Biosensors and Nanotheranostics

1. Introduction

Colorectal cancer (CRC) ranks as the third most frequently diagnosed malignancy and the second leading cause of cancer-related death worldwide, and projections suggest global cancer incidence will exceed 35 million new cases annually by 2050, driven by population growth, ageing and westernised lifestyle transition (Dhanapal et al., 2026; S. Wang et al., 2026). Management rests on a multimodal framework of surgical resection, radiotherapy and systemic chemotherapy, most commonly 5-fluorouracil with oxaliplatin (FOLFOX) or irinotecan (FOLFIRI) (Dhanapal et al., 2026; Li et al., 2024). Surgery remains curative for localised disease, yet it does not reliably eliminate occult micrometastases or prevent systemic recurrence (Dhanapal et al., 2026). Systemic agents and targeted biologics such as cetuximab and bevacizumab carry their own constraints: off-target toxicity, dose-limiting myelosuppression, and the rapid emergence of intrinsic or acquired resistance (Dhanapal et al., 2026; S. Wang et al., 2026). Immune checkpoint inhibition has transformed outcomes for the approximately 15% of patients whose tumours are mismatch-repair deficient or microsatellite instability-high (dMMR/MSI-H), but the remaining ~85% are microsatellite stable (MSS), carry low tumour mutational burden, and sit within an immunologically cold stroma that is largely refractory to checkpoint blockade (Dhanapal et al., 2026; S. Wang et al., 2026). The therapeutic ceiling, in other words, is not set by the potency of available agents. It is set by where those agents go and which cells they reach.

Underlying that ceiling is tumour heterogeneity, operating simultaneously at genetic, transcriptomic, epigenetic, histological and microenvironmental levels (Li et al., 2024; S. Wang et al., 2026; Zhang et al., 2026). CRC progression follows distinct evolutionary routes — chromosomal instability (CIN), microsatellite instability (MSI) and the CpG island methylator phenotype (CIMP) — with recurrent driver mutations in APC, KRAS, TP53, BRAF and PIK3CA (S. Wang et al., 2026). Transcriptomic profiling resolves CRC into four consensus molecular subtypes (CMS1–CMS4), each with its own immune landscape, drug sensitivity and metastatic tendency (Guinney et al., 2015; S. Wang et al., 2026). More consequential still is intratumoural heterogeneity: genetically and phenotypically distinct subclones coexisting within one lesion, providing the substrate for branched clonal evolution under therapeutic selection (Roerink et al., 2018; Zhang et al., 2026). Standard regimens tend to eradicate sensitive clones while permitting expansion of resistant ones, including CD44v9-positive cancer stem cells, which is a plausible mechanistic account of delayed relapse and multi-organ metastasis (Li et al., 2024; Resmi et al., 2026). The nested structure of this variation, and what each level implies for drug evaluation, is summarised in Figure 2 and catalogued in Table 3.

Theranostic nanoparticles were conceived to address precisely this problem of non-specific biodistribution by unifying diagnostic imaging and targeted delivery within a single nanostructure (Dhanapal et al., 2026; Younas et al., 2026). The available chemistries are diverse — lipid nanoparticles, liposomes, polymeric micelles, superparamagnetic iron oxide nanoparticles (SPIONs), mesoporous silica, gold nanostructures and quantum dots — and share a high surface-area-to-volume ratio that supports dense payload loading (Dhanapal et al., 2026; Younas et al., 2026). Co-encapsulating imaging reporters such as MRI contrast agents, near-infrared dyes or photoacoustic probes alongside chemotherapeutics, siRNAs or photosensitisers permits non-invasive, real-time readout of accumulation, biodistribution and therapeutic effect (Dhanapal et al., 2026; Younas et al., 2026). Stimuli-responsive designs go further, exploiting tumour microenvironment (TME) features — acidic pH, elevated glutathione, overexpressed matrix metalloproteinases and cathepsins — to trigger release at the lesion rather than in the circulation (Guo et al., 2026; Younas et al., 2026). Passive accumulation via the enhanced permeability and retention (EPR) effect can be supplemented by active ligand decoration against CD44, EpCAM, EGFR and CEA (Dhanapal et al., 2026; X. Wang et al., 2026; Younas et al., 2026). The architecture of these constructs and the triggers that unlock them are depicted in Figure 4 and classified in Table 2.

Despite this sophistication, most nanomedicines fail clinically, and the gap between preclinical promise and patient benefit is best explained by the models used to generate that promise (Li et al., 2024; Resmi et al., 2026). Two-dimensional monolayers lack three-dimensional architecture, microenvironmental heterogeneity and cell–matrix interaction; they drift genetically with passage and overestimate nanoparticle uptake roughly two-fold relative to primary tissue (Li et al., 2024; Resmi et al., 2026). Patient-derived xenografts (PDXs) retain histology and genotype in vivo but require four to eight months to establish, cost heavily, scale poorly, and progressively substitute murine stroma for human (S. Wang et al., 2026; X. Wang et al., 2026). Patient-derived organoids (PDOs) — self-organising three-dimensional epithelial cultures grown from surgical resections or biopsies — occupy the intervening position, preserving histopathological architecture, driver mutations, copy-number variation, transcriptomic diversity and subclonal structure across extended passage (Drost & Clevers, 2018; Li et al., 2024; Sato et al., 2011; S. Wang et al., 2026). Living PDO biobanks support medium-to-high-throughput screening within two to four weeks and show strong concordance with clinical response (Vlachogiannis et al., 2018). Where each platform falls on the fidelity–throughput trade-off is mapped in Figure 3 and tabulated in Table 1.

Bringing these two developments together — organoid platforms and theranostic nanomedicine — allows nano-bio interactions to be interrogated in a human tissue context (Resmi et al., 2026). Within three-dimensional architecture, nanoparticles acquire a protein corona composed of TME-specific factors such as WNT and EGF, and this adsorbed layer, rather than the engineered surface, establishes the particle’s biological identity and governs internalisation (Resmi et al., 2026). High-resolution profiling in patient-derived colonic organoids has shown subpopulation-specific uptake: stromal and fibroblast compartments favour clathrin- and dynamin-1-dependent endocytosis ending in lysosomal degradation, whereas malignant epithelial subclones exploit caveolin-1-mediated entry that bypasses lysosomal fusion and delivers carriers intact (Resmi et al., 2026). Deliberately pre-modulating the corona raised delivery to chemoresistant CD44v9-positive cells from 3.8% to 70%, an approximately 18-fold enhancement (Resmi et al., 2026). These relationships are illustrated in Figure 5 and the supporting experimental studies are collated in Table 4.

