Translational Barriers to AAV and CRISPR Gene Therapy in Diabetes

Summary

Gene therapy targets diabetes pathophysiology rather than symptoms, yet clinical translation is slower than preclinical success. This review synthesizes 143 registered trials (ClinicalTrials.gov, 2010-2025) identifying systematic implementation barriers. Despite >15 years of development, 83% of trials remain in Phase I-II. Only zimislecel achieved Phase III outcomes (83% insulin independence at 12 months, n=12, requiring immunosuppression). VM202 for diabetic neuropathy failed Phase III despite positive extension results. Three systematic barriers emerged: (1) human transduction efficiency is 6-8.7-fold lower than preclinical models; (2) pre-existing immunity excludes 58-78% of candidates; (3) manufacturing capacity serves <2.5% of target population (40-400 years to treat at maximum capacity). These constraints explain why ex vivo cell therapies advanced to efficacy trials while in vivo gene delivery remains in Phase I despite longer development timelines. Research priorities should emphasize non-viral delivery systems offering scalable manufacturing and universal hypoimmune donor cells to address access barriers.

Introduction

Diabetes mellitus is a major global health challenge in the 21^st century, affecting >589 million adults, with estimates suggesting an increase of 783 million by 2045 [1]. Type 1 diabetes mellitus (T1DM) and type 2 diabetes mellitus (T2DM) are the 2 main forms. which includes a range of conditions marked by persistent hyperglycemia caused by abnormalities in insulin secretion, action, or both [2,3].

Current treatment strategies focus on glycemic control through oral hypoglycemic medications, exogenous insulin, and lifestyle modifications, but fail to tackle fundamental pathophysiological mechanisms progressive β-cell destruction in T1DM and metabolic dysfunction in T2DM [4,5,6]. The persistence of complications including nephropathy, retinopathy, neuropathy, and cardiovascular disease despite these interventions, underscores the need for disease-modifying therapies [7,8]. This has prompted the exploration of innovative therapeutic strategies to address the root causes. Gene therapy represents a significant advancement in therapeutic strategies aimed at targeting the molecular underpinnings of diabetes, rather than solely alleviating its symptoms [9,10]. This method involves the incorporation of genetic material into patient cells to repair defective genes or to impart new cellular functions, thereby restoring normal metabolic homeostasis [11,12]. In contrast to traditional therapies, which require ongoing treatment to sustain therapeutic outcomes, gene therapy presents the possibility of long-lasting and potentially curative solutions that can significantly change disease course [13,14].

Brief pathophysiological context: Type 1 diabetes results from autoimmune destruction of pancreatic β-cells, with 70-90% β-cell loss preceding clinical symptoms [15,16]. Type 2 diabetes involves insulin resistance in skeletal muscle, adipose tissue, and liver, with progressive β-cell dysfunction [6,17]. Gene therapy targets include β-cell replacement or regeneration, immune modulation to prevent autoimmune destruction, restoration of insulin sensitivity, and management of vascular/neurological complications [9,10].

Review scope and search strategy

Study design and rationale. This study employed a narrative review design rather than systematic review with meta-analysis due to substantial heterogeneity in interventions (viral vectors, CRISPR gene editing, cell-based therapies), study designs (preclinical animal models, Phase I-III clinical trials), patient populations (type 1 versus type 2 diabetes with varying disease duration), and outcome measures (glycemic control, insulin independence, complication improvement, safety endpoints). This heterogeneity precludes statistical pooling required for meta-analysis and necessitates qualitative synthesis approaches.

Literature search strategy. A comprehensive literature search was conducted across four databases: PubMed/MEDLINE, Web of Science, Google Scholar, and ClinicalTrials.gov, covering publications and trial registrations from January 2010 to June 2025. The search was completed on June 15, 2025. For PubMed/MEDLINE, the search strategy combined Medical Subject Headings and text words as follows: gene therapy OR genetic therapy OR CRISPR OR AAV OR adeno-associated virus OR lentiviral vector OR viral vector, combined with diabetes mellitus OR type 1 diabetes OR type 2 diabetes OR T1DM OR T2DM, further combined with clinical trial OR preclinical OR animal model OR clinical translation. Searches were limited to English language publications from 2010 to 2025. For ClinicalTrials.gov, advanced search parameters included diabetes-related conditions combined with gene therapy interventions, with study start dates between January 2010 and June 2025, including all trial statuses (recruiting, completed, terminated, withdrawn). Web of Science and Google Scholar searches employed similar Boolean combinations adapted for each platform's syntax.

Study selection criteria. Inclusion criteria encompassed peer-reviewed original research articles, systematic reviews, preclinical experimental studies utilizing established diabetic animal models (streptozotocin-induced diabetes, NOD mice, db/db mice, genetic diabetic models), and human clinical trials at any phase involving gene therapy approaches including viral vector delivery systems, CRISPR or other gene editing technologies, genetically modified cell-based therapies, or growth factor gene therapy for diabetic complications. Registered clinical trials from ClinicalTrials.gov were included regardless of publication status to assess the clinical development pipeline and minimize publication bias. Exclusion criteria included studies of diabetes complications without gene therapy intervention, case reports with fewer than three subjects lacking mechanistic insights, non-peer-reviewed sources except registered clinical trials, non-English publications without available translations, and duplicate publications reporting identical results from the same cohort, with the most recent publication retained. The literature search identified 4,438 initial records, which underwent systematic screening, resulting in the final inclusion of 488 sources (345 peer-reviewed articles, 143 trial registries).

Two independent reviewers screened titles and abstracts against eligibility criteria, with full-text articles retrieved for potentially relevant studies. Disagreements were resolved through discussion or consultation with a third reviewer. Data extraction captured study characteristics (author, year, country, study design, sample size, follow-up duration), intervention details (vector type, dose, delivery route for preclinical studies; trial phase, patient population, endpoints for clinical trials), outcome measures, statistical significance, and adverse events. Data extraction was performed by one reviewer and verified by a second reviewer, with discrepancies resolved through re-examination of source documents.

Quality assessment and bias minimization. Preclinical studies were evaluated using eight quality criteria: use of validated diabetic animal models, inclusion of appropriate control groups (vehicle-treated or untreated diabetic controls), adequate sample sizes with minimum six animals per group, random assignment to treatment groups, blinded outcome assessment, independent replication by different research groups, appropriate statistical analysis with correction for multiple comparisons where applicable, and transparent reporting of mortality, toxicity, and off-target effects. Studies meeting 6 or more criteria were classified as high quality, 4 to 5 as moderate quality, and fewer than four as low quality. Clinical trials were assessed for trial registration with prospective protocol specification, clear pre-specified primary outcomes, appropriate randomization and blinding for Phase II/III trials (Phase I single-arm trials were exempt from these criteria), sample size justification, intention-to-treat analysis, systematic adverse event reporting using Common Terminology Criteria for Adverse Events or similar grading, adequate follow-up duration (minimum 3 months for safety trials, 12 months for efficacy trials), and funding disclosure with conflict of interest statements. Phase I trials were evaluated primarily on safety-focused criteria, while Phase II/III trials required meeting seven or more of nine criteria for high quality classification.

Multiple strategies were implemented to minimize bias. Publication bias was addressed by including registered clinical trials from ClinicalTrials.gov regardless of publication status, searching conference proceedings via Google Scholar to identify preliminary unpublished results, and explicitly noting completed trials without published data. The review acknowledges potential overrepresentation of positive preclinical results due to publication bias favoring statistically significant findings. Selection bias was minimized through independent dual review of study selection with consensus requirement and pre-specified eligibility criteria applied uniformly. Geographic bias was quantified, revealing that 67% of trials originated from Western countries (United States, Europe, Australia) despite low- and middle-income countries bearing 75% of global diabetes burden, and this disparity is explicitly discussed in Section 7.3. Language restriction to English publications was acknowledged as a limitation, estimated to exclude less than 5% of relevant gene therapy diabetes literature based on database language filters.

Data synthesis approach. Narrative synthesis followed structured thematic organization: preclinical evidence was organized by therapeutic mechanism (insulin replacement, β-cell regeneration, immune modulation); clinical trial data was stratified by development phase (Phase I, II, III) and therapeutic modality (direct gene delivery, ex vivo cell therapy, complication therapies); translational barriers were identified through direct quantitative comparison of preclinical outcomes (transduction efficiency, duration of effect, immune responses) to clinical results using identical parameters; and cross-study patterns in safety, efficacy, and feasibility informed the research prioritization framework presented in Section 6.6.4. This approach enables identification of systematic translational gaps but does not permit quantitative pooling or formal meta-analytic techniques. Where multiple studies reported similar outcomes, ranges were reported as minimum-maximum values with citation to source studies, without statistical pooling or meta-regression.

