L1‐positive MM patients, PD‐L1 expression levels remain low (Dodds et al. 2019). Concurrently, MM and related subtypes exhibit upregulation of co‐inhibitory receptors such as TIM‐3 and LAG‐3, which have been linked to enhanced tumor immune evasion and poor prognosis (Karlsson et al. 2020; Schatton et al. 2022; Wiecken et al. 2025).

Compared with CM, non‐cutaneous subtypes (MM and AM) exhibit hallmarks of an immunosuppressive microenvironment: sparse T‐cell infiltration, impaired antigen presentation, and a distinct checkpoint landscape (Table 1). The spatial dynamics and mechanistic interactions between these immune components and tumor cells—leading to the characteristic “cold” phenotype of MM and AM—are visually integrated in Figure 1. These shared immunosuppressive traits not only limit current immunotherapeutic efficacy but also necessitate the development of high‐fidelity preclinical models.

TABLE 1.

Comparison of immune microenvironment characteristics in MM AM and CM.

Feature dimensions	MM	AM	CM
Immune cell infiltration	Neutrophil Percentage ↑	TIGIT+Tregs Percentage ↑	Neutrophil Percentage ↓
	CD8+ T Cells Infiltration ↓	CD8+ T Cells Infiltration ↓	CD8+ Cells Infiltration↑
	CXCL3+ TAMs Infiltration ↑	PI16 + CAFs Infiltration ↑	CXCL3+ TAMs Infiltration ↓
Immune checkpoint expression levels	CTLA4 ↓	CTLA4 ↓	CTLA4 ↑
	PD‐1↓ (16.7%–44%)	PD‐1↓ (13.6%–31%)	PD‐1↑(44.7%–62%)
	TIGTI ↓		TIGTI ↑
Antigen‐presenting Molecule Expression Levels	MHC‐I ↓	MHC‐I ↓	MHC‐I ↑
TMB levels	↓ (2.3–2.7 mutation/MB)	↓ (3.5–5 mutation/MB)	↑ (36.3–49.2 mutation/MB)

FIGURE 1.

Comparison of biological characteristics of some melanoma subtypes. At the levels of both tumor microenvironment and tumor cell, the three subtypes of melanoma exhibit distinct features. In MM, there is an increased proportion of neutrophils, FOXP3⁺Tregs, and CXCL3⁺ TAMs. In contrast, AM shows elevated levels of PI16⁺ CAFs and TIGIT⁺ Tregs. Across different subtypes, immune‐related proteins are differentially expressed, and each subtype is characterized by specific genetic alterations and pathway activities (created with Biorender.com). BRAF, B‐Raf proto‐oncogene, serine/threonine kinase; CDK4, cyclin‐dependent kinase; CXCL3, C‐X‐C motif chemokine ligand 3; EGF, epidermal growth factor; FOXP3, forkhead box P3; GRB2, growth factor receptor‐bound protein 2; HLA, human leukocyte antigen; HOX, homeobox protein hox; IGF‐IR, insulin‐like growth factor 1 receptor; MHC‐1, major histocompatibility complex class I; PD‐L1, programmed death‐ligand 1; PI16, peptidase inhibitor 16; PI3K, phosphoinositide 3‐kinase; RB, retinoblastoma protein; SV, structural variations; TAMs, tumor‐associated macrophages; TIGIT, T cell immunoreceptor with immunoglobulin and ITIM domain; TMB, tumor mutational burden; Tregs, regulatory T cells.

2.4. Implications for Translational Modeling

The distinct clinicopathological, genetic, and immunological features of MM impose specific requirements and challenges for preclinical model development. The pronounced ethnic variability in incidence underscores the necessity of building ethnically diverse patient‐derived resources to ensure global relevance (Mao et al. 2021). Similarly, the enrichment of KIT mutations and CDK4/CCND1 amplifications, alongside the scarcity of BRAF/NRAS alterations, indicates that models derived from CM are inadequate surrogates for MM research (Jung and Johnson 2022; Wang, Banik, et al. 2022).

The immunologically “cold” phenotype—characterized by sparse effector T‐cell infiltration, elevated proportions of FOXP3⁺ regulatory T cells, CXCL3⁺ TAM and neutrophil recruitment, and impaired MHC‐I antigen presentation—necessitates models that can recapitulate the immunosuppressive tumor microenvironment. Moreover, the combination of low TMB and heterogeneous immune checkpoint expression provides further rationale for developing platforms capable of testing novel immune‐modulating agents beyond PD‐1/CTLA‐4 blockade (Liu et al. 2022).

Collectively, these features emphasize that conventional melanoma models are insufficient to capture the biological complexity of MM. Future translational progress will depend on model systems that integrate molecular drivers, tumor–immune interactions, and ethnic diversity, thereby enabling more predictive therapeutic evaluation.

3. Current Therapeutic Landscape and Translation Challenges

MM poses significant therapeutic challenges. Owing to its distinct anatomical locations, patients are often diagnosed at advanced stages (T3–T4) with highly aggressive behavior (Lerner et al. 2017). Current treatment strategies primarily include surgical resection, local ablation, chemotherapy, targeted therapy, immunotherapy, and emerging combination therapies. However, these approaches collectively yield suboptimal outcomes, highlighting an urgent need for sustained advancement in basic research and systematic support from preclinical models. The following sections systematically review the limitations and future translational directions of each therapeutic modality.

3.1. Surgical Resection and Local Therapies

For early‐stage MM, surgical resection remains the primary treatment modality. Unlike CM, surgery for MM frequently arises in anatomically complex regions (e.g., oral cavity, sinonasal tract, or genitourinary mucosa). This creates a dilemma between achieving radical resection and preserving essential functions such as speech, swallowing, or reproduction. The procedure must balance radical resection with functional preservation, emphasizing adequate excision margins to achieve negative mucosal margins (Bachar et al. 2008). Cryotherapy has been utilized for decades in MM management due to the marked sensitivity of melanocytes to cryoablation, which contributes to its therapeutic efficacy (Wu et al. 2025). In select cases of advanced disease, cryotherapy may serve as a palliative cytoreductive measure to prolong survival while enhancing quality of life. However, both surgical resection and cryotherapy face the limitation of incomplete clearance, and postoperative recurrence remains a major risk—particularly in patients with advanced tumors, inadequate margins, or lymphovascular invasion.

3.2. Targeted Therapies

For recurrent and metastatic MM, combination therapy with targeted agents and immunotherapy is typically required. Phase II clinical trials have established Imatinib as a viable therapeutic option for mucosal melanoma (MM). As a KIT receptor tyrosine kinase inhibitor, it demonstrated an overall response rate (ORR) of approximately 22%–30% in patients harboring KIT mutations or amplifications (Guo et al. 2011; Hodi et al. 2013). Subsequent studies identified ponatinib as a more potent inhibitor than imatinib, providing an alternative therapeutic strategy (Han et al. 2019). Patients harboring CDK4 gene mutations may be treated with the CDK4/6 inhibitor dalpiciclib (Shi et al. 2022). However, the therapeutic benefits of targeted agents are often transient, as resistance rapidly develops and responses vary substantially across patients. This reflects the profound genomic complexity of MM, where the co‐occurrence of multiple oncogenic drivers limits the durability of single‐agent approaches. Crucially, evidence suggests that the clinical efficacy of targeting a primary driver can be undermined by concurrent alterations; for instance, while KIT mutations typically predict imatinib sensitivity, the simultaneous presence of CDK4 or CCND1 amplifications has been shown to mediate primary resistance and correlate with significantly poorer survival (Zhou et al. 2019). To address such complexities, pretreatment comprehensive genomic profiling via whole‐exome sequencing (WES) or whole‐genome sequencing (WGS) is essential to identify these high‐risk mutational signatures and guide individualized therapy. Furthermore, the development of robust MM‐specific preclinical models is required to systematically evaluate these specific mutational combinations and validate rational, synergistic drug pairings, thereby bridging molecular complexity with clinical applicability.