A comprehensive synthesis of how PDO screening can address CRC heterogeneity and accelerate theranostic translation nevertheless remains lacking. This review therefore sets out to: (i) evaluate the molecular and cellular heterogeneity of CRC and how interpatient, intratumoural and subclonal variation drives resistance (Guinney et al., 2015; Roerink et al., 2018; S. Wang et al., 2026); (ii) review advances in theranostic nanoparticle design, including TME-responsive release (Dhanapal et al., 2026; Guo et al., 2026; Younas et al., 2026); (iii) analyse the methodological advantages of PDOs over monolayers and animal models (Drost & Clevers, 2018; Li et al., 2024); (iv) examine the mechanisms of PDO-guided screening, from corona modulation to subpopulation-specific endocytosis (Resmi et al., 2026); and (v) identify translational barriers and future directions, including reproducibility, organ-on-a-chip integration, AI-assisted design and regulatory alignment (Dhanapal et al., 2026; Li et al., 2024; S. Wang et al., 2026). The overall logic connecting these objectives is presented in Figure 1.

2. Heterogeneity, Patient-Derived Organoids, and Corona-Engineered Nanoparticles in Colorectal Cancer

2.1 Molecular, cellular and microenvironmental dimensions of heterogeneity

Although early-stage localised CRC is routinely managed by resection and adjuvant chemotherapy, a substantial proportion of patients experience local recurrence or distant metastasis, particularly to liver and peritoneum (Li et al., 2024; S. Wang et al., 2026). The principal obstacle to durable control is biological heterogeneity expressed at interpatient, intralesional and subclonal levels (Li et al., 2024; Roerink et al., 2018).

At the molecular level, the CIN, MSI and CIMP cascades establish distinct evolutionary trajectories, with somatic drivers in APC, KRAS, TP53, BRAF and PIK3CA defining the dominant oncogenic circuitry (S. Wang et al., 2026; Zhang et al., 2026). Transcriptomic classification into CMS1 (immune/MSI-hypermutated), CMS2 (canonical, WNT-activated), CMS3 (metabolic) and CMS4 (mesenchymal, stroma-rich) adds a layer with direct therapeutic consequence (Guinney et al., 2015). CMS4 tumours in particular combine stromal invasion, TGF-β pathway activation and marked resistance to standard cytotoxic therapy (S. Wang et al., 2026). Beyond interpatient variation, multi-region sequencing and single-cell multi-omic profiling demonstrate branched clonal evolution within individual lesions (Roerink et al., 2018; Zhang et al., 2026). Under selective pressure from chemotherapy or anti-EGFR biologics, dominant sensitive clones are removed and minor resistant subclones expand (Li et al., 2024; S. Wang et al., 2026). Cancer stem cells carrying CD44 and the CD44v9 variant isoform are central to this resistant fraction, sustaining self-renewal and distant colonisation (Li et al., 2024; Resmi et al., 2026). Multi-omic comparison of primary lesions with matched peritoneal metastases has further identified glycosylphosphatidylinositol anchor attachment 1 (GPAA1) as a metastasis-associated protein whose contribution arises through protein-level dysregulation rather than genomic amplification (Zhang et al., 2026). The determinants reviewed here, together with the model-selection strategy each implies, are set out in Table 3 and depicted schematically in Figure 2.

2.2 From monolayers and xenografts to patient-derived organoids

Preclinical oncology has long relied on immortalised two-dimensional lines (Li et al., 2024; X. Wang et al., 2026). These do not reproduce three-dimensional architecture,

Table 1: Comparative evaluation of preclinical colorectal cancer platforms for drug and nanomedicine screening. Four platform classes are assessed across architectural complexity, retention of inter- and intratumoural heterogeneity, fidelity of nanomedicine transport and uptake, throughput, and principal translational barriers. The comparison establishes why monolayer potency data transfer poorly to patients, why patient-derived xenografts cannot support screening at scale, and why patient-derived organoids occupy the position from which subclone-resolved nanoparticle behaviour becomes observable. Tumour organoid-on-a-chip systems are included as the emerging configuration that restores vascular and mechanical variables absent from static organoid culture. Compiled from Dhanapal et al. (2026), Drost and Clevers (2018), Li et al. (2024), Resmi et al. (2026), Sato et al. (2011), Vlachogiannis et al. (2018), S. Wang et al. (2026), X. Wang et al. (2026) and Younas et al. (2026).

Table 2: Classification, design principles and theranostic functionality of nanoparticle platforms in colorectal cancer. Five material classes are organised by their diagnostic imaging modality, therapeutic payload and mechanism of action, targeting and tumour microenvironment responsiveness, physiological advantages, and outstanding safety or manufacturing concerns. The table shows how imaging and therapy are integrated within single constructs and how acidic pH, glutathione gradients and protease overexpression are exploited as release triggers. It also makes visible the trade-off recurring across classes, in which the features conferring targeting precision — surface functionalisation, membrane camouflage, multi-step synthesis — are the same features that complicate reproducible manufacture and regulatory characterisation. Compiled from Dhanapal et al. (2026), Guo et al. (2026), Qu et al. (2022), Resmi et al. (2026) and Younas et al. (2026).

Figure 1. Conceptual architecture of the review: from tumour heterogeneity to patient-stratified theranostic delivery. The schematic traces the logical sequence connecting the five stages of the argument developed across Sections 1 to 5. Intrinsic biological variation in colorectal cancer (stage I) produces the clinical bottleneck of relapse and immunotherapy refractoriness (stage II), to which theranostic nanoparticle engineering is the designed response (stage III). Patient-derived organoid screening (stage IV) supplies the resolution at which corona identity, subclonal tropism and endocytic routing become measurable, yielding subtype- and subclone-matched nanomedicine (stage V). The dashed return path indicates the iterative loop in which screening outcomes feed back into formulation design against organoids from the same patient.