Challenges and knowledge gaps. This narrative review employed modified PRISMA documentation for literature identification and selection but lacks formal PRISMA protocol registration and did not employ validated risk-of-bias assessment tools such as Cochrane Risk of Bias 2.0 or ROBINS-I. Dual independent screening was applied to title/abstract selection and data extraction verification but not to full-text assessment. Quality assessment was qualitative rather than employing quantitative scoring systems with inter-rater reliability metrics. Substantial heterogeneity in interventions, study designs, patient populations, and outcome measures precluded meta-analysis and calculation of summary effect estimates with confidence intervals. Publication bias likely results in overrepresentation of positive preclinical results, as studies demonstrating therapeutic failure or adverse outcomes are less frequently published. Geographic bias is evident, with 67% of trials originating from Western countries (United States, Europe, Australia) despite low- and middle-income countries bearing 75% of global diabetes burden, limiting generalizability of findings to diverse healthcare systems and patient populations. Language restriction to English publications excluded an estimated 5% of relevant literature based on database filters. Most clinical trials have follow-up durations less than 2 years, preventing assessment of long-term treatment durability, late adverse events, or sustained efficacy beyond initial observation periods. Single reviewer data extraction with verification rather than duplicate independent extraction may introduce extraction errors, though consensus review of discrepancies mitigates this risk. The rapidly evolving gene therapy landscape means data are current only through June 2025; recent developments in vector engineering, editing technologies, or clinical trial outcomes after this date are not captured.

Part I: Foundational technologies and mechanisms

Gene therapy delivery technologies for diabetes

Viral vector delivery systems

Viral vectors remain the most clinically developed delivery platforms for diabetes gene therapy, each with distinct advantages and limitations [18,19].

Adeno-associated virus (AAV) vectors exhibit low immunogenicity, sustained transgene expression (>4 years in preclinical models), and tissue-specific tropism via multiple serotypes [20,21]. AAV2 demonstrates preferential transduction of pancreatic β-cells in preclinical models [22], while AAV8 exhibits enhanced liver tropism [23]. Critical limitations include restricted payload capacity (4.7 kb), pre-existing neutralizing antibodies in 58.5% of candidates for AAV2 (30-60% across serotypes), and 6-8.7-fold lower transduction efficiency in human tissue compared to rodent models [23,24].

Lentiviral vectors enable stable genomic integration, facilitating permanent genetic modifications advantageous for ex vivo cell engineering [25]. Contemporary self-inactivating (SIN) designs mitigate insertional mutagenesis risks, and tissue-specific promoters restrict transgene expression to target cells [26]. Applications include hepatic insulin gene delivery and pancreatic cell reprogramming via transcription factor delivery [27,28]. Integration-associated genotoxicity remains a theoretical concern requiring long-term monitoring [29].

Adenoviral vectors achieve high transduction efficiency and rapid transgene expression but induce strong immunogenicity limiting repeated administration and expression duration (1-2 weeks) [30,31]. Applications are restricted to proof-of-concept studies and short-term interventions rather than durable diabetes therapy [32].

Non-viral delivery systems

Non-viral systems including lipid nanoparticles (LNPs) and liposomes offer reduced immunogenicity, unlimited payload capacity, and simplified manufacturing compared to viral vectors [33]. Successful applications include delivery of insulin genes, CRISPR-Cas9 components, and regulatory RNAs for metabolic pathway modulation [34]. However, transduction efficiency (5-20%) remains substantially lower than optimized viral vectors (40-70% in preclinical models), necessitating further optimization before clinical competitiveness [35]. Manufacturing costs are 10-100-fold lower than AAV ($500-$5,000 per dose vs. $50,000-$100,000), representing the most viable pathway to scalable population-level access if efficacy can be demonstrated [36].

CRISPR-Cas9 gene editing platforms

Technological framework. CRISPR-Cas9 is a ground-breaking gene-editing technology that provides exceptional accuracy for targeting specific DNA sequences and altering genetic components related to diabetes mellitus [37]. The CRISPR-Cas9 system functions via a programmable dual mechanism, employing a guide RNA (gRNA) to direct Cas9 nucleases to complementary genomic sequences, resulting in precise double-strand breaks [38]. Breaks can be repaired through two main cellular repair pathways: nonhomologous end joining (NHEJ), which frequently leads to gene knockout via insertions or deletions, and homology-directed repair (HDR), which allows for accurate gene correction when a donor template is available [38]. This technology signifies a shift from conventional gene therapy methods by facilitating direct genome editing, instead of simply introducing exogenous genetic material. It is especially beneficial for rectifying genetic defects associated with monogenic diabetes or altering cellular functions in complex diabetic pathophysiologies. A comprehensive overview of the four principal gene therapy strategies discussed in this review. These approaches are not mutually exclusive; combination therapies integrating multiple modalities may ultimately prove most effective for achieving durable glycemic control and preventing disease progression.

Applications in type 1 diabetes. The CRISPR-Cas9 technology has shown considerable promise in protecting pancreatic β-cells from autoimmune damage and facilitating their regeneration in T1DM [37]. The following discussion begins with preclinical proof-of-concept research in animal models, followed by an examination of translational progress toward human applications, differentiating between laboratory results and clinical preparedness.

Preclinical evidence. Research has demonstrated that the targeted deletion of specific genes can confer resistance to autoimmune attacks in pancreatic β-cells. Additionally, results demonstrates that cells deficient in particular stress-response genes showed increased resilience to inflammatory damage and destruction mediated by cytotoxic T cells [39]. This approach constitutes a novel therapeutic strategy that targets the underlying autoimmune aspect of T1DM instead of solely focusing on the replacement of lost insulin production.

Toward clinical translation. Recent advancements in CRISPR-based methodologies have focused on modulating the immune system to inhibit the progression of T1DM through intricate cellular engineering techniques [40]. Ex vivo genetic modifications of human embryonic stem cells (hESCs) have been used to generate immune-evasive pancreatic cells for transplantation. This was achieved through the creation of edited clonal hESC lines that are deficient in the β2-microglobulin (B2M) gene, a critical element of the MHC-I pathway [41]. This genetic modification allows engineered cells to avoid allogeneic rejection while preserving their ability to differentiate into functional insulin-producing cells, potentially offering a renewable source of transplantable β-cells for patients with T1DM.

Applications in type 2 diabetes. The CRISPR-Cas9 technology can be used to tackle the intricate metabolic dysfunction associated with T2DM through the precise alteration of genes related to glucose metabolism, insulin sensitivity, and inflammatory pathways [42]. This approach has the potential to rectify underlying metabolic defects instead of simply addressing their symptoms, signifying a significant shift in treatment strategies for T2DM. The analysis starts by reviewing preclinical mechanistic findings that pertain to T1DM, and then proceeds to address translational challenges that are unique to T2DM due to the metabolic complexity of the disease.

One especially promising application focuses on addressing chronic inflammation, which is a significant contributor to insulin resistance in T2DM. The NLRP3 inflammasome is a significant therapeutic target in the pathogenesis of T2DM, as its activation results in the release of pro-inflammatory cytokines, including IL-1β and IL-18, which directly disrupt insulin signaling pathways [43,44]. CRISPR-based methods for targeting NLRP3 are being investigated, and delivery techniques mediated by nanoparticles seem to hold promise for gene editing specific to macrophages [45]. Reduced release of inflammatory cytokines and increased insulin sensitivity would result from a successful deployment.

In addition to targeting inflammasomes, novel metabolic gene therapy approaches focus on improving lipid dysregulation associated with insulin resistance. The knockdown of the PCSK9 gene or CRISPR-mediated disruption exhibits potential for the management of diabetic dyslipidemia, since preclinical studies indicate durable reductions in LDL levels and enhanced insulin sensitivity [46]. Similarly, modification of the ANGPTL3 and ANGPTL4 pathways enhances lipid metabolism and may improve glucose homeostasis in T2DM mice [47]. CRISPR interference (CRISPRi) and activation (CRISPRa) technologies provide reversible transcriptional regulation without permanent genomic alteration, providing adjustable control over metabolic gene expression [48]. These strategies may facilitate the modulation of glucose transporter expression, insulin receptor signaling components, or incretin pathway genes to reestablish metabolic equilibrium.

These technological platforms enable diverse therapeutic strategies targeting different aspects of diabetes pathophysiology, as detailed in the following section.

Therapeutic mechanisms and preclinical evidence

While foundational mechanisms (hepatic insulin production, transcription factor reprogramming) were established in 2007-2013 studies, this section synthesizes them to contextualize why direct translation has failed (Section 6.6) and inform the research priority framework (Section 6.6.4). The synthesis connects established mechanisms to current translational barriers not addressed in original publications.

Insulin replacement strategies

Gene therapy for diabetes can be classified into two main strategies: replacement strategies aimed at restoring insulin production and protective strategies designed to maintain existing β-cell function [49,50]. Replacement strategies seek to address diminished insulin production by developing alternative cell types capable of insulin synthesis and secretion, whereas protective strategies concentrate on averting additional β-cell loss via immune modulation or cytoprotection. The key challenge in replacement strategies is the replication of the complex glucose-sensing and insulin-secreting functions of pancreatic β-cells in other cell types. This necessitates not only the production of insulin but also the capacity for appropriate glucose responsiveness and regulated secretion [9].

The complexity of this challenge is a result of the intricate molecular machinery of native β-cells, which includes glucose transporters (GLUT2 in rodents; GLUT1 and GLUT3 predominantly in humans), glucokinase that functions as a glucose sensor, and calcium-dependent exocytosis mechanisms that enable precise insulin production in response to physiological fluctuations in glucose levels [51]. The majority of the GLUT2 literature is based on rodent models, necessitating that translational strategies consider species-specific differences in glucose transport mechanisms.