3.3. Immunotherapy

While immune checkpoint inhibitors (ICIs) demonstrate significant efficacy in CM, their therapeutic benefit in MM remains limited. Monotherapy with PD‐1/PD‐L1 inhibitors yields an ORR of merely 15%–20%, with a median progression‐free survival (mPFS) of 4.33 months and a 12‐month PFS rate of 28.3% (D'Angelo et al. 2017; Uhara et al. 2021). The median overall survival (mOS) is 18.3 months, with a 12‐month OS rate of 64.0% (Dimitriou et al. 2022; Rose et al. 2021). Combination immunotherapy, such as PD‐1 plus CTLA‐4 blockade, can raise ORR to around 30%, but this improvement comes at the cost of substantial toxicity, with over half of patients developing grade ≥ 3 immune‐related adverse events (Nakamura et al. 2021). This stark trade‐off illustrates the core dilemma in MM immunotherapy: limited efficacy with monotherapy versus prohibitive toxicity with aggressive combinations. The integration of ICIs with anti‐angiogenic agents has emerged as a pivotal therapeutic strategy. In a phase II trial, atezolizumab plus bevacizumab achieved an ORR of 45% and mPFS of 8.2 months, demonstrating pronounced clinical benefit in NRAS‐mutant patients, likely due to VEGFA/FLT4 overexpression (Dai et al. 2025; Mao et al. 2016). Nevertheless, the modest efficacy of ICIs overall, combined with high toxicity and heterogeneous responses, underscores the urgent need for advanced preclinical and computational models. Such systems are required to dissect the immunosuppressive microenvironment of MM, predict patient‐specific responses, and guide rational combinations beyond empirical trial‐and‐error.

3.4. Chemotherapy and Conventional Systemic Approaches

Conventional chemotherapy in MM is primarily reserved for advanced stages or salvage settings after failure of immune/targeted therapies. A phase III trial involving 204 surgically resected MM patients evaluated adjuvant chemotherapy, comparing temozolomide plus cisplatin versus interferon α‐2b administered over 1 year. Preliminary evidence indicated improved relapse‐free and overall survival with chemotherapy at a median follow‐up of 24 months (Weber et al. 2023). Nevertheless, the therapeutic impact of chemotherapy in MM remains modest, and it should not be considered a standard frontline treatment. Its role is largely restricted to postoperative adjuvant settings or to patient ineligible for targeted or immune‐based approaches (Tang et al. 2022). When radiotherapy is considered for MM, it is most commonly delivered as an adjuvant treatment after surgical resection. A large retrospective cohort study showed that the addition of postoperative radiotherapy improved local control by 26% (Benlyazid et al. 2010; Caspers et al. 2018; Dréno et al. 2017), and additional investigations have reported relative reductions in local‐recurrence risk of up to 45% (Wushou et al. 2015). Nevertheless, a growing body of evidence, including several systematic reviews and meta‐analyses, indicates that adjuvant radiotherapy may fail to translate local control into an overall survival benefit, and in some contexts, might even correlate with suboptimal long‐term outcomes (Li et al. 2015; Owens et al. 2003). This dual reality‐limited efficacy yet selective utility‐highlights chemotherapy's continuing relevance as a benchmark or control arm in translational studies. Reliable preclinical models remain essential to clarify its comparative role and to identify rational combinations that could enhance its otherwise limited benefit.

3.5. Emerging and Combination Therapies

Beyond standard approaches, multiple investigational regimens are being evaluated for MM. Promising strategies focus on synergistic combinations, such as the pairing of PD‐1 inhibition with anti‐angiogenesis. For instance, toripalimab (anti‐PD‐1) plus axitinib (a VEGFR‐1/2/3 inhibitor) has shown clinical activity by potentially normalizing tumor vasculature and enhancing T‐cell infiltration (Zhang et al. 2021). Similarly, apatinib (a selective VEGFR2 inhibitor) or ICIs combined with the alkylating agent temozolomide are being explored to exploit the potential of chemotherapy to induce immunogenic cell death and deplete immunosuppressive cells. Furthermore, radiotherapy plus ICI is being investigated specifically for head‐and‐neck primary disease, leveraging the abscopal effect where localized radiation triggers a systemic immune response that enhances the efficacy of immunotherapy (Sergi et al. 2023). Melanoma brain metastasis (MBM), one of the most lethal complications, develops in 19.5% of MM patients within 5 years (Wang, Lian, et al. 2022). CTLA‐4 and PD‐1 are critical checkpoints in MBM progression, and dual blockade significantly outperforms either agent alone (Wang, Feng, and Liu 2025). Nevertheless, these therapeutic advances remain constrained by small cohort sizes, heterogeneous responses, and substantial toxicity. This paradox of encouraging signals but limited evidence highlights the urgent need for robust preclinical and computational models to predict patient‐specific benefit, optimize combination strategies, and guide rational trial design.

3.6. Translational Bottlenecks and the Need for Advanced Modeling

To illustrate these therapeutic advances and their translational gaps, Table 2 summarizes the major treatment approaches in MM, their representative outcomes, limitations, and the corresponding model requirements.

TABLE 2.

Therapeutic Approaches in MM—Advances, limitations, and model needs.

Therapeutic approach	Key advances	Representative outcomes	Limitations/Bottlenecks	Clinical evidence level	Translational/Model needs
Surgery/Local therapy	Radical resection; cryotherapy as adjunct	Margin‐negative resection improves RFS; cryo improves QoL	Functional impairment; high recurrence after incomplete clearance	Retrospective studies, institutional series	PDO/PDX capturing local recurrence; in silico models simulating anatomical constraints
Targeted therapy	KIT inhibitors (imatinib, ponatinib); CDK4/6 inhibitors	ORR ~30% (KIT+); variable benefit for CDK4/6 blockade	Rapid resistance; heterogeneous mutational profiles	Phase II trials	Molecularly annotated PDX/PDO; computational models predicting resistance
Immunotherapy	PD‐1/PD‐L1 ± CTLA‐4; PD‐1 + anti‐VEGF	ORR 15%–20% (monotherapy); 30% (dual blockade); 45% (PD‐1+VEGF)	Limited efficacy; high toxicity with combinations; response heterogeneity	Phase II–III trials	Immunocompetent models, humanized mice; virtual simulations of immune‐escape
Chemotherapy	Temozolomide + cisplatin improved RFS/OS vs. interferon	Improved 2‐year RFS; OS benefit in adjuvant setting	Not standard; modest efficacy; limited durability	Phase III trial (China)	Benchmark arm in PDC/PDX studies; models for rational combination
Emerging therapies	Toripalimab + axitinib; ICI+ apatinib+ TMZ; RT + ICI; dual blockade for MBM	ORR up to 45%; survival benefit in MBM with dual blockade	Small cohorts; toxicity; limited reproducibility	Early‐phase trials	Integrative models combining immune and angiogenic pathways; in silico modeling to predict subgroup benefit