Figure 2. Nested levels of colorectal cancer heterogeneity and their consequences for nanomedicine evaluation. Four hierarchically nested layers of biological variation are shown alongside the specific methodological demand each imposes on preclinical screening. Level 1 establishes the mutational substrate through chromosomal instability, microsatellite instability and CpG island methylator phenotype pathways; level 2 resolves transcriptomic subtype; level 3 captures branched subclonal and stem-cell architecture; and level 4 supplies the microenvironmental context that determines protein corona composition. The levels are nested rather than parallel, since a single biopsy carries all four simultaneously, which is the reason any one-dimensional model systematically misreads drug response. Synthesised from Guinney et al. (2015), Roerink et al. (2018), S. Wang et al. (2026) and Zhang et al. (2026).

cell–cell contact, extracellular matrix (ECM) barriers or microenvironmental gradients, and decades of passage on rigid plastic produce genetic drift, loss of parental subclonal diversity and altered phenotype (Li et al., 2024; Resmi et al., 2026). Candidates potent in monolayer assays therefore fail disproportionately in trials, in part because such assays overestimate internalisation and omit physiologically relevant transport barriers (Resmi et al., 2026; Younas et al., 2026).

PDX models, generated by implanting patient tumour fragments into immunodeficient mice, address several of these deficits, maintaining parental histology, genotype, copy-number profile and stromal architecture in vivo, and permitting study of tumour–stroma crosstalk and metastatic behaviour (S. Wang et al., 2026; X. Wang et al., 2026). Their utility for screening is nonetheless limited by establishment times of four to eight months, high cost, variable engraftment across molecular subtypes, progressive replacement of human stroma by murine elements, and an immunodeficient host that precludes evaluation of human-specific immunotherapies (S. Wang et al., 2026; X. Wang et al., 2026).

PDO technology represents the substantive methodological advance (Drost & Clevers, 2018; Sato et al., 2011). Derived from primary or metastatic resections, biopsies or circulating tumour cells, PDOs self-organise within basement membrane extract into three-dimensional epithelial structures that retain histopathological differentiation, driver mutations, epigenetic landscape and subclonal heterogeneity across long-term culture (Li et al., 2024; Vlachogiannis et al., 2018; S. Wang et al., 2026). Living biobanks can be expanded for medium-to-high-throughput assays within clinically actionable windows of two to four weeks, with demonstrated concordance against cytotoxic, targeted and chemoradiotherapeutic responses (Vlachogiannis et al., 2018), and medium-throughput screening of peritoneal metastasis organoids has been used to direct individualised therapy (Narasimhan et al., 2020). A direct comparison across all four platform classes is provided in Table 1, and their relative positioning in Figure 3.

2.3 Theranostic nanoparticle engineering and TME-responsive release

In parallel, nanomedicine has produced platforms intended to overcome non-specific biodistribution, poor aqueous solubility and systemic toxicity (Dhanapal et al., 2026; Younas et al., 2026). Theranostic constructs co-encapsulate therapeutic agents — 5-fluorouracil, paclitaxel, doxorubicin, siRNA — with diagnostic reporters including magnetic resonance contrast agents, near-infrared dyes and photoacoustic probes, enabling real-time visualisation of accumulation, distribution, release kinetics and efficacy (Dhanapal et al., 2026; Guo et al., 2026; Younas et al., 2026).

Material classes span lipid nanoparticles and liposomes, polymeric micelles such as PLGA and poly-L-glutamic acid, inorganic SPIONs, gold nanostructures, mesoporous silica, and semiconductor quantum dots including silicon (SiQDs) and graphene quantum dots (GQDs) (Dhanapal et al., 2026; Resmi et al., 2026; Younas et al., 2026). Passive targeting via EPR is unreliable where vascularisation is irregular and interstitial fluid pressure is elevated, as in dense desmoplastic stroma (Guo et al., 2026; X. Wang et al., 2026). Active targeting therefore employs surface ligands — antibodies, aptamers, hyaluronic acid, peptides — directed at CD44, EGFR, VEGFR and integrins (Guo et al., 2026; Younas et al., 2026).

The most consequential design advance is stimuli-responsiveness (Guo et al., 2026; Younas et al., 2026). Such carriers remain stable at physiological pH 7.4 yet undergo disassembly, charge reversal or bond cleavage on encountering tumour-specific triggers. Three dominate. Mildly acidic pH (6.2–6.8 extracellularly, 4.5–5.5 in endolysosomes) cleaves hydrazone, acetal or maleic acid amide linkages and drives protonation-dependent disassembly (Guo et al., 2026; Younas et al., 2026). Elevated intracellular glutathione, 100- to 1000-fold above extracellular concentrations, reduces disulfide bonds to release payload intracellularly (Guo et al., 2026; Younas et al., 2026). Overexpressed proteases including MMP-2/MMP-9, legumain and hyaluronidases cleave peptide motifs such as PLGVRG or AAN, stripping PEG shielding, altering particle morphology or liberating drug at the tumour margin (Guo et al., 2026). These mechanisms are illustrated in Figure 4; the corresponding platform classes, payloads, advantages and safety concerns appear in Table 2.

2.4 Nano-bio interactions: protein corona dynamics and subpopulation-specific delivery

Nanoparticles do not meet cells as pristine synthetic objects. On entering biological fluid they adsorb a dynamic

Figure 3. Positioning of preclinical platforms on the fidelity–throughput trade-off. Four model classes are plotted against screening scalability on the horizontal axis and fidelity to human tumour transport biology on the vertical axis, with annotations giving characteristic turnaround and principal limitation. Two-dimensional monolayers occupy the high-throughput, low-fidelity corner and overestimate nanoparticle uptake approximately two-fold; patient-derived xenografts recover physiological fidelity but require four to eight months per model; patient-derived organoid biobanks recover much of that fidelity at a fraction of the turnaround; and organoid-on-a-chip systems achieve the highest transport realism at substantial cost in throughput. The figure illustrates why a tiered configuration, rather than substitution of one platform for another, is the configuration this review recommends.