Hepatic insulin production

The liver is a potential target for insulin gene therapy considering its key function in glucose metabolism, vital protein synthesis capacity, and ease of gene delivery [9]. Hepatocytes offer distinct benefits for insulin gene therapy due to their natural glucose-sensing capabilities via glucokinase expression. They can be genetically engineered to secrete insulin in response to variations in glucose levels, thereby replicating the physiological insulin secretion patterns of pancreatic β-cells [52].

Preclinical studies have shown significant therapeutic efficacy with hepatic insulin gene therapy methods. Ren et al. conducted a significant study utilizing a lentiviral vector featuring an HIV/murine stem cell virus hybrid long terminal repeat promoter to transduce hepatocytes for insulin expression. This approach incorporates furin-cleavable sequences, facilitating appropriate insulin processing and maturation. The research demonstrated transduction efficiencies of 87 ± 1.2% in vitro and 60 ± 3.2% in vivo, indicating the viability of effective hepatic gene delivery. Following transduction, diabetic rats exhibited a significant reduction in blood glucose levels, normalizing to the same level of those of non-diabetic controls within 5 days, demonstrating rapid therapeutic efficacy [27]. The durability of this effect is significant, as sustained glucose control was observed throughout the study period, indicating potential for long-term therapeutic benefit.

Alternative cell type targeting

Multiple types of cells have been studied as surrogate insulin-producing cells, each presenting unique benefits for diabetes gene therapy applications. Alternative targets comprise three cell types: keratinocytes, which provide an accessible and renewable autologous cell source; muscle cells, which offer high protein synthesis capacity and efficient vector transduction; and enteroendocrine K cells, which possess natural endocrine properties similar to pancreatic β-cells [23,53,54].

Keratinocytes are notable candidates partly due to their self-renewal capabilities and potential for regulated insulin delivery. Tian et al. illustrated the viability of this method by genetically modifying human epidermal keratinocytes to synthesize a complex fusion protein comprising proinsulin with furin-cleavage sites, facilitating appropriate insulin maturation and processing. This system integrated rapamycin-inducible regulation, enabling controllable insulin production with precise temporal regulation in response to external stimuli, potentially facilitating patient-controlled insulin delivery [54]. This controllable system signifies a notable improvement over constitutive insulin expression, as it enables the adjustment of insulin production based on individual patient requirements and glucose concentrations.

Physiologically appropriate insulin replacement requires glucose-responsive secretion mechanisms replicating native β-cell function. Strategies include co-delivery of insulin and glucokinase genes enabling glucose-sensing in hepatocytes [23], synthetic glucose-responsive promoters [22], or targeting of cells with endogenous glucose-sensing machinery. Species differences in glucose transporters (GLUT2 in rodent β-cells vs. GLUT1/GLUT3 in human β-cells) complicate translation of glucose-responsive systems validated in preclinical models [51].

β-cell regeneration and transdifferentiation

Transcription factor-mediated reprogramming

Rather than introducing insulin genes into non-endocrine cells, an alternate approach is reprogramming existing pancreatic or hepatic cells into functioning β-cell-like cells. This method utilizes the developmental transcription factors that inherently regulate pancreatic cell fate determination.

Key transcription factors

The integration of critical transcription factors PDX-1, NEUROG3, and MAFA has shown considerable effectiveness in converting pancreatic α-cells into functional β-cells, using the developmental pathways that intrinsically govern pancreatic cell fate determination [55]. This technique is especially intriguing since it utilizes indigenous pancreatic cells that already possess certain molecular components essential for hormone production and glucose detection, hence simplifying the engineering process in comparison to hepatocyte or keratinocyte methods.

Therapeutic applications - hepatocyte transdifferentiation

The PDX-1 (pancreatic duodenal homeobox 1) functions as a principal regulator of pancreatic development and β-cell activity, establishing it as an essential target for reprogramming approaches. Fodor et al. investigated the therapeutic potential of PDX-1-mediated reprogramming through hepatocyte transdifferentiation, utilizing lentiviral vectors that encode PDX-1 regulated by an elongation factor-1α (EF1α) promoter [28].

The transduced hepatocytes differentiated into insulin-producing cells that demonstrated glucose-responsive insulin secretion, with insulin release quantified at 48±16 ng per mg of cell lysate protein over 2 hours following stimulation with 25 mM glucose. The glucose responsiveness observed is significant (p<0.05), indicating that the reprogrammed cells have gained both the ability to produce insulin and the complex glucose-sensing mechanisms required for physiologically appropriate insulin release. This indicates the practicality of transcription factor-mediated conversion as a legitimate gene therapy approach for diabetes.

Growth factor gene therapy

Growth factor gene therapy is a targeted strategy aimed at mitigating diabetic complications, specifically those related to vascular dysfunction, neuropathy, and impaired wound healing prevalent among diabetes patients. This strategy aims to deliver genes that encode therapeutic growth factors, which can facilitate tissue repair, improve vascularization, and restore normal cellular functions in tissues affected by diabetes.

Clinical studies of intramuscular injections of VM202, a plasmid expressing hepatocyte growth factor, and bicistronic VEGF165/HGF plasmids have shown notable therapeutic advantages in diabetic patients with complications. Reported outcomes of these treatments include significant pain reduction in diabetic neuropathy, enhanced healing of ischemic lesions in diabetic foot syndrome, and increased angiogenesis in poorly vascularized tissues [56,57]. The therapeutic mechanism comprises two complementary growth factors: VEGF (vascular endothelial growth factor), which specifically promotes the formation of new blood vessels and enhances tissue perfusion, and HGF (hepatocyte growth factor), which has diverse roles in cell growth, motility, tissue regeneration, and angiogenesis.

The efficacy of this combination of growth factors is notable, as diabetic complications frequently encompass both vascular insufficiency and compromised tissue repair mechanisms. This approach simultaneously promotes angiogenesis and enhances cellular regenerative capacity, addressing the underlying pathophysiology of diabetic complications rather than merely managing symptoms.

Having established the technological platforms and molecular mechanisms, the following sections examine preclinical validation in animal models, clinical translation to human trials, and systematic barriers limiting translational success.

Part II: Preclinical evidence and Proof-Of-concept

Preclinical studies in animal models

This section synthesizes evidence from preclinical animal studies to evaluate the translational pathway of diabetes gene therapy from laboratory proof-of-concept to clinical application. Quantitative meta-analysis was not performed due to: (1) heterogeneity in therapeutic approaches (viral vectors, cell therapies, CRISPR) preventing pooled analysis; (2) majority of trials (83.2%) in Phase I-II without comparable efficacy endpoints; (3) absence of head-to-head comparisons between modalities. This review instead provides qualitative synthesis identifying common translational barriers (Section 6) applicable across therapeutic modalities.

Preclinical animal studies

Preclinical animal studies have established the foundational evidence for the biological feasibility and therapeutic potential of gene therapy approaches in diabetes, employing various animal models that closely mimic human diabetic pathophysiology. These studies have been crucial in demonstrating proof-of-concept for clinical translation and refining delivery strategies prior to human trials.

Long-term efficacy studies

Long-term efficacy studies have shown notably positive outcomes. Callejas et al. conducted an innovative study that involved the intramuscular administration of adeno-associated viral vectors containing human insulin and glucokinase genes to streptozotocin-induced diabetic mice. The research demonstrated sustained glycemic control and metabolic stability for more than four years, indicating long-term enhancement in glucose homeostasis and insulin production, with no significant adverse effects [23]. The incorporation of glucokinase with insulin is essential, as this glucose-sensing enzyme facilitates glucose-responsive insulin production, thereby more accurately reflecting physiological β-cell function.

Lentiviral hepatic insulin production

Elsner et al. achieved significant therapeutic results through lentiviral-mediated hepatic insulin gene therapy in diabetic rats, resulting in sustained normalization of blood glucose levels for over a year. This research demonstrated diabetes reversal in animal models, with treated subjects sustaining euglycemia similar to that of non-diabetic controls [58]. Further research indicates therapeutic advantages extending beyond glucose regulation, such as enhanced wound healing via VEGF-C gene therapy, which facilitated healing by around 20% in diabetic animal models [59].

The preclinical studies highlight the potential of gene therapy for T1DM and T2DM, showing improved glycemic control, increased insulin production, and significant reduction of diabetes-related complications, including impaired wound healing and vascular dysfunction.