Despite progress in surgery, targeted therapy, and immunotherapy, outcomes for MM remain dismal. According to treatment guidelines for MM, enrollment in clinical trials is recommended when eligible studies are available. A search of the WHO International Clinical Trials Registry Platform (ICTRP, http://apps.who.int/trialsearch/) for registered MM trials from its inception until August 25, 2025, identified a total of 113 trials. Of these, 103 were interventional studies and 10 were observational. Recent trials have increasingly focused on combination drug therapies. Table 3 summarizes clinical trials related to MM registered in the past 3 years. A key obstacle lies in the lack of preclinical systems that faithfully reproduce MM's molecule lar heterogeneity, IME, and mucosa‐specific growth patterns. Existing CM–derived platforms inadequately model MM biology, while the scarcity of MM‐specific cell lines, PDXs, and organoids restricts reproducibility and drug testing. Immunocompetent, mucosa‐relevant animal models are also underdeveloped, limiting mechanistic insights into tumor–immune interactions and resistance.

TABLE 3.

Registered MM clinical trials in the past 3 years.

Main ID	Country	Study type	Sample size	Phase	Intervention(s)	Time
NCT07076550	Australia	non‐RCT	50	I	Drug: [225Ac] Ac‐A9‐3408	2025‐06‐27
NCT06999980	Australia	RCT	297	II	Nivolumab, Relatlimab Ipilimumab	2025‐05‐23
NCT06797297	China	RCT	180	II	Biological: Pembrolizumab or IBI363	2025‐01‐22
ChiCTR‐2400093001	China	non‐RCT	47	II	Chemoradiotherapy with immunotherapy	2024‐11‐27
ChiCTR‐2400090453	China	/	/	II	Cadunilumab	2024‐09‐30
ChiCTR‐2400089885	China	Observational study	200	NA	NA	2024‐09‐19
CTIS2023‐509451‐14‐00	Australia	non‐RCT	76	I/II	Drug:DYP688	2024‐06‐27
CTIS2024‐513027‐16‐00	France	non‐RCT	60	II	Drug: LENVIMA	2024‐06‐07
NCT06424626	China	non‐RCT	60	I	Drug: AK112+Axitinib/AK104+Axitinib	2024‐05‐16
NCT06319196	Canada	non‐RCT	54	II	Nivolumab, Opdualag	2024‐02‐14
ChiCTR2400079387	China	non‐RCT	20	NA	DNV3, Toripalimab Combined chemotherapy	2024‐01‐02
ChiCTR2300078151	China	non‐RCT	40	IV	Adebrelimab, Bevacizumab	2023‐11‐29
NCT06041724	China	non‐RCT	46	II	Envafolimab combined recombinant human endostatin and first‐line chemotherapy	2023‐09‐12
ChiCTR2300073726	China	non‐RCT	30	II	Dalpiciclib combined with Camrelizumab	2023‐07‐19
NCT05661955	China	non‐RCT	202	I/II	Drug: BGB‐A445, Tislelizumab	2022‐12‐15
NCT05655312	America	non‐RCT	264	I/II	Drug: [203Pb] VMT01 [212Pb] VMT01 Nivolumab	2022‐11‐15
NCT05628883	America	non‐RCT	17	I	Biological: TBio‐4101 Drug: Interleukin‐2, Cyclophosphamide, Fludarabine	2022‐11‐14
KCT0007745	Korea	non‐RCT	30	II	Pembrolizumab, Vactosertib	2022‐09‐29
NCT05545969	Australia	non‐RCT	NA	II	Lenvatinib, Pembrolizumab	2022‐09‐06
NCT05482074	America	non‐RCT	NA	II	Drug: Olaparib	2022‐07‐27
NCT05436990	Korea	non‐RCT	14	II	Drug: Pembrolizumab, Vactosertib	2022‐06‐16
JPRN‐UMIN000048036	Japan	NA	25	NA	NA	2022‐06‐13
NCT05415072	Australia	non‐RCT	66	I/II	Drug: DYP688	2022‐06‐08
NCT05420324	China	non‐RCT	20	II	Drug: Pembrolizumab, YH003, albumin paclitaxel	2022‐06‐07
NCT05384496	America	non‐RCT	20	II	Radiation, Drug: Axitinib, Ni‐volumab or Ipilimumab	2022‐05‐17
NCT05341349	America	non‐RCT	1	I	Radiation, Axitinib, Nivolumab or Ipilimumab, Pembrolizuma	2022‐04‐11

To address these bottlenecks, emerging virtual strategies—including in silico simulations, multi‐omics–driven machine learning, and patient‐specific mechanistic modeling—offer scalable and cost‐effective tools. By simulating tumor–immune–drug dynamics with limited biospecimens, these approaches complement physical models and provide a multi‐tiered framework for translational research.

4. Preclinical Models of Mucosal Melanoma

The establishment of robust preclinical models for MM remains a major challenge owing to its clinical rarity, limited tissue availability, anatomical complexity, and immunologically cold phenotype. Unlike CM, MM lacks abundant cell line resources and standardized modeling protocols, hampering mechanistic studies and translational progress. Nevertheless, significant advances have been achieved. Recent platforms now encompass PDOs, xenografts (PDXs), patient‐derived cells (PDCs), and cancer cell lines (CCLs), complemented by canine and zebrafish models that provide comparative and high‐throughput insights.

Historically, progress has followed a stepwise trajectory: the first oral MM cell line in the 1980s (Tagawa et al. 1981), the introduction of genetically engineered zebrafish in the 2000s (Patton et al. 2005), and the maturation of organoid culture systems by the 2020s (Sun et al. 2023). More recently, immune‐humanized and refined murine models have expanded the scope of MM research, enabling deeper investigation of therapeutic resistance and metastatic biology. Collectively, these advances mark a transition from simple in vitro cultures to integrated, multi‐scale modeling frameworks that now underpin the evaluation of targeted, immune, and combination therapies (Figure 2).

FIGURE 2.

This timeline delineates key advancements in melanoma research from 1981 to 2025, encompassing the establishment of cell lines, applications of preclinical models (e.g., PDX), exploration of genetic and therapeutic targets (such as TP53 and PD‐L1), and efficacy evaluations of therapeutic agents (including CAR‐T therapies and PD‐1 inhibitors). It illustrates the translational research trajectory from basic scientific discoveries to clinical applications (created with Biorender.com). CAR‐T, chimeric antigen receptor T‐cell immunotherapy; CCND1, cyclin D1; CDK4, cyclin‐dependent kinase 4; EZH2, enhancer of zeste homolog 2; MM, multiple myeloma; PDC, patient—derived circulating tumor cells; PD‐L1, programmed death ligand; PDO, patient‐derived organoids; PDX, patient—derived xenograft; POM121, pore membrane protein 121; PTEN, phosphatase and tensin homolog; RTK, receptor tyrosine kinase.