Figure 4. Theranostic nanoparticle architecture and the tumour microenvironment triggers that unlock payload release. The concentric construct on the left shows the canonical theranostic design, comprising a therapeutic payload core, a co-encapsulated imaging reporter, a stimuli-labile linker layer, a polyethylene glycol stealth shell and surface targeting ligands directed at receptors overexpressed on colorectal epithelium. The three panels on the right summarise the endogenous triggers exploited for site-specific release: mildly acidic interstitial and endolysosomal pH, the steep intracellular glutathione gradient, and overexpressed tumour proteases including matrix metalloproteinases, legumain and cathepsin B. External stimuli such as near-infrared irradiation, focused ultrasound and alternating magnetic fields may be layered onto these endogenous triggers. Synthesised from Dhanapal et al. (2026), Guo et al. (2026) and Younas et al. (2026).

protein layer — the corona — that alters hydrodynamic diameter, surface charge, colloidal stability and receptor affinity, conferring a biological identity distinct from the synthetic one and governing endocytosis, immune clearance and tissue accumulation (Resmi et al., 2026; Younas et al., 2026). Corona research historically concentrated on abundant serum proteins such as albumin, immunoglobulins, transferrin and apolipoproteins, which left unexamined the question of how proteins secreted within a patient’s own tumour microenvironment shape particle behaviour (Resmi et al., 2026).

Work in patient-derived colonic organoids (PDCOs) has begun to close that gap (Resmi et al., 2026). Using material-distinct nanoprobes, specifically SiQDs and GQDs, Resmi et al. (2026) reported that conventional monolayers overestimate uptake approximately two-fold relative to ex vivo PDCOs. PDCO monolayers preserve crypt-like organisation, self-assembling into node and non-node regions that contain stem-like, proliferative and differentiated compartments (Resmi et al., 2026). High-dimensional spectral flow cytometry with single-cell dimensionality reduction showed that uptake kinetics, cellular tropism and endocytic mechanism differ substantially between cell lines, fibroblasts and organoid subpopulations (Resmi et al., 2026).

Critically, the surrounding protein environment determines whether carriers accumulate in particular subclones. Pre-incubating SiQDs in organoid growth medium containing TME signalling factors produced a distinct, highly crosslinked corona unlike that formed in serum-rich medium (Resmi et al., 2026). Pre-engineering the corona with WNT, a secreted glycoprotein driving stemness, or with EGF, raised uptake in chemoresistant CD44v9-positive epithelial subpopulations from 3.8% in non-engineered particles to 70% and 66.85% respectively — an approximately 18-fold enhancement in delivery to the fraction associated with relapse (Resmi et al., 2026).

PDO screening additionally revealed that intracellular fate depends on cell-type-specific endocytic routing (Resmi et al., 2026). Primary organoid epithelial cells and normal fibroblasts internalise via clathrin- and dynamin-1-dependent endocytosis, directing carriers to late endosomes and lysosomes for acidic degradation. Malignant epithelial cells instead use caveolin-1-mediated uptake, bypassing lysosomal fusion and routing particles through caveosomes toward the endoplasmic reticulum and cytosol (Resmi et al., 2026). This divergence offers a mechanistic explanation for why carriers degraded in normal tissue may still deliver intact payload within malignant cells (Guo et al., 2026; Resmi et al., 2026). Figure 5 summarises these relationships, and the underlying experimental studies are compiled in Table 4.

2.5 Methodological challenges and emerging platforms

Several obstacles stand between this rationale and clinical translation (Li et al., 2024; S. Wang et al., 2026; Younas et al., 2026). Standard epithelial PDOs cultured in Matrigel or basement membrane extract lack functional vasculature, shear stress, interstitial flow and resident immune populations including T lymphocytes, tumour-associated macrophages and cancer-associated fibroblasts (Li et al., 2024; S. Wang et al., 2026). Non-epithelial components are progressively lost across passage, which may lead to underestimation of physical transport barriers and immune-mediated clearance (Li et al., 2024; Younas et al., 2026). Batch-to-batch variability in naturally derived hydrogels, labour-intensive protocols and the absence of standardised high-content imaging further complicate cross-laboratory reproducibility (Li et al., 2024; S. Wang et al., 2026).

Bioengineering responses are emerging. Tumour organoid-on-a-chip (ToOC) microfluidic systems combine three-dimensional organoids with perfusable endothelial channels, biomimetic matrices and controlled flow, reproducing extravasation, haemodynamic shear and tissue penetration while permitting real-time optical and electrochemical monitoring of carrier assembly and disassembly (S. Wang et al., 2026; Younas et al., 2026). Heterotypic co-cultures combining epithelial organoids with autologous peripheral blood mononuclear cells, tumour-infiltrating lymphocytes or cancer-associated fibroblasts enable functional assessment of immunomodulatory nanomedicines, tumour–immune crosstalk and macrophage repolarisation (Li et al., 2024; S. Wang et al., 2026). Finally, artificial intelligence and machine learning trained on high-content organoid imaging, single-cell transcriptomes and physicochemical descriptors can predict corona composition, uptake efficiency and patient-specific sensitivity in silico, allowing formulation design to be triaged before experimental validation (Dhanapal et al., 2026; Younas et al., 2026). Coupling AI-driven design with PDO validation offers a plausible route to systematic optimisation and patient stratification (Dhanapal et al., 2026; Li et al., 2024; S. Wang et al., 2026). The comparative strengths and remaining limitations of each platform, including ToOC, are detailed in Table 1.

3. Methods

3.1 Review design and its rationale

This work is a narrative, integrative review rather than a systematic review or meta-analysis. The choice was deliberate and follows from the nature of the question. The literature being synthesised spans three largely separate research communities — colorectal cancer molecular pathology, organoid methodology and nanomaterial engineering — which differ in study design, outcome definition and reporting convention to a degree that precludes meaningful statistical pooling. Nanoparticle uptake, for instance, is variously reported as percentage-positive cells, mean fluorescence intensity, mass per cell or fold-change against an internal control, with no shared denominator. Under these conditions a meta-analytic estimate would convey false precision. A narrative synthesis, conducted transparently, was judged the more defensible format (Li et al., 2024; S. Wang et al., 2026).

Transparency, however, is not optional simply because the design is narrative. The procedures below are reported at a level of detail intended to let an independent group reconstruct the evidence base, identify what was excluded and extend the search forward. The complete workflow is presented in Figure 6.