Translational challenges and limitations

Although these preclinical studies provide proof-of-concept, major limitations hinder translation to clinical trials. Transduction effectiveness varies significantly between rodent and human tissue, with rat studies achieving 60-87% hepatocyte transduction and human tissue showing ~10% peripheral islet cell transduction. This suggests that therapeutic thresholds may need greater vector dosages and increased immunogenicity risks. Second, animal models do not fully reproduce human pathophysiology. Streptozotocin-induced diabetes results in acute β-cell death, rather than the persistent autoimmune mechanisms seen in human T1DM. It also lacks immunological memory and regulatory failure. NOD mice (db/db) acquire diabetes at a faster pace than human T2DM, which typically takes decades to develop. Only 25% of the assessed preclinical research used immunologically relevant models. Glucose metabolism variations between rodents and humans (GLUT2 vs. GLUT1/3) impact glucose-sensing dynamics, leaving appropriate human regulation aspects unknown. Follow-up lengths of more than 4 years in Callejas et al. study are inadequate to evaluate long-term safety and effectiveness criteria relative to human lifespan requirements. Rare adverse events are difficult to identify due to small sample numbers (n=6–12) [33]. Efficacy may be overstated due to publication bias, since 83% of preclinical trials reported outcomes that were significant. Despite its importance, preclinical success does not adequately predict clinical results, as shown by these limitations.

Table 1A summarizes the key preclinical studies establishing proof-of-concept, including study design, quality assessment scores, and translational significance.

Table 1A.

Foundational preclinical studies supporting clinical translation.

Study	Intervention	Animal Model	Study Duration	Study Design	Control Group	Primary Outcomes	Key Results	Significance for Translation
Callejas et al. 2013 (33)	AAV-insulin + glucokinase gene therapy via intramuscular injection	STZ-induced diabetic mice	>4 years	Randomized, controlled	STZ-diabetic untreated mice	Long-term glycemic control, safety assessment	Sustained normoglycemia >4 years; no adverse effects; glucose-responsive insulin secretion	Longest duration preclinical data demonstrating durability; co-delivery of glucokinase with insulin enables glucose-responsiveness; translational barrier: 60-87% transduction in mice vs ∼10% in human islets
Ren et al. 2007 (37)	Lentiviral hepatic insulin gene therapy with furin-cleavable sequences	STZ-induced diabetic rats	1 year	Controlled efficacy study	STZ-diabetic untreated rats	Glucose normalization, duration of effect	87±1.2% in vitro transduction, 60±3.2% in vivo; normoglycemia within 5 days; sustained >1 year	Demonstrates rapid therapeutic effect and glucose-responsive secretion from hepatocytes; translational barrier: species differences in GLUT2 (rodent) vs GLUT1/3 (human) glucose-sensing
Fodor et al. 2007 (38)	Lentiviral PDX-1-mediated hepatocyte transdifferentiation	Diabetic mice (ex vivo modified rat hepatocytes transplanted)	2 hours glucose stimulation assay	Ex vivo transduction followed by transplant	Non-transduced hepatocytes	Glucose-responsive insulin secretion	48±16 ng insulin per mg protein over 2h with 25 mM glucose (p<0.05 vs basal)	Proof-of-concept for transcription factor-mediated reprogramming; translational barrier: no human clinical trials of in vivo reprogramming initiated as of 2025

AAV: adeno-associated virus, GLUT: glucose transporter, PDX-1: pancreatic and duodenal homeobox 1, STZ: streptozotocin.

Preclinical studies demonstrate biological feasibility of gene therapy for diabetes across multiple modalities. However, translation to human clinical trials has revealed systematic barriers not apparent in animal models, as detailed in the following sections.

Part III: Clinical translation and human trials

Clinical trial development and outcomes

Direct in vivo gene delivery approaches

Direct AAV-mediated gene therapy: Translational delays

Despite promising preclinical outcomes demonstrating glucose-responsive hepatic insulin expression in mice [22] and >4-year glycemic control through AAV-insulin/glucokinase co-delivery [23], direct AAV-insulin approaches have not advanced to published Phase II/III efficacy trials as of mid-2025, reflecting substantial translational barriers not apparent in animal models.

First-in-human type 1 diabetes trial

The domain of diabetes gene therapy has achieved a significant milestone with the beginning of the first clinical trials aimed at evaluating gene therapy strategies for T1DM, signifying a shift from years of preclinical investigation to human implementation.

Australian AAV-islet trial: first-in-human enrollment without published data. The Australian trial (GARV-AAV2-A20-Islet product) represents the first registered Phase I investigation of direct AAV-mediated delivery to type 1 diabetes patients [60]. As of June 2025, this trial remains in early enrollment without peer-reviewed publications, conference presentations, or interim data releases. As a Phase I safety study, it appropriately focuses on safety and feasibility rather than efficacy. The trial's significance lies in demonstrating regulatory approval for direct AAV-insulin approaches is achievable, not in proven clinical success. Without published transduction efficiency and clinical outcome data, it remains uncertain whether the 6-8.7-fold efficiency gap between rodent models (60-87% hepatocyte transduction) and human tissue (~10% peripheral islet cells) [27,61] has been successfully addressed.

Reasons for limited clinical progression

Direct AAV-insulin gene therapy approaches remain in Phase I exploration despite over 15 years of preclinical development due to convergent biological and economic barriers. The 6-8.7-fold reduction in transduction efficiency between rodent models and human tissue creates uncertainty about whether sufficient insulin-producing cells can be generated to achieve glycemic control. Pre-existing neutralizing antibodies against AAV2 in 58.5% of the population, and against other serotypes in 30-60% of candidates, substantially restricts eligible patient populations. Manufacturing costs of $50,000-$100,000 per therapeutic dose limit trial enrollment and feasibility studies. Most critically, the absence of published human efficacy data from ongoing Phase I trials prevents assessment of whether these approaches can overcome the efficiency and immunity barriers identified in preclinical-to-clinical translation. Until published data demonstrate therapeutically adequate transduction rates and glucose-responsive insulin production in human subjects, advancement to efficacy-stage trials remains speculative rather than evidence-based.

Ex vivo cell-based gene therapy approaches

Direct AAV vector methods contrast fundamentally with CRISPR-edited cell therapies (VCTX210) and stem cell-derived islet transplantation (zimislecel), which utilize ex vivo procedures enabling validation of transduction efficiency, glucose-responsiveness, and transgene expression before transplantation [62,63]. These alternative approaches have published Phase II/III efficacy data: zimislecel demonstrates 83% insulin independence at 12 months in 12 patients [62], while VCTX210 reports preliminary immune evasion signals [63]. The clinical development disparity, ex vivo therapies reaching efficacy-stage trials while in vivo AAV-insulin delivery remains in Phase I enrollment underscores technical challenges inherent to achieving sufficient transduction efficiency, physiologically appropriate glucose-responsive control, and sustained expression despite neutralizing antibodies (58.5% anti-AAV2 seroprevalence) in human subjects [61,64].

CRISPR-based cell therapy trials

The collaboration between ViaCyte and CRISPR Therapeutics has initiated a significant first-in-human study (NCT05210530) assessing VCTX210, an innovative method that integrates cell therapy with gene editing technology [63]. This trial utilizes CRISPR/Cas9-modified human embryonic stem cell-derived pancreatic endoderm cells that have been genetically modified to avoid immune system detection by eradicating β2-microglobulin, thereby reducing MHC class I expression. The gene-edited cells are secured within a biocompatible delivery device that is intended to protect them from immune system attack while simultaneously facilitating the exchange of glucose and insulin. Preliminary reports from the ongoing trial indicate early signals consistent with immune evasion and detectable C-peptide production in some recipients [66], though comprehensive safety and efficacy data await peer-reviewed publication.

Stem cell-derived islet replacement therapy

Phase I/II/III outcomes

Vertex Pharmaceuticals' zimislecel (VX-880), a fully differentiated islet cell therapy derived from allogeneic stem cells, has shown significant potential in the Phase 1/2/3 FORWARD-101 trial (NCT04786262). Recent findings discussed at the American Diabetes Association 85^th Scientific Sessions (June 2025) and concurrently published in the New England Journal of Medicine indicate that all 12 patients (Part B and C) administered full-dose zimislecel met ADA-recommended glycemic targets (HbA1c <7.0% and >70% time-in-range) at the 12-month follow-up [62]. Notably, 10 out of 12 patients (83%) became completely insulin independent, and the average amount of exogenous insulin used by all participants decreased by 92%. All patients exhibited a complete cessation of severe hypoglycemic events from Day 90 onwards during the evaluation period [62]. Data from the 61^st European Association for the Study of Diabetes Annual Meeting (September 2025) corroborated these findings, showing that participants attained 95% time-in-range compared to 55.5% at baseline, alongside the total cessation of exogenous insulin needs (mean baseline: 36.1 units/day) [65]. The trial has increased enrollment to around 50 patients and is advancing toward pivotal Phase 3 development [62].

Critical limitations and considerations

Essential factors to consider: Although these findings indicate significant progress in the treatment of T1DM, various limitations must be acknowledged. Zimislecel necessitates chronic immunosuppressive therapy to prevent allogeneic rejection, which results in a considerable medication burden and associated risks such as increased susceptibility to infections, malignancy, and nephrotoxicity. Vertex's parallel encapsulated islet therapy program (VX-264), designed to eliminate the need for immunosuppression by utilizing stem cell-derived islets within an immunoprotective device, was discontinued after not meeting performance milestones at the 90-day mark [66], highlighting the technical challenges associated with immune evasion strategies. Additionally, the existing zimislecel data originates from a limited cohort (n=12 at full dose) with a one-year follow-up; the long-term durability and applicability to larger populations have yet to be determined. Third, two patient deaths occurred during the trial; FDA preliminary assessment concluded no treatment causality (Source: Vertex Pharmaceuticals Clinical Trial Update, FDA Breakthrough Therapy Designation documentation, June 2025; formal FDA safety report pending peer-reviewed publication) [62]. Hepatic portal vein infusion presents procedural risks and necessitates specialized interventional expertise, which may restrict accessibility.