4.1. PDO Models

PDOs are three‐dimensional (3D) miniature organoid models established through in vitro culture techniques using patient tumor tissue or liquid biopsy samples (Yang et al. 2023). They effectively preserve the 3D architecture, cellular heterogeneity, genetic features (e.g., mutations and copy number variations), and key components of the TME—including stromal and immune cells—of the original human tumors, representing a major advance in cancer research. By accurately recapitulating human tumor biology, PDOs also circumvent certain ethical and physiological limitations associated with in vivo models. Consequently, they are widely used for drug sensitivity testing and predicting patient responses to chemotherapy, targeted therapy, and immunotherapy (Tong et al. 2024). The US FDA Modernization Act 2.0 (2022) officially recognized organoid models as novel alternative methods in drug development, marking a regulatory milestone that underscores their scientific importance (Han 2023). Numerous studies have successfully generated PDOs from various human cancers, including breast, colon, kidney, ovarian, pancreatic, and liver cancers (Li et al. 2022). In melanoma, PDOs have emerged as a critical platform for optimizing targeted‐immunotherapy combinations due to their applicability to scarce clinical samples and high fidelity in modeling the immune microenvironment, though successfully established MM PDO models remain limited. In 2022, Shi et al. reported a 3D organoid model of MM along with its culture methodology and applications (Shi, Gu, et al. 2024). In 2023, Lingling Ou et al. established melanoma PDOs (MPDOs) using two innovative culture methods—collagen gel embedding and Matrigel encapsulation—providing a valuable reference for MM research (Ou et al. 2023). For therapeutic screening, Letizia Porcelli et al. utilized BRAF (V600E) mutant and BRAF (V600E/K601Q) PDO models to investigate targeted therapy, demonstrating that Nirogacestat inhibits Notch signaling and enhances antitumor responses, thereby enabling PDO‐based efficacy evaluation in rare metastatic melanomas (Porcelli et al. 2022). While these models were primarily derived from cutaneous rather than mucosal origins, they establish the essential technical benchmarks and methodological proof‐of‐concept necessary to advance organoid‐based research in mucosal melanoma. In 2023, Sun et al. established OMM (oral mucosal melanoma) organoids, performed molecular characterization and drug screening, and identified significant upregulation of receptor tyrosine kinase (RTK) signaling—particularly high NGFR (nerve growth factor receptor) expression in anti‐PD‐1‐resistant organoids. Combining anti‐PD‐1 with anlotinib or NGFR knockdown enhanced CD8+ T cell cytotoxicity, promoted IFN‐γ and TNFα secretion, and reduced tumor cell survival (Sun et al. 2023). Future PDO research should focus on improving establishment efficiency and success rates (currently ranging from 31% to 90%), reducing costs, and advancing personalized therapy (Wensink et al. 2021).

In recent years, promoted by institutions such as the NIH and aligned with the “3Rs” (Replacement, Reduction, and Refinement) principles for animal research, ethical and application standards for animal models have become more stringent. The NIH has explicitly advocated reducing reliance on traditional animal experiments and prioritizing high‐fidelity human tissue‐derived in vitro models. Among these, PDOs are becoming central tools for modeling solid tumors due to their advantages in preserving tumor heterogeneity, structural integrity, and functional characteristics. For MM, PDOs overcome limitations such as tissue scarcity and xenograft rejection, proving particularly suitable for drug screening, resistance mechanism studies, and immune co‐culture systems. Future efforts should promote the development of MM‐specific PDO biobanks, integrate emerging technologies like single‐cell omics and spatial transcriptomics, and construct more representative and scalable in vitro modeling platforms for MM. These advances will align with global trends in research ethics and accelerate translation toward precision medicine.

4.2. PDX Models

The PDX model is a preclinical platform established by directly transplanting tumor tissue from MM patients into immunodeficient animals (e.g., mice and zebrafish) (Hidalgo et al. 2014; Kim et al. 2009). This model retains key genomic characteristics—including mutation profiles and gene expression patterns—as well as the pathological heterogeneity of the original tumor, enabling faithful recapitulation of tumor behavior and drug response. By preserving these features, PDX models demonstrate 80%–90% accuracy in predicting drug efficacy and can model processes such as recurrence, metastasis, and drug resistance, providing a valuable tool for precision oncology and individualized therapy screening (Blanchard et al. 2025). While early PDX development was constrained by engraftment efficiency, contemporary refinements in transplantation techniques and genomic characterization have solidified its role as a cornerstone of precision oncology, providing a high‐fidelity surrogate for individualized therapy screening in rare malignancies like MM.

In MM, PDX models have played an instrumental role in uncovering disease mechanisms and developing personalized therapies. In 2019, Han et al. used KIT‐mutant melanoma PDX models to demonstrate that ponatinib had superior efficacy compared to imatinib, suggesting a new therapeutic alternative (Han et al. 2019). Shi et al. identified CDK4 amplification as a common genetic event in nearly half of MM cases, highlighting CDK4 as a promising target (Shi et al. 2021). That same year, Zhou et al. performed WGS on 65 MM samples and linked genetic alterations in POM121 to elevated proliferation and poor prognosis (Zhou et al. 2019). They further employed PDX models to validate the efficacy of palbociclib in CDK4‐amplified MM. Subsequent studies by Shi et al. and Xu et al. confirmed the anti‐tumor activity of other CDK4/6 inhibitors, including dalpiciclib and PD‐0332991, in MM‐PDX models (Shi, Ju, et al. 2024; Xu et al. 2019). In 2023, Shi et al. reported that an antibody‐drug conjugate targeting MUC18 (AMT‐253) exhibited potent anti‐tumor effects in MM‐PDX models, particularly when combined with anti‐angiogenic agents (Shi et al. 2023). In 2025, Zhang et al. developed CAR‐T cells targeting MUC18, which induced tumor regression in PDX models without notable toxicity, revealing a promising immunotherapeutic strategy for MM (Zhang et al. 2025). Furthermore, PDX models have supported the identification of new therapeutic targets such as EZH2 (Du et al. 2024) and enabled noninvasive prediction of treatment response—for instance, through advanced MRI metrics to assess the effects of c‐KIT inhibitors in sinonasal MM (Wang, Niu, et al. 2025).

A major limitation of conventional PDX models is the lack of a human immune microenvironment, restricting their utility in immunotherapy research. To address this, immune‐humanized PDX (huPDX) models have been developed by reconstituting immunodeficient mice with human immune cells or tissues. In the field of melanoma, these models have proven critical for evaluating novel immunotherapies. In the context of melanoma, huPDX models have provided critical insights into immune escape and therapeutic response. For instance, Jespersen et al. established a sophisticated huPDX platform by co‐engrafting autologous TILs and PDX tumors, which successfully recapitulated clinical responses to anti‐PD‐1 therapy (Jespersen et al. 2017). Similarly, Ny et al. used immune‐reconstituted NOG mice to simulate the human immune landscape and evaluate therapeutic efficacy (Ny et al. 2020). Furthermore, huPDX systems have been used to uncover novel therapeutic vulnerabilities; for example, Vendramin et al. utilized humanized melanoma models to demonstrate that mitoribosome‐targeting antibiotics can overcome immunotherapy resistance.