3.2 Information sources and search window

Five bibliographic resources were interrogated: PubMed/MEDLINE, Scopus, Web of Science Core Collection and Embase for peer-reviewed literature, together with the bioRxiv and medRxiv preprint servers to capture recent methodological work not yet through peer review. Preprints were admitted because several of the organoid–nanoparticle interaction studies central to this review exist currently only in that form; they are identified as preprints wherever cited, and conclusions drawn from them are flagged accordingly in Section 5. The search window extended from database inception to 30 June 2026. Reference lists of all included articles were hand-searched, and forward citation tracking was performed in Google Scholar on the three most central sources to capture work the database queries may have missed.

3.3 Search strategy

Searches combined controlled vocabulary with free-text terms across four concept blocks, joined by the Boolean operator AND within and OR between synonym sets. The PubMed strategy, adapted syntactically for the other platforms, was constructed as follows:

Block 1 (disease): “Colorectal Neoplasms”[MeSH] OR colorectal cancer[tiab] OR colorectal carcinoma[tiab] OR colon cancer[tiab] OR rectal cancer[tiab] OR CRC[tiab]

Block 2 (model): “Organoids”[MeSH] OR organoid*[tiab] OR “patient-derived organoid”[tiab] OR PDO[tiab] OR ”tumour organoid”[tiab] OR ”organ-on-a-chip”[tiab] OR ”patient-derived xenograft”[tiab] OR PDX[tiab]

Block 3 (intervention): “Nanoparticles”[MeSH] OR “Theranostic Nanomedicine”[MeSH] OR nanoparticle[tiab] OR nanomedicine[tiab] OR theranostic[tiab] OR “quantum dot“[tiab] OR liposom[tiab] OR micelle[tiab] OR SPION[tiab]

Block 4 (mechanism): “protein corona”[tiab] OR endocytosis[tiab] OR “tumour heterogeneity”[tiab] OR “intratumoral heterogeneity”[tiab] OR “drug resistance”[tiab] OR “stimuli-responsive”[tiab] OR “consensus molecular subtype*“[tiab]

Blocks 1 and 3 were mandatory in every query. Blocks 2 and 4 were applied in rotation so that, for example, disease plus intervention plus mechanism retrieved corona and resistance literature that an organoid-restricted query would have missed. Language was restricted to English. No restriction was placed on publication type at the search stage.

3.4 Eligibility criteria

Records were eligible if they met all of the following: (i) addressed colorectal cancer specifically, or reported a mechanism explicitly transferable to it; (ii) reported original experimental data, methodological development, or a synthesis of either, concerning preclinical model systems, theranostic nanoparticle design, or nano-bio interaction; (iii) described outcomes attributable to an identifiable model system, such that the biological context of any quantitative result could be determined; and (iv) were available in English in full text.

Records were excluded if they: (i) concerned non-colorectal malignancies without a mechanism generalisable to CRC — one exception was retained and is noted below; (ii) were conference abstracts, editorials, commentaries or correspondence without retrievable primary data; (iii) reported nanoparticle performance without specifying the biological model in which it was measured; or (iv) duplicated a dataset already represented by a more complete report. The single deliberate exception to criterion (i) is Qu et al. (2022), a biomimetic nanoparticle study conducted outside colorectal disease, retained because its three-dimensional tricellular spheroid methodology and magnetic hyperthermia mechanism bear directly on the transport questions this review addresses; readers should weigh its indication-specific findings with that caveat in mind.

3.5 Selection and data charting

Titles and abstracts were screened against the criteria above, and records surviving that pass were examined in full text. For each included source, information was charted onto a standardised extraction template capturing: bibliographic details; model system and its provenance; nanoparticle composition, size and surface chemistry where applicable; the biological outcome reported and its quantitative value; the mechanism proposed; and stated limitations. Charted entries were then assigned to one or more of four analytic axes — comparative model benchmarking, nanoparticle design and responsiveness, determinants of heterogeneity, and empirical nano-bio interaction — which correspond directly to Tables 1 through 4 respectively and structure the synthesis presented in Section 4.

3.6 Approach to synthesis and appraisal

Synthesis proceeded by narrative integration within each axis, followed by cross-axis comparison to identify convergence and contradiction. Quantitative findings are reported with their original model context attached rather than abstracted from it, since a figure such as the two-fold overestimation of uptake in monolayer culture is meaningful only in relation to the comparator against which it was measured (Resmi et al., 2026). Where a single source supplies a pivotal claim, that dependence is stated explicitly in the text rather than obscured by aggregation.

No formal risk-of-bias instrument was applied. Established tools for randomised or observational clinical research do not map onto bench studies of nanoparticle transport, and no consensus instrument exists for this literature. In place of formal scoring, included sources were appraised qualitatively against four criteria: whether the model system was fully specified; whether quantitative outcomes were reported with dispersion or replicate number; whether mechanistic claims were supported by intervention, such as pathway inhibition, rather than by correlation alone; and whether the authors themselves identified limitations. Sources that satisfied fewer of these criteria are cited with correspondingly hedged language.

3.7 Declared limitations of the method

Three limitations follow directly from the design and should condition interpretation of everything that follows. First, no protocol was registered prospectively, so the possibility of post hoc refinement of the search cannot be formally excluded. Second, screening and charting were not performed in duplicate by independent reviewers, which leaves selection open to single-assessor bias. Third, restricting to English-language records introduces a known and unquantified geographic skew. These constraints are characteristic of narrative reviews generally; stating them is the minimum condition under which such a review can be read critically (Figure 6).

4. Preclinical Model Choice and Corona Engineering as Determinants of Theranostic Nanomedicine Development in Colorectal Cancer

The evidence assembled here converges on a proposition that is, on reflection, fairly simple: the choice of preclinical model is not a technical preliminary to nanomedicine development but a determinant of what that development can discover (Dhanapal et al., 2026; Li et al., 2024; Resmi et al., 2026; S. Wang et al., 2026). The findings are organised below along the four analytic axes defined in Section 3.5.