Comparison: ex vivo vs. in vivo approaches

Ex vivo and in vivo gene therapy strategies represent fundamentally different translational paradigms with distinct advantages and limitations. Ex vivo approaches engineer cells outside the body, enabling validation of transduction efficiency, glucose-responsive insulin secretion, and transgene expression before transplantation into patients. This pre-validation reduces uncertainty and allows rejection of inadequately modified cell batches before clinical administration. Quality control measures can confirm that cells meet predefined functional criteria, and patients receive only cells demonstrating appropriate glucose sensing and insulin production in controlled laboratory assays. In contrast, in vivo approaches deliver vectors directly to patients, where transduction efficiency, immune responses, and functional outcomes cannot be validated until after administration. The clinical development disparity reflects these differences: ex vivo cell therapies initiated in 2018 achieved Phase III outcomes by 2025, whereas in vivo AAV-insulin delivery begun in 2008 remains in Phase I enrollment without published efficacy data, demonstrating that pre-validation capabilities substantially accelerate clinical translation despite apparent simplicity of direct vector administration. Table 1B summarizes clinical trials for type 1 diabetes gene therapy, including design characteristics, primary outcomes, and quality assessment.

Table 1B.

Clinical trials of gene therapy for Type 1 diabetes.

	Intervention	Study/Trial	Phase	Sample Size (n) & Follow-up	Study Design	Control Group	Blinding	Primary Outcomes	Key Results	Endpoint Met	Clinical Significance
CRISPR-edited hESC-derived pancreatic cells (B2M knockout)	VCTX210 (ViaCyte/CRISPR) NCT05210530	I/II	Not disclosed; ≥6 months	Single-arm, open-label	None (single-arm)	Open-label	Safety, immune evasion, C-peptide production	Preliminary data: early signals of immune evasion and detectable C-peptide levels in subset of recipients	Pending	First CRISPR-edited cell therapy for diabetes
Genetically modified pancreatic islet cells via AAV vectors (GARV-AAV2-A20-Islet)	Australian AAV-islet trial	I	Enrolling; follow-up pending	Dose-escalation, single-arm	None (Phase I safety)	Open-label	Safety, feasibility of AAV-mediated islet modification	Ongoing enrollment, preliminary safety data positive	Pending	Direct pancreatic targeting approach validation
Allogeneic stem cell-derived islet cells	Zimislecel/VX-880 (Vertex) NCT04786262	I/II/III	n=12 (full dose); 12 months	Single-arm, open-label	Prospective matched controls	Open-label	Elimination of severe hypoglycemic events, HbA1c <7%	All 12 achieved HbA1c <7%, >70% TIR; 10/12 (83%) insulin-independent; 100% elimination of SHEs	Yes	Demonstrated functional β-cell replacement with glucose-responsive insulin production

AAV: adeno-associated virus, B2M: β2-microglobulin, HbA1c: hemoglobin A1c, hESC: human embryonic stem cell, SHE: severe hypoglycemic event, T1DM: type 1 diabetes mellitus, TIR: time in range.

Gene therapy for diabetic complications

Gene therapy strategies targeting diabetic complications have progressed more rapidly to late-phase trials than direct insulin gene delivery approaches. VM202 (hepatocyte growth factor plasmid) demonstrated acceptable safety in a Phase III randomized controlled trial (n=500) for diabetic peripheral neuropathy, though primary efficacy endpoints were not met in the main study [67]. A 12-month extension study (n=101) showed statistically significant pain reduction compared to placebo, particularly in patients not receiving concurrent gabapentin or pregabalin, with sustained symptom relief for eight months post-treatment [67]. For critical limb ischemia in type 2 diabetes patients, intramuscular injection of bicistronic VEGF165/HGF plasmid demonstrated statistically significant improvements in ankle-brachial index, serum VEGF levels, rest pain scores, and vascularization on imaging within three months [56]. These trials provide proof-of-concept for gene therapy applications addressing vascular and neurological complications beyond glycemic control. Detailed trial characteristics, outcomes, and endpoints are presented in Table 1C.

Table 1C.

Clinical trials of gene therapy for diabetic complications.

Study/Trial	Intervention	Target Population	Phase	Sample Size (n) & Follow-up	Study Design	Control Group	Blinding	Primary Outcomes	Key Results	Endpoint Met	Clinical Significance
VM202 Neuropathy Trial (Kessler et al.)	Plasmid DNA expressing dual HGF isoforms (intramuscular)	Diabetic peripheral neuropathy	III	n=500 (9 months); n=101 extension (12 months)	Randomized, placebo-controlled	Placebo injections (saline)	Double-blind (main study); open-label (extension)	Pain reduction (24-h NRS pain score), neurological function improvement	Primary study (n=500) did not meet efficacy endpoint; extension study (n=101) showed significant pain reduction vs placebo, especially in patients not on gabapentin/ pregabalin; pain relief sustained 8 months post-treatment	Mixed: No (primary); Yes (extension)	First gene therapy for diabetic neuropathy; demonstrated safety and potential disease-modifying effects in subset analysis, though mixed Phase III results require additional validation
VEGF/HGF critical limb ischemia (Barc et al.)	Bicistronic plasmid (pIRES/ VEGF165/ HGF) (intramuscular)	T2DM with critical limb ischemia	II	Not disclosed; 3 months	Single-arm, open-label	None (single-arm Phase II)	Open-label	Vascularization, limb salvage, pain reduction	Improved ABI from baseline, elevated serum VEGF, notable decreases in rest pain scores, enhanced vascularization on imaging	Yes	Vascular regeneration approach for limb-threatening diabetic complications

ABI,: ankle-brachial index, HGF: hepatocyte growth factor, ITT: intent-to-treat, T2DM: type 2 diabetes mellitus, VEGF: vascular endothelial growth factor.

Emerging clinical programs

Type 2 diabetes gene therapies (GLP-1 gene therapy)

RJVA-001, an investigational AAV-based gene therapy aimed at facilitating endogenous pancreatic GLP-1 production, has obtained regulatory approval to commence clinical trials in patients with T2DM in 2026 [68]. This single-administration method seeks to achieve prolonged glucose regulation and weight management without the need for ongoing medication. The therapy mitigates the limitations of existing GLP-1 receptor agonists, such as the requirement for daily or weekly injections and high costs, by engineering a physiologically responsive GLP-1 secretion mechanism. Permanent genomic modification presents issues related to dose titration flexibility, reversibility, and long-term safety, necessitating thorough assessment during clinical development.

Cell therapy for diabetic kidney disease

Rilparencel (REACT), a renal autologous cell treatment, is being evaluated in patients with chronic kidney disease who have type 2 diabetes in the Phase 3 REGEN-006 (PROACT 1) trial (NCT05099770) [69]. This signifies a change in focus from aiming to manage blood sugar levels to treating complications of diabetes via stem cell treatment. Efficacy results are still awaited, but the autologous technique removes the need for immunosuppression.

The various therapeutic modalities presently under clinical development comprising direct viral gene delivery, ex vivo gene-edited cell transplantation, allogeneic stem cell-derived islets, and targeted therapies for diabetic complications constitute complementary strategies aimed at achieving clinically effective diabetes treatments.

Current clinical development pipeline

Analysis of 143 registered trials was conducted through ClinicalTrials.gov search (terms: ‘gene therapy’ AND ‘diabetes’, January 2010-September 2024). A complete trial registry table is available upon request but was excluded due to manuscript length constraints. Analysis indicates that diabetes gene therapies are primarily in early-phase development, with 83.2% in Phases I-II and only 11.2% in Phase III/IV. This suggests that the field remains exploratory despite recent advancements.

Type 1 diabetes constitutes 69.2% of clinical trials, with China at the forefront of trial registration at 33.3%, followed by the United States and Europe. Nonetheless, representation from low- and middle-income countries is still inadequate, despite these regions shouldering 75% of the global diabetes burden. Additionally, emerging research from Asian centers increasingly reveals cost-effective manufacturing methods that merit greater consideration in forthcoming systematic reviews to guarantee equitable global access.

Amarna Therapeutics' AM510, an immunomodulatory gene therapy for T1DM, received FDA guidance in July 2024, with an Investigational New Drug (IND) submission expected within two years [70].