Other challenges of PDX models include low engraftment rates, long latency, and high costs, which limit large‐scale drug screening. Ongoing improvements involve Mini‐PDX platforms (Zhang et al. 2018), partial resection for rapid tumor regrowth (Liu et al. 2020), and the use of zebrafish PDX models as a cost‐effective alternative (Xiao et al. 2020). However, the loss of tumor heterogeneity during engraftment remains a significant concern; for example, the selective pressure of the murine environment can lead to clonal selection and the replacement of human stromal cells with murine counterparts, potentially skewing the molecular profile and drug response of the original malignancy (Karnik et al. 2023).

Future directions include integrating artificial intelligence (AI) with PDX models to accelerate data interpretation, shorten drug development timelines, and enable real‐time monitoring of tumor dynamics via imaging and computational algorithms (Blanchard et al. 2025).

4.3. CCL Models

CCL models are immortalized cell models established from patient tumor tissues through optimized two‐dimensional culture conditions. These models can be utilized in vitro or in vivo to simulate tumor growth, metastasis, and drug response. CCLs retain some molecular features of the primary tumor and offer advantages such as relatively low cost, short experimental cycles, straightforward establishment and handling, and controllable timelines, making them suitable for large‐scale preliminary drug screening (Viegas and Sarmento 2024). However, significant limitations are apparent: many cells undergo early senescence or growth arrest during passaging, and long‐term in vitro culture often leads to loss of original tumor heterogeneity—including cellular subpopulations and spatial architecture—thereby failing to reflect the complexity of patient tumors. Although more advanced preclinical models have been developed, tumor cell lines remain widely used in preclinical research.

In MM, the scarcity and limited availability of biopsy samples pose challenges for isolating MM cells. Additionally, most cultures exhibit early senescence or proliferative arrest during passaging. Consequently, reports on MM‐CCLs are relatively scarce. According to Cellosaurus, the world's largest cell line database, only 32 MM cell lines had been reported as of June 2025 (https://web.expasy.org/cellosaurus/) (Table 4). Representative examples include the NM78‐AM and NM78‐MM cell lines established by Taka Nakahara et al. in 2010 from buccal MM, which were used for drug screening (Nakahara et al. 2010). In 2013, Lourenço et al. successfully generated one OMM cell line, designated MEMO, from a hard palate tumor (Lourenço et al. 2013). In 2016, Zhang Yan et al. patented a method for establishing COMM‐1, a cell line derived from Chinese oral melanoma, and used it to enrich cancer stem cells, providing a valuable resource for studying the characteristics and therapies of oral melanoma in this population (Lin et al. 2022). In 2022, Shi et al. systematically reported the first triple‐wild‐type (TWT) MM‐CCL, named MM9H‐1, derived from a female Chinese Han OMM patient (Shi et al. 2022). The MM9H‐1 cell line exhibits stable spindle‐shaped morphology across passages, contains abundant melanin, and expresses melanoma markers such as HMB45 and Melan‐A. It maintains proliferative, migratory, and tumorigenic capabilities beyond 10 passages, forms spheroids in vitro, and induces tumor growth and metastasis in mice. Using this cell line, high‐throughput drug screening identified the proteasome inhibitor bortezomib as a potential therapeutic agent for TWT MM. The establishment and application of MM9H‐1 provide a readily adoptable cellular tool for future MM research.

TABLE 4.

Mucosal melanoma cell lines.

Cell line name	Species	PMID	Research content	Time	Country
HMG	Human	6807872	A melanoma cell line was established from a case of gingival melanoma (Tagawa et al. 1981).	1981	Japan
HM 162	Human	1882124	A hard palate MM‐CCL was derived from a lymph node metastasis (Yoshikawa and Sakuda 1991).	1991	Japan
SSM‐1	Human	8583160	A MM‐CCL was established to investigate the efficacy of radiotherapy in human OMM. The study revealed that the SSM‐1 OMM‐ cell line exhibited higher radiosensitivity compared to CM cell lines (Mimura 1995).	1995	Japan
ME	Human	11287286	Establishment of a new cell line from oral palatal gingival MM (Chang et al. 2001).	2001	China
Bear3	Dog	12944992	Overexpression of FasL effectively induces apoptosis in Fas(+) melanoma cells and supports the notion that priming immune effector cells with apoptotic tumor cells may enhance antitumor responses (Bianco et al. 2003).	2003	America
MMG1	Human	15160996 15467732	1. Overexpressed wild‐type BRAF drives melanoma cell growth. 2. Patient CTLs showed low IFN‐γ across peptides, indicating immunosuppression in metastatic melanoma (Akiyama et al. 2004; Tanami et al. 2004).	2004	Japan
Ma‐Mel‐76	Human	16827748	Proposing a model in which TGF‐β signaling drives the shift from low‐ to high‐metastatic melanoma by upregulating genes for angiogenesis, matrix remodeling, and Wnt inhibitors while suppressing Wnt target genes (Hoek et al. 2006).	2006	Switzerland
Ma‐Mel‐152aI	Human	/	/	2006	/
CML‐10	Dog	15915147	Within the context of a phase II clinical trial utilizing spontaneous canine melanoma, the efficacy of a xenogeneic vaccination strategy targeting the human melanoma differentiation hgp100 was evaluated (Itakura et al. 2005).	2006	Japan
SMYM‐PRGP	Human	17488338	New acral melanoma cell line suggested cyclin D1 overexpression acts as a facilitator rather than directly driving cell cycle (Murata et al. 2007).	2007	Japan
M40	Human	19047099	KIT aberrations in MM cells were analyzed; imatinib's effects were tested via viability and apoptosis assays (Jiang et al. 2008).	2008	America
Mel‐18 Mel‐2	Human	19035443 22578220	Pathologically activated KIT was confirmed in metastatic AM/MM, suggesting sunitinib's potential benefit (Ashida et al. 2009; Furney et al. 2012).	2009	Japan
SM2‐1 SM3	Human	19015443 22578220	Experiments demonstrated pathological KIT activation in a substantial proportion of metastatic AM and MM and indicated that sunitinib offers potential therapeutic benefit for these tumors (Furney et al. 2012; Webb and Chang 2008).	2009	America
NM78‐AM NM78‐MM	Human	20590915	Novel cell lines from buccal melanoma: NM78‐AM (cluster‐forming, non‐pigmented) and NM78‐MM (adherent, pigmented); drug sensitivity was tested (Nakahara et al. 2010).	2010	Japan
LAU‐T1257A LAU‐T1257C	Human	20862285 20862275	Mutational analysis of CT‐X MAGE genes (MAGEA1, A4, C1, C2, E1) was performed on cell lines and patient blood samples (Caballero et al. 2010; Stamatopoulos et al. 2010).	2010	America
YUHOIN	Human	21543894	miRNA expression differs by melanoma subtype and is influenced by genetic variation, contributing to biological diversity (Chan et al. 2011).	2011	America
YUSAN	Human	22842228	CM had more UV‐like C>T mutations; novel recurrent mutations in PPP6C and RAC1 were identified (Krauthammer et al. 2012).	2012	America
MEMO	Human	23249835	Cell lines were established from MM on the hard palate and maxillary alveolar ridge (Lourenço et al. 2013).	2013	Brazil
M990203	Human	24581590	Metastatic melanoma cells promoted macrophage IL‐1β production despite not secreting it themselves, via mechanisms including deficient inflammasome expression and DNA methylation (Gehrke et al. 2014).	2014	Switzerland
COMM‐1	Human	36004461	Two novel Chinese oral mucosal melanoma (COMM) cell lines were established to study heterogeneity loss in long‐term culture and its link to metastasis (Lin et al. 2022).	2016	China
MEL1	Human	/	/	2016	/
M140325	Human	/	/	2021	/
Dog‐OralMel‐18249	Dog	35053440	Integrated genomics/transcriptomics of 32 canine MMs revealed two subtypes: low‐SV tumors (immune‐rich microenvironment) and high‐SV tumors (pigmentation/oncogene pathways, e.g., TERT). Genes MDM2, CDK4, TRPM7, GABPB1, and SPPL2A were proposed as novel candidate oncogenes for human MM (Prouteau et al. 2022).	2022	France
Dog‐OralMel‐18333	Dog	35053440		2022
Dog‐OralMel‐18395	Dog	35053440		2022
Dog‐OralMel‐18657	Dog	35053440		2022
Dog‐OralMel‐18848	Dog	35053440		2022
MM9H‐1	Human	35666052	A novel triple‐wild‐type (TWT) mucosal cell line (MM9H‐1) without common driver mutations was established as a unique preclinical model (Shi et al. 2022).	2022	China