4.1 Comparative benchmarking: the platforms differ in kind, not merely in degree

Set side by side, the four platform classes diverge sharply in fidelity, transport realism and screening utility (Table 1; Figure 3). Two-dimensional lines such as HT29, HCT116 and COLO201 are inexpensive and fast, turning around an assay in one to three days, but they lack spatial geometry, ECM diffusion barriers and cell polarisation (Li et al., 2024). The practical consequence is quantitative and now reasonably well characterised: monolayers overstate cellular nanoparticle uptake by roughly two-fold relative to patient tissue models, and they reproduce neither endosomal–lysosomal trafficking nor subpopulation-specific tropism (Resmi et al., 2026). Whether the two-fold figure generalises across all nanomaterial classes is not yet established — it derives principally from quantum dot probes — but the direction of the error appears consistent.

PDX models retain parental genotype, copy-number variation and desmoplastic stroma in vivo (S. Wang et al., 2026; X. Wang et al., 2026). Their constraints are logistical as much as biological: low throughput, high cost, engraftment windows of four to eight months, and progressive replacement of human stroma by murine elements (S. Wang et al., 2026; X. Wang et al., 2026). Because standard PDX hosts are immunodeficient, immunomodulatory and checkpoint-directed nanomedicines cannot be meaningfully evaluated in them at all (S. Wang et al., 2026).

PDO biobanks sit between these positions and, on the evidence reviewed, capture most of what matters. They maintain self-organising crypt–villus architecture preserving interpatient, intralesional and subclonal heterogeneity across passage (Drost & Clevers, 2018; Sato et al., 2011; S. Wang et al., 2026), and they support screening within two to four weeks with demonstrated concordance against clinical outcome (Vlachogiannis et al., 2018). Coupling organoids to perfusable endothelial channels in ToOC devices restores haemodynamic shear, interstitial pressure and extravasation barriers, permitting real-time tracking of carrier disassembly and penetration (Dhanapal et al., 2026; Younas et al., 2026) — at the cost of throughput and considerable technical complexity (Table 1).

4.2 Theranostic design: responsiveness delivers, though the evidence base is uneven

Nanomedicine engineering has produced a broad repertoire unifying imaging with site-specific delivery (Table 2). Co-encapsulating reporters — SPIONs for T₂-weighted MRI, near-infrared dyes, photoacoustic agents — with payloads such as 5-FU, oxaliplatin, paclitaxel, doxorubicin or siRNA permits real-time readout of biodistribution and response (Dhanapal et al., 2026; Guo et al., 2026; Younas et al., 2026).

The clearest gains come from stimuli-responsive release (Figure 4). pH-responsive systems exploit acidic extracellular tumour pH of 6.5–6.8 and endolysosomal pH of 4.5–5.5 to trigger charge reversal or ester and hydrazone cleavage (Guo et al., 2026; Younas et al., 2026). Enzyme-responsive systems use MMP-2/MMP-9, cathepsin B and hyaluronidases to cut peptide linkers such as GPLGVR motifs; MMP-responsive ferritin nanoparticles shed protective coatings site-specifically, reducing particle size from 15 nm to 9 nm on cleavage and inducing 40.93% apoptosis in resistant models with limited systemic toxicity (Guo et al., 2026) (Table 4).

Multi-stimuli systems combining endogenous triggers with external physical stimuli produce the largest reported effects. In tricellular three-dimensional spheroids of Caco-2 cells, fibroblasts and endothelial cells, melittin-functionalised magnetic iron oxide nanoparticles under an alternating magnetic field disrupted mitochondrial cristae, depleted ATP to 14–23% of control and activated caspase-9-dependent death (Qu et al., 2022). Photothermal multi-enzyme probes achieved 39.1% photothermal conversion efficiency with GSH depletion and amplified reactive oxygen species, while dual pH/MMP-2-responsive nanodroplets reported 78.1% primary tumour growth inhibition and a 93% reduction in pulmonary metastatic niches (Guo et al., 2026; Younas et al., 2026) (Table 4). These are striking numbers. They should be read with care: they derive from heterogeneous model systems, often single studies, and in the case of Qu et al. (2022) from a non-colorectal indication. The consistency of direction across chemistries is more persuasive than any individual magnitude.

4.3 Multi-omic determinants of subclonal resistance

Paired analyses of primary lesions and distant metastases confirm that intratumoural heterogeneity and branched clonal evolution underpin resistance (Li et al., 2024; S. Wang et al., 2026; Zhang et al., 2026) (Table 3). CMS profiling shows CMS1 (MSI-high) tumours responding well to checkpoint inhibition while CMS4 tumours combine desmoplastic stroma, active TGF-β signalling and pronounced resistance to FOLFOX and FOLFIRI (Guinney et al., 2015; S. Wang et al., 2026).

Integrated whole-exome sequencing with quantitative proteomics on paired primary and peritoneal metastatic tissue identified GPAA1, the catalytic subunit of the GPI transamidase complex, as a driver of peritoneal spread (Zhang et al., 2026). Elevated GPAA1 protein in peritoneal metastases was associated with shorter recurrence-free survival (HR = 1.72, p = 3.4 × 10⁻⁵) and with disease-specific survival on multivariable analysis (HR = 1.83, p = 0.027) (Zhang et al., 2026). Single-cell RNA sequencing and spatial transcriptomics localised GPAA1 to metabolically active malignant cells with enhanced oxidative phosphorylation, spatially co-located with stromal fibroblasts and myeloid infiltrate, and gain- and loss-of-function assays implicated AKT phosphorylation in its promotion of proliferation, migration and epithelial–mesenchymal transition (Zhang et al., 2026). GPAA1 thus presents as both biomarker and candidate nanomedicine target — though it should be said that no nanoparticle has yet been reported against it, and the proposal remains prospective (Figure 2; Table 3).

4.4 Nano-bio interaction in organoids: the corona as the operative variable

The empirical organoid work yields what is arguably the central finding of this review (Table 4; Figure 5). When SiQDs and GQDs enter biological environments they rapidly adsorb surrounding proteins into a corona that establishes their biological identity (Resmi et al., 2026). Conventional culture relies on fetal bovine serum dominated by non-specific albumin; organoid growth media instead contain patient-relevant TME factors including WNT and EGF, which form a densely crosslinked, tumour-tailored corona (Resmi et al., 2026).