Fig. 1 synthesizes the clinical development landscape across therapeutic modalities, revealing systematic patterns in translational progression. Panel A demonstrates that the temporal gap between preclinical proof-of-concept and clinical efficacy trials varies dramatically by approach: direct AAV-insulin delivery (2008 preclinical milestone) has not advanced beyond Phase I enrollment in 17 years, whereas ex vivo CRISPR-edited cell therapies (2018 development) achieved Phase III outcomes within 7 years. This disparity directly reflects the translational barriers quantified in Table 2, particularly the inability to address the 6-8.7-fold transduction efficiency gap and 58-78% pre-existing immunity through incremental optimization of in vivo delivery approaches. Panel B reveals that 83.2% of diabetes gene therapy trials remain in Phase I-II, with only diabetic complication therapies (VM202, VEGF/HGF) reaching Phase III completion. The concentration of trials in early phases indicates the field is predominantly exploratory rather than approaching near-term regulatory approval, with realistic estimates for first FDA approval in 2028-2030 contingent on successful Phase III expansion of advanced therapies like zimislecel. The pipeline distribution underscores the conclusion from Section 6.6.4 that research priorities should emphasize ex vivo cell-based approaches with demonstrated translational feasibility over continued optimization of in vivo viral delivery approaches that have not overcome fundamental biological barriers despite extended development timelines.

Fig. 1.

Timeline of clinical development and current pipeline for diabetes gene therapy. A timeline detailing significant milestones from 2000 to 2025, illustrating the evolution from initial vector development and proof-of-concept studies to contemporary clinical trials. (A) Orange circles represent preclinical studies, while red circles denote clinical trials. Key milestones encompass initial AAV/lentiviral studies in 2000, methodological development for glucose-responsive insulin expression in 2008, a four-year efficacy demonstration in 2013, the development of CRISPR/immune-evasive cells in 2018, Phase III complication trials in 2021, the introduction of VCTX210 CRISPR-edited cell therapy in 2023, and over 15 active Phase I-III studies projected for 2025. (B) Current therapeutic approaches are categorized by strategy: direct insulin gene therapy, CRISPR-edited cell therapy, management of diabetic complications, β-cell regeneration, and immune modulation. The color-coding system denotes the following development phases: gray reflects Preclinical, green represents Phase I, yellow signifies Phase II, red indicates Phase III, blue corresponds to Regulatory considerations, purple pertains to Future directions. Each approach can include several phases across various trials. Abbreviations: AAV, adeno-associated virus; CRISPR, clustered regularly interspaced short palindromic repeats; HGF, hepatocyte growth factor; PDX-1, pancreatic and duodenal homeobox 1; T1DM, type 1 diabetes mellitus; VCTX210, ViaCyte CRISPR-edited cell product; VEGF, vascular endothelial growth factor; VM202, hepatocyte growth factor plasmid gene therapy. The figure was created using Microsoft PowerPoint.

Table 2.

Preclinical-to-clinical translation gaps in diabetes gene therapy.

Parameter	Preclinical Achievement	95% CI / SD1	Preclinical Quality Score2	Human Clinical Reality	95% CI / Sample Size	Clinical Data Quality3	Gap Analysis
Hepatocyte transduction	60-87% (Ren 2007: 60±3.2%; Callejas 2013: ∼70%)	56.8-63.2% (Ren); NR (Callejas)	7/8 (High)	∼10% peripheral cells only (Leibowitz 1999)	NR (descriptive study)	Moderate	6-8.7× efficiency gap
Duration to therapeutic effect	5 days normoglycemia (Ren 2007)	NR (n=8-12/group)	7/8 (High)	Not yet reported in published human trials	N/A	N/A	Data gap prevents assessment of translation success
Sustained efficacy	>4 years in mice (Callejas 2013)	NR (n=6-8/group)	8/8 (High)	<12 months follow-up (Phase I trials)	N/A (ongoing trials)	Low (Phase I)	Temporal scaling uncertain
Immunogenicity (anti-AAV antibodies)	Minimal (0-5%) in SPF mice	NR (pathogen-free conditions)	5/8 (Moderate)4	58.5% anti-AAV2 (Mingozzi 2011); 30-60% across serotypes	55.2-61.8% (Mingozzi, n=180)	High (seroprevalence n=180-500)	Species difference: 50-60% higher baseline immunity
Glucose responsiveness	Demonstrated with GLUT2/GCK system (Callejas 2013)	Blood glucose: 120±18 mg/dL at 4 years	7/8 (High)	Human GLUT1/3 system untested in vivo	N/A	N/A	Regulatory element mismatch
Manufacturing scale (cost per dose)	Proof-of-concept: <$5,000/batch	N/A (academic production)	N/A	$50-100K per clinical dose	Industry estimates	High	10-20× cost increase
Sample size	n=6-12 per group (typical)	Power calculations rarely reported	6/8 (Moderate)5	Phase I: n=3-12; Phase III: n=12 (zimislecel)	Small sample sizes limit statistical power	Low-Moderate	Inadequate power for rare adverse events

¹Preclinical 95% CI/SD: Confidence intervals or standard deviations reported in original publications. “NR” = not reported in source publication; “N/A” = not applicable for the parameter. Many preclinical studies report only p-values without CIs or SDs.

²Preclinical Quality Score: Based on 8 criteria (see Section 1.1): validated animal model, appropriate controls, adequate sample size (n≥6/group), randomization, blinding, reproducibility, statistical rigor, adverse event reporting. Score: ≥6/8 = High; 4-5/8 = Moderate; <4/8 = Low.

³Clinical Data Quality: High = Phase III RCT with peer-reviewed outcomes (n>100); Moderate = Phase II trials or published Phase I data (n=20-100); Low = Phase I enrollment only (n<20); N/A = No published human data available.

⁴Reduced score: Non-representative animal models (young, specific-pathogen-free, no prior AAV exposure) do not reflect 30-60% human anti-AAV seroprevalence.

⁵Reduced score: Acute STZ-induced diabetes does not replicate chronic autoimmune T1DM pathophysiology; small sample sizes (n=6-12) limit detection of rare adverse events.

⁶Data availability limitation: The majority of preclinical gene therapy studies published between 2007-2013 report only mean ± SD or p-values without calculating 95% confidence intervals. For clinical trials, confidence intervals are available for large seroprevalence studies (anti-AAV, anti-Cas9 antibodies) but not for early-phase gene therapy trials with small sample sizes (n=3-12). Where confidence intervals were not reported in original publications, this is noted as “NR.” This limitation reflects historical reporting standards in the field rather than methodological deficiency of individual studies.

Key Findings: The most robustly quantified translational gaps with available statistical confidence are: (1) pre-existing anti-AAV2 immunity affects 58.5% (95% CI: 55.2-61.8%) of candidates, excluding the majority from AAV2-based therapies; (2) anti-Cas9 immunity affects 58-78% depending on ortholog, with narrow confidence intervals indicating high certainty; (3) transduction efficiency shows 6-8.7-fold reduction though original publications lack formal confidence intervals for this comparison. Manufacturing cost increases (10-20-fold) are based on industry data rather than statistical estimates. The absence of confidence intervals for many parameters reflects data gaps requiring prospective human studies with adequate statistical power.

AAV: adeno-associated virus, Cas9: CRISPR-associated protein 9, CI: confidence interval, GCK: glucokinase, GLUT: glucose transporter, N/A: not applicable, NR: not reported, RCT: randomized controlled trial, SaCas9: Staphylococcus aureus Cas9, SD: standard deviation, SpCas9: Streptococcus pyogenes Cas9, SPF: specific pathogen-free, STZ: streptozotocin, T1DM: type 1 diabetes mellitus.

Clinical trials reveal a striking disparity in translational success: ex vivo cell therapies have advanced to Phase III outcomes while direct in vivo gene delivery remains in Phase I exploration despite longer development timelines. The following sections quantify the systematic barriers explaining this disparity and propose strategic research priorities to overcome them.

Part IV: Translational barriers and critical analysis

Systematic translational barriers from preclinical to clinical application

Quantitative analysis of preclinical-to-clinical gaps

Table 2 quantifies translational gaps across key parameters with statistical confidence intervals where available in published literature. The most robustly quantified barriers are pre-existing immunity to AAV (58.5%, 95% CI: 55.2-61.8%, n=180) and Cas9 proteins (58-78%, narrow confidence intervals), which exclude the majority of potential candidates. The 6-8.7-fold transduction efficiency gap, while consistently observed across multiple studies (Ren 2007: 60±3.2% in rats; Leibowitz 1999: ~10% in human islets), lacks formal confidence intervals due to historical reporting standards in the gene therapy field. The absence of confidence intervals for several critical parameters particularly glucose responsiveness in human tissue and clinical transduction efficiency reflects fundamental data gaps rather than methodological limitations of existing studies. These gaps highlight the need for prospective human trials with adequate statistical power (n≥30-50 per group) to generate precise translational estimates with narrow confidence intervals, enabling evidence-based decisions regarding clinical development pathways.

Biological and technical barriers

Transduction efficiency: The 6-8.7-Fold Gap

Preclinical vs. human tissue efficiency

The most significant quantitative barrier is the transduction efficiency gap between preclinical models and human tissue. Preclinical studies demonstrate 60-87% hepatocyte transduction in rodent models: Ren et al. (2007) achieved 87±1.2% in vitro and 60±3.2% in vivo hepatocyte transduction in diabetic rats using lentiviral vectors [27], while Callejas et al. (2013) demonstrated sustained therapeutic efficacy in mice with AAV-mediated gene delivery achieving comparable transduction rates [23]. In contrast, human pancreatic tissue shows dramatically lower transduction efficiency. Leibowitz et al. demonstrated that effective transduction of cells within pancreatic islets occurs exclusively at the periphery, affecting approximately 10% of cells, while core cells remain untransduced [61]. This represents a 6-8.7-fold efficiency gap that directly impacts therapeutic dosing requirements and clinical feasibility.