Beyond high‐throughput drug screening, CCLs are also widely used to investigate other disease characteristics, including proliferative, migratory, and invasive capacities, as well as molecular profiles. For instance, Ma et al. performed miRNA microarray analyses on MM tissues and three MM cell lines, revealing consistent downregulation of miR‐23a‐3p in all samples and cell lines (Ma et al. 2019). Its downregulation was associated with poor prognosis in MM, mediated through suppression of the cAMP and MAPK pathways via ADCY1, offering novel insights for therapeutic development.

The search results from the Cellosaurus database for MM cell lines are summarized chronologically in Table 4.

4.4. PDC Models

The PDC model uses cells obtained by culturing patient tumor tissue or PDX tumor tissue in vitro. Alongside tumor cells, these cultures can retain microenvironmental elements—fibroblasts and immune cells—providing a closer approximation of intratumoral heterogeneity than conventional cell lines. Compared with such lines, PDCs maintain the mutation spectrum, copy‐number alterations, and gene‐expression signature of the originating tumor. They are compatible with high‐throughput drug screens, allowing direct measurement of patient‐derived tumor cell sensitivity to chemotherapeutics or targeted agents and facilitating personalized therapy design (Mitra et al. 2013). Because MM specimens are scarce, establishing PDCs from PDX tumors has become a practical alternative. PDX models conserve the primary tumor's genetics, and serial in vivo passage enriches tumor content; tumor cells purified from PDX thus yield PDCs more reliably than those from primary biopsies. Nevertheless, current efforts have often yielded MM‐PDCs that tend to enter senescence and cease proliferation after 5–8 passages, potentially limiting their scale‐up and broader use (Huang et al. 2023).

4.5. Canine‐Derived Preclinical Models

The relatively low incidence of MM in human patients limits the scope of research in this field. Moreover, conventional GEMMs models often fail to fully recapitulate the complexity of spontaneously arising tumors. Canine models offer several advantages for MM research. First, dogs naturally develop MM in an immunocompetent setting. Second, the anatomical distribution—primarily occurring in the oral mucosa—and histopathological features closely resemble those in humans. Canine MM also mirrors human disease in its inter‐ and intra‐tumoral heterogeneity, metastatic behavior, recurrence, and therapy resistance (Khanna et al. 2006). At the molecular level, both species exhibit rare BRAF mutations, frequent KIT protein expression (∼85%) without a direct correlation to mutation status (Simpson et al. 2014), and concurrent activation of the MAPK and PI3K/AKT/mTOR pathways (Fowles et al. 2015). Additionally, the shared environmental exposures of dogs and humans further enhance the relevance of canine models for studying cancer development and treatment (Wei et al. 2016).

In 2019, Wong et al. performed cross‐species genomic analysis of 46 human MMs, 65 canine oral melanomas, and 28 equine mucosal melanomas (Wong et al. 2019). They identified shared mutations (e.g., NRAS) and copy number variations (e.g., MDM2 amplification and B2M deletion) between canine and human tumors, although key driver mutations such as SF3B1 and ATRX were absent in dogs. In 2025, Li et al. successfully established a canine OMM cell line, COMM6605, along with corresponding xenograft models including orthotopic, subcutaneous, and metastatic variants, providing a valuable tool for comparative studies (Li, Liu, et al. 2024). Bih‐Rong Wei et al. evaluated trametinib and sapanisertib, both as monotherapies and in combination, in canine MM cell lines (van Rooijen et al. 2017). Their findings support dual inhibition of the Ras/MAPK and PI3K/Akt/mTOR pathways as a promising strategy, particularly for patients lacking canonical driver mutations.

Spontaneous canine MM thus represents a highly relevant preclinical model for identifying novel therapeutic targets, evaluating drug efficacy and toxicity, and predicting clinical outcomes, holding significant promise for advancing MM research.

4.6. Zebrafish Models

Zebrafish share considerable anatomical, genetic, and pathway homology with humans (> 85%), including similarities in melanocyte development and the genetic basis of melanoma. Practical advantages include low‐cost drug screening via immersion in small volumes of solution, enabling efficient compound testing and combination studies. Their small size facilitates high‐throughput phenotype‐based drug screening. Additionally, genetic manipulation via microinjection at the single‐cell embryo stage is highly feasible.

The first zebrafish model of melanoma, developed by Patton et al. in 2005 (Berghmans et al. 2005), carried a mutation in the tumor suppressor gene TP53. In 2018, Leonard Zon's group knocked down SPRED1, a negative regulator of MAPK signaling, in zebrafish and demonstrated its tumor‐suppressive role, particularly in the context of KIT mutation—a finding with potential clinical implications for MM. Previously, Langenau et al. established a transparent immunodeficient adult zebrafish model capable of long‐term engraftment with human and murine cells, maintained at human physiological temperature with high viability (Yan et al. 2019).

In MM research, Babu et al. developed a zebrafish model in 2025 to investigate the effects of specific genetic alterations: overexpression of CCND1 coupled with loss of PTEN and tp53 in all melanocytes (Babu et al. 2025). Melanomas in this model arose specifically from melanocytes lining internal organs, mimicking the common sites of human MM. Both human patients and this zebrafish model showed upregulation of neural crest‐related migration genes and downregulation of antigen presentation genes, consistent with enhanced metastatic potential and reduced immune therapy sensitivity.