Spectral flow cytometry with single-cell clustering demonstrated that pre-engineering the SiQD corona with WNT or EGF redirected particles specifically to chemoresistant CD44v9-positive populations (Resmi et al., 2026). Non-engineered SiQDs achieved 3.8% uptake in CD44v9-positive cells; WNT- and EGF-functionalised coronas raised this to 70% and 66.85% respectively, approximately an 18-fold gain in delivery to the fraction associated with relapse (Resmi et al., 2026). It is worth being explicit that this result currently rests on a single preprint report in one organoid system, and independent replication across additional patient lines and nanomaterial classes has not yet been published. The mechanism is coherent and the effect size is large; neither guarantees generality.

Single-cell pathway analysis further showed endocytic mechanism to be strictly cell-type-dependent (Resmi et al., 2026). Normal epithelial cells and fibroblasts internalise via clathrin- and dynamin-1-mediated endocytosis into late endosomes and lysosomes for acidic degradation. Malignant subclones exploit caveolin-1-mediated uptake, bypassing lysosomal fusion and routing carriers through caveosomes to the endoplasmic reticulum and cytosol (Resmi et al., 2026). This divergence determines whether a carrier is destroyed in healthy tissue or delivered intact in malignant tissue — which is to say, it is the mechanistic substrate of a therapeutic window, and it is invisible in monolayer culture (Figure 5).

4.5 Convergence across the four axes

Read together, the axes point the same way. Heterogeneity determines which cells must be reached (Table 3); nanoparticle chemistry determines what can be built (Table 2); model choice determines whether delivery to those particular cells can be observed at all (Table 1); and the organoid experiments demonstrate that when it is observed, the governing variable is the adsorbed corona rather than the engineered surface (Table 4). The practical implication is that corona composition belongs among the specified design parameters of a theranostic construct, alongside size, charge and ligand density — and that it can only be specified against a model carrying the relevant secreted-factor milieu (Figure 1; Figure 5).

5. Reframing Organoids as Design Instruments for Corona-Engineered Nanomedicine in Colorectal Cancer

5.1 Principal interpretation

The conventional sequence in nanomedicine development places the complex model late: formulations are optimised in monolayers, advanced to animals, and organoids, if used at all, serve as confirmation before in vivo work. The evidence reviewed here suggests that ordering is inverted relative to where the information actually lies (Resmi et al., 2026) (Figure 1). If the adsorbed protein corona governs biological identity, and if corona composition depends on the secreted-factor environment, then a model lacking that environment cannot report the variable that most determines outcome. Optimising in serum-rich monolayer culture is optimising against a corona no tumour will ever present (Figure 5).

This reframing is more than rhetorical. It reclassifies the organoid from validation step to design instrument — the stage at which corona composition, subclonal tropism and endocytic routing are selected rather than merely confirmed (Table 1; Table 4).

5.2 Why the corona finding carries particular weight

Three features make the corona result consequential. It is mechanistic rather than phenomenological: WNT and EGF were selected because of their established roles in stemness and proliferative signalling in CRC, not identified by screening, so the effect has a prior rationale

Figure 5. Protein corona pre-engineering redirects nanocarriers to chemoresistant subclones, and cell identity determines their intracellular fate. Panel A contrasts the generic corona formed in serum-rich medium with the densely crosslinked, tumour-tailored corona formed when silicon quantum dots are pre-incubated in organoid medium containing WNT or EGF. Panel B shows the divergence in endocytic routing, in which normal epithelium and fibroblasts internalise carriers via clathrin and dynamin-1 into degradative lysosomes while malignant subclones use caveolin-1 entry that bypasses lysosomal fusion and preserves payload. Panel C quantifies the consequence, with uptake in CD44v9-positive chemoresistant cells rising from 3.8% without corona engineering to 66.85% and 70% with EGF and WNT coronas respectively. These relationships are not resolvable in monolayer culture. Data from Resmi et al. (2026).

Figure 6. Evidence identification, screening and synthesis workflow for this narrative review. The diagram documents the four-stage procedure described in Section 3, from database identification through title and abstract screening, full-text eligibility appraisal and thematic charting onto the four analytic axes that structure Section 4 and Tables 1 to 4. Exclusion grounds applied at each stage are shown on the right. The closing panel states the declared scope of the method explicitly: this is a transparent narrative synthesis rather than a systematic review, conducted without protocol registration, without duplicate screening and without a formal risk-of-bias instrument, with search strings, date limits and charting framework reported so that the evidence base can be reconstructed and extended by other investigators.

(Resmi et al., 2026). Its magnitude is large enough to be clinically meaningful — moving from 3.8% to roughly 70% uptake is a difference in kind, not a marginal optimisation (Figure 5). And its target is the correct one: CD44v9-positive cells constitute precisely the chemoresistant fraction implicated in relapse (Li et al., 2024; Resmi et al., 2026) (Table 3).

The divergence in endocytic routing matters for a related but distinct reason. Clathrin- and dynamin-1-dependent entry in normal cells terminating in lysosomal degradation, against caveolin-1-mediated entry in malignant cells bypassing that fate, supplies a mechanistic account of selectivity that does not depend on differential binding at all (Resmi et al., 2026). A carrier taken up equally by both compartments could still act selectively if only one compartment preserves the payload (Guo et al., 2026) (Figure 5). That widens the design space considerably.

5.3 How far the evidence actually extends

Enthusiasm should be tempered, and it is better to be explicit about where. The corona findings rest substantially on one preprint reporting one organoid system with two quantum dot chemistries (Resmi et al., 2026). Quantum dots are convenient probes — small, photostable, traceable — but they are not representative of the liposomes, polymeric micelles and ferritin cages that dominate clinical translation (Table 2). Whether corona pre-engineering transfers to a 100 nm PEGylated liposome, where shielding is itself designed to resist protein adsorption, is an open question that the current data do not answer.

Second, the quantitative claims in Table 4 derive from heterogeneous systems. The 40.93% apoptosis figure for MMP-responsive ferritin, the 39.1% photothermal conversion efficiency, the 78.1% growth inhibition and 93% metastatic niche reduction each come from distinct models under distinct protocols (Guo et al., 2026; Younas et al., 2026). The melittin-magnetic nanoparticle work was performed in a non-colorectal context (Qu et al., 2022). These are not grounds for dismissal, but they do preclude the kind of cross-study comparison that a reader might otherwise be tempted to make.