Critically, the Australian AAV-islet trial (Section 5.1.2), VCTX210, and other ongoing trials have not published strategies or preliminary data demonstrating successful bridging of this efficiency gap in human subjects. Without published human transduction efficiency data from these trials, it remains uncertain whether rodent-optimized vectors can achieve therapeutically adequate transduction in humans or whether fundamental re-engineering is required.

Strategies to address the efficiency gap (dose escalation limitations, capsid engineering, tissue alternatives)

Achieving therapeutic efficacy in humans will likely require one or more of the following approaches, none of which have been validated in published human trials as of mid-2025: (1) Substantially higher vector doses to compensate for lower per-cell transduction efficiency. However, this approach increases immunogenicity risk, as larger quantities of vector capsids trigger stronger innate and adaptive immune responses, potentially leading to vector neutralization and elimination of transduced cells. Additionally, manufacturing costs scale linearly with dose: current AAV production costs $50,000-$100,000 per therapeutic dose, meaning 6-fold higher doses to match rodent transduction rates would increase per-patient costs to $300,000-$600,000 for vector manufacturing alone, before accounting for clinical administration, monitoring, and management. (2) Novel capsid engineering for enhanced human β-cell or hepatocyte tropism beyond serotypes optimized in rodent models. Directed evolution, ancestral sequence reconstruction, and structure-guided rational design offer pathways to develop capsid variants with: (a) improved binding to human cell surface receptors; (b) enhanced intracellular trafficking and nuclear entry in human cells; (c) evasion of neutralizing antibodies through epitope modification; and (d) reduced innate immune recognition. However, capsid engineering requires extensive screening (typically 10⁶-10⁹ variants), validation in human tissue models, and demonstration of safety profiles comparable to natural serotypes, a development timeline of 5-10 years before clinical translation. (3) Alternative target tissues such as hepatocytes rather than pancreatic islet cells. Hepatocytes demonstrate higher transduction efficiency than islet cells in preclinical models and are more accessible via portal vein delivery. However, hepatocytes utilize different glucose-sensing mechanisms than pancreatic β-cells (GLUT1/3 and glucokinase expression patterns differ), creating uncertainty about physiologically appropriate glucose-responsive insulin secretion from engineered hepatocytes. The lack of published human data from hepatic insulin gene therapy trials prevents assessment of whether this tissue-switching strategy successfully addresses the efficiency gap. (4) Acceptance of lower transduction thresholds if engineered glucose-sensing mechanisms can enable adequate insulin production from smaller transduced cell populations. This approach requires demonstrating that 10-20% transduction efficiency is sufficient for glycemic control, contingent on: (a) highly efficient glucose-sensing systems that respond proportionally to physiological glucose fluctuations (4-10 mM); (b) sufficient insulin production per transduced cell to compensate for smaller transduced population; and (c) durability of transgene expression without silencing or immune clearance over multi-year timescales. No published human data validates that low transduction rates can achieve therapeutic outcomes comparable to native β-cell function.

Species-specific differences in immune responses

Beyond transduction efficiency, immune responses vary substantially between species. Humans exhibit markedly elevated pre-existing immunity to viral vectors (58.5% anti-AAV2 seroprevalence, 30-60% across serotypes) and CRISPR components (58-78% anti-Cas9 antibodies; see Section 5.2.1) compared to young, specific-pathogen-free laboratory animals with minimal environmental exposure to wild-type AAV or pathogenic bacteria. This pre-existing immunity can neutralize administered vectors before transduction occurs, eliminate transduced cells through antibody-dependent or T-cell-mediated cytotoxicity, or trigger inflammatory responses that cause systemic toxicity.

Metabolic and physiological complexity

The metabolic complexity and heterogeneity of human diabetes surpass those of animal models. Streptozotocin-induced diabetes produces acute, uniform β-cell ablation without the gradual autoimmune destruction, epitope spreading, and immunological memory that characterize human type 1 diabetes. Genetic models (NOD mice, db/db mice) develop diabetes on accelerated timelines (weeks to months) compared to human disease progression (months to years for T1DM, decades for T2DM), potentially missing important pathophysiological features. Additionally, glucose-sensing mechanisms differ between species: rodents predominantly utilize GLUT2 in β-cells, while humans utilize GLUT1 and GLUT3, affecting the translational relevance of glucose-responsive regulatory elements validated in rodent models.

Limited clinical data constrains assessment

Despite promising preclinical findings spanning 2007-2025, clinical translation is constrained by limited sample sizes (typical Phase I trials: n=6-12 patients), short follow-up durations (most ongoing trials have <2 years follow-up), and absence of randomized controlled comparisons in early-phase studies. Multiple factors contribute to slow clinical progression: stringent regulatory requirements for IND-enabling toxicology and biodistribution studies, manufacturing challenges in scaling from research-grade to GMP-grade vector production, and patient recruitment difficulties given restrictive eligibility criteria (exclusion based on anti-vector antibodies, age limits, BMI restrictions, complication severity thresholds).

Critical interpretation: Speculation versus evidence

Until ongoing trials publish transduction efficiency data in human subjects paired with corresponding clinical outcomes (glycemic control measured by HbA1c and continuous glucose monitoring, C-peptide levels indicating functional insulin production, insulin independence rates), the expectation of clinical success for direct in vivo gene delivery approaches remains speculative rather than evidence-based.

The disparity in clinical development between therapeutic modalities is striking: ex vivo cell therapy approaches have advanced to published Phase III outcomes (zimislecel: n=12, 12-month follow-up, 83% insulin independence), while in vivo vector delivery approaches remain in Phase I enrollment without published efficacy data despite >15 years of preclinical development since foundational studies (Ren 2007, Callejas 2013). This disparity suggests that the translational barriers documented in Table 2 particularly the 6-8.7-fold transduction efficiency gap represent fundamental biological constraints rather than engineering challenges amenable to incremental optimization of existing approaches.

Successful clinical translation of in vivo gene delivery for diabetes may require paradigm shifts in delivery technology (novel vectors, non-viral systems), target tissue selection (hepatocytes vs. islets), or acceptance that this approach will serve niche populations where ex vivo cell therapies are contraindicated, rather than achieving the broad applicability initially envisioned in early preclinical studies.

Immunological barriers to clinical translation

Pre-existing anti-AAV neutralizing antibodies

Pre-existing adaptive immunity to both viral vectors and CRISPR components represents a systematic barrier affecting patient eligibility across all gene therapy approaches for diabetes. Unlike acquired immunity that develops after therapeutic administration, pre-existing immunity results from prior environmental exposures to wild-type AAV (ubiquitous in human populations) and bacterial species expressing Cas9 orthologs (Staphylococcus aureus, Streptococcus pyogenes).

Seroprevalence of neutralizing antibodies varies by AAV serotype and geographic region, with global rates ranging from 30% to 60% of the population. AAV2, the most extensively studied serotype for pancreatic applications due to its demonstrated β-cell tropism in preclinical models, exhibits approximately 58.5% seroprevalence with significant geographic variation higher prevalence in developing countries correlates with greater wild-type AAV exposure in childhood. This high baseline seroprevalence excludes the majority of potential candidates from AAV2-based therapies unless alternative serotypes with lower cross-reactive immunity or immunosuppressive protocols during vector administration are employed.

Anti-Cas9 adaptive immunity

Pre-existing immunity to Cas9 proteins reflects common bacterial exposures throughout life. Studies demonstrate anti-Cas9 antibodies in 58% of donors for Streptococcus pyogenes Cas9 (SpCas9) and 78% for Staphylococcus aureus Cas9 (SaCas9), the two most commonly utilized orthologs in therapeutic applications. T-cell memory against SpCas9 is present in 67% of assessed individuals and against SaCas9 in 78%, indicating both humoral and cellular adaptive immune responses that could neutralize therapeutic delivery or eliminate transduced cells [71].

Implications for patient selection and trial eligibility

Comprehensive immunological screening is mandatory for patient selection in gene therapy trials, requiring: (1) quantitative measurement of anti-AAV neutralizing antibody titers for relevant serotypes; (2) anti-Cas9 antibody and T-cell response assays; and (3) HLA typing for allogeneic cell therapies. However, this screening may exclude 50-70% of potential candidates based on pre-existing immunity thresholds established in current trial protocols.

This necessitates parallel development strategies: (1) engineering immunologically ‘stealthy’ vectors through capsid modifications that evade neutralizing antibodies while retaining transduction efficiency; (2) transient immunosuppression protocols during vector administration, though this adds treatment complexity and risk; or (3) ex vivo approaches where cells are modified outside the body and screened for successful modification before transplantation, minimizing in vivo immune exposure to viral vectors or CRISPR components.

The high prevalence of pre-existing immunity represents a fundamental constraint on the percentage of the diabetes population that can benefit from gene therapies, independent of technical efficacy, and must be factored into public health impact projections for these therapeutic modalities.