4.7. Comparative Overview of Preclinical Models

A comparative overview of preclinical models for MM is summarized in Table 5, outlining their respective strengths, limitations, applications, and relevance to MM biology. This integrated comparison highlights the complementary nature of existing systems and underscores the necessity of combining multiple platforms—and extending them with emerging virtual strategies—to fully capture the complexity of MM.

TABLE 5.

Comparative overview of preclinical models for MM.

Model type	Strengths	Limitations	Applications	Relevance to MM
PDO	Preserves 3D architecture, heterogeneity, immune co‐culture	Limited success rate, costly, technically demanding	Drug sensitivity testing, personalized therapy	High—suitable for rare samples and immune studies
PDX	Retains genomic/phenotypic features, mimics resistance, and recurrence	Expensive, time‐consuming, lacks immunity unless humanized	Precision therapy validation, biomarker discovery	High—robust translational fidelity
PDC	Maintains mutation/CNV spectrum, scalable assays	Short lifespan, senescence after passages	Personalized screening, mechanistic studies	Moderate—feasible via PDX‐derived sources
CCL	Easy to establish, cost‐effective, suitable for high‐throughput	Loss of heterogeneity, long‐term drift	Drug screening, basic biology	Moderate—foundational but less representative
Canine models	Spontaneous OMM, immunocompetent, histological similarity	Species differences, limited genetic overlap	Comparative oncology, drug toxicity	High—recapitulates tumor behavior
Zebrafish	Optical transparency, high‐throughput, genetic tractability	Different immune context	Migration studies, rapid compound screening	Moderate—complementary model

Note: This table summarizes the strengths, limitations, and applications of major preclinical model systems—including PDOs, PDXs, PDCs, CCLs, canine‐derived models, and zebrafish models—and evaluates their relevance to MM research.

4.8. Toward Integrated and Virtual Modeling

No single preclinical model can fully capture the molecular heterogeneity, immune suppression, and anatomical constraints of MM. PDOs, PDXs, PDCs, and CCLs each contribute unique advantages, yet all remain limited in scalability and fidelity (Saglam‐Metiner et al. 2019). Comparative oncology and zebrafish models provide complementary insights but still fall short of reflecting the complexity of human MM (Barbosa et al. 2025). These limitations highlight the need for an integrated framework, where multiple model systems are combined to provide complementary perspectives—from cellular and genetic mechanisms to immune interactions and therapeutic resistance.

Beyond physical models, emerging virtual strategies offer a transformative extension. By leveraging multi‐omics data, machine learning, and mechanistic simulations, in silico models can recapitulate tumor–immune–drug dynamics at scale, even in the context of scarce patient samples (Aziz 2024; Chakraborty et al. 2024; Li, Wu, et al. 2024). Such integration of traditional experimental platforms with computational modeling may establish a multi‐tiered translational pipeline, where hypotheses generated in silico are validated in organoids, xenografts, or comparative animal models, and vice versa. This synergy holds the potential to accelerate therapeutic discovery, refine patient stratification, and ultimately close the gap between MM biology and precision medicine.

The following section therefore focuses on these emerging computational and virtual strategies, highlighting their potential to complement traditional platforms and reshape the modeling landscape for MM.

5. Emerging Virtual Strategies for Modeling MM

In recent years, rapid advances in AI, biological systems modeling, and multi‐omics integration have positioned emerging virtual strategies and digital twins as pivotal frontier approaches in cancer modeling and personalized therapy research (Asghar and Chung 2025; Pati et al. 2024; Perez‐Lopez et al. 2024). Although direct applications in MM remain limited, such strategies integrate multi‐omics data, biological behavioral rules, and computational simulations to enable virtual experimentation, therapeutic screening, and mechanistic hypothesis generation, particularly where physical models are constrained [92]. For a rare and highly heterogeneous cancer like MM, emerging virtual strategies not only open new technical pathways for preclinical research but also present translational opportunities for optimizing targeted and immune therapies.

The rarity of patient samples, pronounced heterogeneity, and immunologically “cold” phenotype make robust preclinical modeling of MM exceptionally difficult (Dimitriou et al. 2022). Conventional systems such as PDOs, PDXs, PDCs, and CCLs have advanced the field, yet remain constrained by issues of scalability, reproducibility, and fidelity in recapitulating the tumor–immune microenvironment (Swayden et al. 2019). Against this backdrop, emerging virtual strategies—including machine learning, network‐based systems biology, and mechanistic simulations—are increasingly recognized as transformative complements. By reconstructing tumor–immune–drug dynamics in silico, these approaches require minimal biospecimens, enable virtual clinical trials, and extend the reach of scarce physical models, thereby opening new translational avenues for optimizing targeted and immune therapies in MM (Li, Wu, et al. 2024; Yeo and Selvarajoo 2022; Zheng et al. 2025).

To provide a clear comparative overview of these complementary virtual approaches, their key characteristics are summarized in Table 6.

TABLE 6.

Comparison of emerging virtual modeling strategies for MM.

Strategy	Core principle	Primary data input	MM applications	Advantages	Key challenges
Machine/Deep Learning (ML/DL)	Pattern recognition and predictive modeling from complex datasets.	Multi‐omics, digital pathology, clinical records.	Prognostic stratification; biomarker discovery; spatial TME quantification.	High predictive power; handles high‐dimensional data.	Demands large datasets; limited interpretability; risk of overfitting.
Network Analysis	Construction of biomolecular interaction networks to identify key pathways.	Interaction databases, gene expression, prior knowledge.	Mapping signaling pathways (e.g., MAPK); identifying drug combinations.	Systems‐level insight; integrates prior knowledge; interpretable.	Static representation; dependent on knowledgebase quality.
Mechanistic Modeling	Mathematical simulation of biological processes based on physical principles.	Kinetic parameters, imaging biomarkers (e.g., DKI), drug properties.	Simulating tumor‐immune dynamics; in silico drug trials.	Dynamic simulation; mechanistic insight; enables hypothesis testing.	Complex model building; parameterization difficulty; computational cost.
Integrative Paradigms	Translating natural language rules into executable computational models.	Biological rules, multi‐omics data, literature.	Building personalized “virtual MM” for therapy simulation.	Democratizes modeling; synthesizes fragmented knowledge.	Early developmental stage; challenge of rule formalization.

5.1. Machine Learning/Deep Learning

Machine Learning/Deep Learning (ML/DL) algorithms identify patterns from multi‐omics data (genomic, transcriptomic, and imaging) for patient stratification and prognosis prediction (Vera et al. 2022). In 2022, Lai et al. integrated genomics data with deep learning to identify a gene signature for prognostic classification of melanoma patients (Lai et al. 2022). Their approach combined network analysis and an autoencoder‐based DL model to reduce genomic data dimensionality into patient score profiles. These scores delineated three subtypes with distinct survival outcomes in TCGA melanoma cohorts. SHAP coefficients were used to quantify and rank the impact of genomic features, enhancing model interpretability. This strategy holds promise for MM, particularly in identifying subtypes with rare molecular features such as CDK4 amplification, KIT mutations, and loss of MHC expression. Recently, Andrew et al. (2025) developed MelanoMAP, a multimodal framework that integrates TME‐derived digital biomarkers from > 3500 histology slides with clinicopathological data to refine metastatic risk prediction in melanoma. Using SHAP analysis, the study underscored the pivotal role of TME features in driving disease progression. This fusion of digital pathology and multi‐omics provides a vital template for mucosal melanoma (MM) research; specifically, the automated quantification of MM‐specific spatial immune landscapes—such as T‐cell exclusion phenotypes—can mitigate the constraints of limited clinical specimens and facilitate more precise biomarker discovery (Andrew et al. 2025).