Third, the GPAA1 proposal remains at the biomarker stage. The survival associations are statistically robust and the functional data are supportive (Zhang et al., 2026), but no nanocarrier directed at GPAA1 has been reported, and the step from target identification to deliverable construct is routinely where such programmes stall (Table 3).

5.4 What organoids still cannot do

Organoids have real and consequential blind spots (Table 1). Standard epithelial PDOs lack functional vasculature, shear stress, interstitial flow and resident immune populations (Li et al., 2024; S. Wang et al., 2026). For nanomedicine this is not a minor omission: a substantial fraction of an administered dose is cleared by the mononuclear phagocyte system before reaching tumour at all, and that process is simply absent from an epithelial organoid. A carrier that performs well in PDO screening may still fail in vivo for reasons the organoid was never positioned to detect (Younas et al., 2026). Progressive loss of non-epithelial components across passage compounds this, risking underestimation of both physical barriers and immune clearance (Li et al., 2024).

Reproducibility presents a further obstacle. Batch variation in naturally derived hydrogels, labour-intensive protocols and the absence of standardised high-content imaging impede cross-laboratory comparison (Li et al., 2024; S. Wang et al., 2026). If PDO screening is to function as a decision gate, this is not a peripheral concern — a gate that different laboratories apply differently is not a gate.

5.5 Plausible routes forward

ToOC systems address the vascular and mechanical deficits directly, restoring perfusion, shear and extravasation barriers with real-time monitoring (S. Wang et al., 2026; Younas et al., 2026) (Table 1). The sensible configuration is probably tiered rather than substitutive: PDO biobanks for breadth across patients and formulations, ToOC for transport validation of the shortlist. Heterotypic co-culture with autologous PBMCs, TILs or cancer-associated fibroblasts extends the system to immunomodulatory agents and macrophage repolarisation (Li et al., 2024; S. Wang et al., 2026) — of obvious relevance to the MSS majority in whom checkpoint blockade fails (Dhanapal et al., 2026) (Table 3).

Artificial intelligence may help, with a caveat. Models trained on organoid imaging, single-cell transcriptomes and physicochemical descriptors can predict corona composition, uptake and patient-specific sensitivity in silico (Dhanapal et al., 2026; Younas et al., 2026). The caveat is that such models inherit the biases of their training data: an algorithm trained predominantly on monolayer

Table 3: Molecular, cellular and microenvironmental determinants of colorectal cancer heterogeneity and therapy resistance. Five dimensions of heterogeneity are characterised by their genomic, transcriptomic and epigenetic features, their pathophysiological function and subclonal phenotype, their therapeutic and clinical consequences, and the preclinical model configuration each demands. Reading the final two columns together clarifies the central methodological argument of this review: each distinct mechanism of resistance requires a model that preserves the specific biological feature responsible, so that CMS4 stromal barriers, branched subclonal architecture, CD44v9-positive stem-like fractions and immune exclusion each imply a different screening design. Compiled from Dhanapal et al. (2026), Guinney et al. (2015), Li et al. (2024), Narasimhan et al. (2020), Resmi et al. (2026), Roerink et al. (2018), S. Wang et al. (2026) and Zhang et al. (2026).

Table 4: Empirical preclinical studies of organoid and 3D-model nanoparticle interactions. Five experimental systems are summarised by nanoparticle construct and surface chemistry, protein corona composition and resulting biological identity, cellular uptake mechanism and intracellular trafficking route, and principal quantitative outcome. The first entry documents the finding central to this review, in which corona pre-engineering with WNT or EGF raised delivery to chemoresistant CD44v9-positive cells from 3.8% to approximately 70% while monolayer culture overestimated uptake two-fold. The remaining entries illustrate stimuli-responsive mechanisms in three-dimensional models. Readers should note that these outcomes derive from heterogeneous systems and protocols and are not directly comparable; the entry from Qu et al. (2022) originates outside colorectal disease and is included for its methodological relevance. Compiled from Guo et al. (2026), Qu et al. (2022), Resmi et al. (2026) and Younas et al. (2026).

measurements will reproduce the two-fold overestimate rather than correct it (Resmi et al., 2026). The value of AI here depends on organoid-derived data being generated first, at sufficient scale — which returns the problem to standardisation.

5.6 Implications for practice and reporting

Several implications follow for how this work might be conducted and reported. Nanoparticle studies should specify the protein environment in which uptake was measured, since reporting uptake without that context is reporting a number whose meaning cannot be recovered (Resmi et al., 2026). Corona composition warrants treatment as a tunable design parameter rather than an unavoidable confounder — arguably the most actionable single conclusion available here (Figure 5). Where a construct is intended for a defined subclone, screening should be conducted against models retaining that subclone, which in practice means organoids (Table 1; Table 4). And model provenance, passage number and medium composition belong in the methods of every such study, since without them a result cannot be interpreted, let alone replicated.

Whether PDO screening will in fact improve clinical translation rates is, it must be conceded, not yet demonstrated. No nanomedicine has to our knowledge completed the full path from PDO-guided optimisation to clinical endpoint. The argument advanced here is one of mechanism and plausibility, not of proven outcome, and it should be read on those terms.

6. Conclusion

Colorectal cancer resists durable control not because effective agents are unavailable but because tumours are internally plural, and conventional models render that plurality invisible. Patient-derived organoids change what is observable. By reconstituting the secreted-factor environment of human tumour tissue, they reveal that the adsorbed protein corona, rather than the engineered nanoparticle surface, governs which subpopulation a carrier reaches, and that malignant and non-malignant cells route identical particles through divergent endocytic pathways with opposite consequences for payload survival. Pre-engineering the corona with WNT or EGF increased delivery to chemoresistant CD44v9-positive cells roughly eighteen-fold. These findings reposition the organoid from a late validation step to an early design instrument, and they argue for treating corona composition as a specified design parameter. Substantial obstacles remain — absent vasculature and immunity, hydrogel batch variability, and reliance on a narrow and largely unreplicated empirical base — but the direction of travel is clear enough to act upon.

References