5.2. Network Analysis

This approach reconstructs gene/protein interaction networks to identify core pathways driving MM progression (Vera et al. 2022). Kuenzi et al. constructed a drug‐response network to predict the effects of combination therapies on melanoma cell survival. This methodology offers a framework for investigating microenvironment‐associated signaling pathways (e.g., TGF‐β and IL‐10) in MM (Kuenzi et al. 2020).

5.3. Mechanistic Modelling

Based on biophysical principles, dynamic models simulate tumor‐immune interactions or drug responses (Vera et al. 2022). In 2013, Passante et al. developed a computational model to predict melanoma cell sensitivity to apoptosis (Passante et al. 2013). Vetma et al. demonstrated that baseline expression levels of key apoptotic regulators could predict responses to TRAIL and dacarbazine in melanoma cell lines (Vetma et al. 2020). In silico clinical trials leveraging metastatic melanoma patient data have demonstrated promising predictive potential for estimating response rates to TRAIL variants and Birinapant. While the ultimate utility of this approach hinges on clinical efficacy validation, integrating MM‐specific microenvironment factors—such as immunosuppressive cell enrichment and MHC‐I downregulation—into such models provides a valuable framework for guiding future prospective trials. Integrating MM‐specific microenvironment factors—such as immunosuppressive cell enrichment and MHC‐I downregulation—into such models could provide a valuable platform for prospective validation of combination immunotherapies. Beyond genomic simulations, functional imaging based on physiological parameters offers robust predictive utility. For instance, quantifying intra‐tumoral features through advanced techniques such as DKI and IVIM enables the in silico modeling of c‐KIT inhibitor sensitivity, yielding results highly concordant with experimental drug‐response assays (Wang, Niu, et al. 2025).

5.4. Challenges and Future Directions for Virtual Modeling in MM

Despite advantages such as high throughput, and scalability, virtual modeling in MM faces challenges like data scarcity, limited generalizability, and underrepresentation of rare mutations such as GNAQ/GNA11. Recently, Johnson et al. proposed a conceptual framework—Cell Behavior Hypothesis Grammar (CBHG)—which uses natural language descriptions of cellular rules to construct mathematical models (Johnson et al. 2025). This approach systematically integrates biological knowledge and multi‐omics data to generate in silico virtual cell models. For MM, CBHG could integrate features such as CDK4 amplification and MHC‐I downregulation to build individualized “virtual MM” models, enabling simulation of tumor–immune–drug interactions and guiding experimental validation (Broit et al. 2021; Mengoni et al. 2025).

Future development should prioritize: (1) dedicated MM databases integrating multi‐omics and clinical data; (2) interpretable ML frameworks; (3) spatiotemporal models combining spatial and single‐cell omics; and (4) adoption of CBHG‐like methods to democratize virtual model construction.

Ultimately, as virtual strategies evolve from data‐driven AI to mechanism‐based simulations and natural language–driven “virtual cells,” they are expected to synergize with physical models and drive a paradigm shift in MM research.

6. Challenges and Future Directions

Conventional models such as 2D cell lines and PDXs often fail to accurately predict drug response (Viegas and Sarmento 2024). Most are derived from single biopsy sites, limiting their ability to capture tumor evolutionary heterogeneity (Sun et al. 2023). Furthermore, PDX models are costly and time‐consuming, while organoids lack vascularization, impairing drug penetration studies and translational relevance. To overcome these limitations, clinically oriented multi‐omics approaches—integrating genomic, transcriptomic, epigenomic, and proteomic data—are needed to elucidate tumor‐driving mechanisms. For example, UDA‐seq (Chinese Academy of Sciences) enables high‐throughput single‐cell multi‐omics profiling (e.g., RNA–ATAC), facilitating identification of rare resistant subpopulations such as ITGB1⁺PREX1⁺ T cells and associated regulatory networks (Li, Huang, et al. 2025).

Multi‐omics profiling (genomic, transcriptomic, epigenomic, and proteomic) combined with high‐throughput platforms such as UDA‐seq and CRISPR‐based screening can uncover rare resistant subpopulations and actionable targets (Augustin and Koh 2017; Xu et al. 2022). Emerging bioengineering approaches—3D bioprinting, organ‐on‐a‐chip, vascularized organoids—may improve physiological fidelity. At the same time, mini‐PDX systems and virtual modeling offer solutions to overcome sample scarcity and cost barriers. Virtual cell laboratories and frameworks such as Cell Behavior Hypothesis Grammar can simulate MM‐specific tumor–immune–drug dynamics, providing a scalable “digital testbed” to guide hypothesis generation and therapy optimization (Hernandez‐Boussard et al. 2021; Laubenbacher et al. 2022).

Looking forward, three priorities should be emphasized: (i) development of site‐specific models (e.g., oral and sinonasal mucosa, incorporating microbiota and mechanical stress) (Rambhia et al. 2019); (ii) establishment of functional validation systems to standardize drug response and immune‐mimicking benchmarks; and (iii) creation of population‐specific resources, particularly MM models derived from Asian cohorts where incidence is highest. Integrating these directions with emerging virtual strategies will yield a more robust and representative modeling ecosystem, accelerating the translation of precision therapies for this rare yet lethal malignancy.

7. Conclusion

This review has focused on the clinical characteristics, current treatments, and advances in preclinical modeling of MM, systematically outlining the types, strengths, and limitations of existing model systems, and discussing the potential value of emerging in silico strategies. These insights provide a reference for researchers in selecting context‐appropriate models for specific study aims. Although considerable progress has been made in recent years—from traditional cell lines and patient‐derived models (PDC, PDX, and PDO) to alternative animal models—significantly advancing mechanistic studies and therapy development, current systems still fall short of fully capturing the molecular heterogeneity, immune microenvironment, and tissue specificity of human MM. This limits the precision and predictive power of therapeutic development. Uniquely, this review underscores two underexplored but critical aspects: (i) the persistent translational bottlenecks caused by the absence of standardized, MM‐specific model repositories, and (ii) the emerging role of virtual strategies and computational simulations as complementary tools to overcome data scarcity and enable scalable hypothesis testing. Future work should emphasize multi‐model integration, coupling with multi‐omics and spatial omics data, and the systematic incorporation of in silico approaches alongside experimental platforms. Such efforts will provide a more robust foundation for translating precision‐targeted, immune, and combination therapies into clinical practice.

Author Contributions

Yuantai Zhu: investigation. Xiangjie Jin: writing – review and editing, writing – original draft. Zhiyuan Zhang: conceptualization, supervision, project administration. Yuhan Zhang: writing – original draft. Chaoji Shi: conceptualization, supervision, project administration.

Funding

This review is supported by the National Natural Science Foundation of China (82573697), the Shanghai's Top Priority Research Center (2022ZZ01017).

Conflicts of Interest

The authors declare no conflicts of interest.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.