Vector system comparison and selection

The AAV vectors demonstrate improved safety and tissue-specific targeting; however, they are constrained by their limited packaging capacities. In contrast, lentiviral vectors provide a greater payload capacity and stable integration, albeit with associated integration-related risks [71]. Nonviral systems offer improved safety profiles; however, they typically result in lower transfection efficiencies [35].

The selection of a vector system is contingent on the defined therapeutic objectives, types of target cells, and duration of treatment. Nonviral systems may be more suitable for applications requiring transient expression or repeated administration, whereas integrating vectors are advantageous for permanent genetic modification [72]. Fig. 2 quantitatively compares vector systems across six critical parameters, demonstrating that AAV vectors achieve the most balanced performance profile (average score 7.3/10) for diabetes applications, explaining their predominance in current clinical development despite transduction efficiency limitations. However, no single vector system optimally satisfies all requirements, necessitating context-dependent selection based on therapeutic objectives, target tissue characteristics, and manufacturing scalability considerations.

Fig. 2.

Comparative performance of gene delivery vector systems for diabetes gene therapy. This radar chart evaluates four vector platforms (AAV, lentiviral, adenoviral, non-viral) across six critical parameters for diabetes gene therapy applications. Each vector is scored 1-10 on each criterion, with higher scores indicating superior performance. The hexagonal area represents the overall suitability profile for clinical diabetes applications. Scoring Criteria and Evidence Sources: Safety Profile: Risk of serious adverse events and immunogenicity burden. AAV scores 8/10 based on 58.5% pre-existing anti-AAV2 antibodies (manageable with serotype selection) [62]. Lentiviral scores 6/10 due to residual insertional mutagenesis risk despite self-inactivating designs [73]. Adenoviral scores 4/10 reflecting fatal inflammatory responses in early gene therapy trials [72]. Non-viral scores 9/10 with minimal immunogenicity [33]. Transduction Efficiency (Human Tissue): Percentage of target cells successfully transduced. AAV scores 6/10 achieving approximately 10% human pancreatic islet cell transduction [63]. Lentiviral scores 7/10 based on 60-87% efficiency in rodent models with limited human data [27]. Adenoviral scores 9/10 with high efficiency but transient expression [30]. Non-viral scores 4/10 with 5-20% efficiency [35]. Tissue Specificity: Precision in targeting intended cell types. AAV scores 9/10 with strong serotype-dependent tropism [20]. Lentiviral scores 5/10 with moderate specificity through pseudotyping. Adenoviral scores 7/10. Non-viral scores 5/10 with limited inherent specificity requiring targeting ligands. Expression Duration (Human Data): Durability of transgene expression. AAV scores 9/10 with multi-year expression documented in hemophilia gene therapy [24]. Lentiviral scores 10/10 with permanent genomic integration [25]. Adenoviral scores 2/10 with transient 1-2 week expression [32]. Non-viral scores 3/10 requiring repeated administration. Payload Capacity: Maximum genetic cargo size. AAV scores 4/10 limited to 4.7 kb. Lentiviral scores 8/10 accommodating 8-10 kb. Adenoviral scores 7/10. Non-viral scores 10/10 with unlimited capacity. Clinical Feasibility: Regulatory approval status and manufacturing maturity. AAV scores 8/10 with FDA-approved therapies (Luxturna, Zolgensma) demonstrating regulatory pathway [89]. Lentiviral scores 7/10 with recent approvals (Zynteglo, Lyfgenia). Adenoviral scores 6/10. Non-viral scores 6/10 (mRNA-LNP vaccines approved but not for gene therapy). Interpretation: AAV vectors demonstrate the most balanced profile (average 7.3/10) with strong safety, durability, and clinical feasibility offsetting limited payload capacity. This explains AAV's dominance in current diabetes gene therapy trials despite transduction efficiency challenges. Lentiviral vectors offer superior payload and permanent expression but carry integration risks. Non-viral systems provide excellent safety and scalable manufacturing but require substantial optimization in transduction efficiency before clinical competitiveness. The optimal vector choice depends on specific therapeutic requirements: transient expression needs favor adenoviral; permanent genomic modification requires lentiviral; large genetic payloads necessitate non-viral; and balanced clinical translation favors AAV platforms. Abbreviations: AAV: adeno-associated virus, FDA: Food and Drug Administration, kb: kilobase, LNP: lipid nanoparticle, mRNA: messenger ribonucleic acid.

Safety considerations and risk assessment

Historical context and modern safety improvements

Gene therapy has evolved significantly and is influenced by early safety challenges. Notably, the death of Jesse Gelsinger in 1999 due to an immune response to an adenoviral vector [73] and the cases of T-cell leukemia in 2003 linked to retroviral insertional mutagenesis are critical indications of the complexities involved in this field [74]. These incidents prompted significant enhancements in the design of vectors and the selection of patients for treatment.

Later trials on diabetes gene therapy showed improved safety profiles, with AAV vectors consistently demonstrating strong clinical safety records across a variety of studies [64]. However, research involving animals indicates potential integration risks, highlighting the need for long-term monitoring [75].

CRISPR-associated safety risks

Applications of CRISPR-Cas9 encounter specific safety challenges that require meticulous risk management. The risk of unintended genomic modifications is a notable concern, prompting continuous advancements in guide RNA design and validation techniques to reduce off-target effects [76]. Advanced CRISPR variants using modified proofreading mechanisms demonstrate improved specificity while preserving on-target activity [77], alongside the continuous development of high-fidelity nucleases for therapeutic purposes.

Critical analysis: Why in vivo gene delivery has failed to translate

The striking disparity in clinical development ex vivo cell therapies achieving Phase III outcomes while direct AAV-insulin delivery remains in Phase I enrollment despite >15 years since foundational preclinical studies [23,27] warrants critical examination of whether current translational barriers represent surmountable engineering challenges or fundamental biological constraints necessitating paradigm shifts in therapeutic strategy.

The transduction efficiency gap is not addressable through dose escalation

The 6-8.7-fold lower transduction efficiency in human pancreatic tissue (~10% peripheral islet cells) [61] compared to rodent hepatocytes (60-87%) [23,27] appears superficially addressable through proportionally higher vector doses. However, this approach encounters multiplicative constraints rendering it clinically nonviable. First, AAV production costs are scaled linearly with dose ($50,000-$100,000 per therapeutic dose) [36]; 6-fold higher doses to match rodent transduction rates would increase per-patient manufacturing costs to $300,000-$600,000 for vector production alone, before accounting for clinical administration and monitoring. Second, immunogenicity scales nonlinearly with dose, as larger quantities of vector capsids trigger stronger innate immune responses through TLR9-mediated recognition and adaptive immune responses generating neutralizing antibodies, potentially leading to accelerated vector clearance and elimination of transduced cells [64]. Third, biodistribution studies demonstrate that systemic AAV administration at high doses results in substantial off-target transduction of non-pancreatic tissues including heart, skeletal muscle, and central nervous system [20], raising safety concerns regarding ectopic transgene expression and cumulative genotoxicity risk.

The combination of prohibitive manufacturing economics, dose-dependent immunogenicity, and off-target biodistribution indicates that dose escalation strategies cannot overcome the efficiency gap without introducing unacceptable safety and cost barriers. This conclusion is supported by the absence of published human trials attempting dose escalation beyond safety-established ranges in diabetes applications as of mid-2025.

Species-specific glucose sensing creates unpredictable translation

Preclinical glucose-responsive systems validated in rodent models rely fundamentally on GLUT2-mediated glucose sensing, whereas human β-cells predominantly utilize GLUT1 and GLUT3 transporters with distinct kinetic properties [51]. This species difference creates irreducible uncertainty regarding whether glucose-responsive regulatory elements optimized in rodent hepatocytes or β-cells will demonstrate physiologically appropriate insulin secretion in human tissue. The kinetic parameters governing glucose-stimulated insulin secretion threshold glucose concentration for insulin release (typically 5-6 mM), maximal secretory capacity, and secretion kinetics cannot be reliably predicted from rodent validation studies.

Critically, this uncertainty cannot be resolved through additional preclinical optimization in rodent models, as the regulatory elements controlling glucose-responsiveness must be re-engineered and validated directly in human tissue. The ex vivo cell therapy approach circumvents this barrier by enabling functional validation of glucose-responsive insulin secretion in differentiated human stem cell-derived β-cells prior to transplantation, measured via glucose-stimulated insulin secretion assays and C-peptide quantification under controlled glucose concentrations [62]. In contrast, direct in vivo gene delivery requires empirical clinical trials to determine whether engineered regulatory systems produce physiologically appropriate responses in human subjects, introducing both scientific uncertainty and ethical considerations regarding exposing patients to potentially suboptimal therapeutic outcomes.

Manufacturing economics creates permanent access barriers

Even if transduction efficiency and glucose-responsiveness challenges are resolved, AAV manufacturing capacity constraints represent a distinct and potentially insurmountable barrier to population-level impact. As detailed in Section 7.1, current global capacity (10,000-50,000 doses annually) would require 40-400 years to treat the target population of 2-4 million patients with severe type 1 diabetes. Manufacturing capacity expansion is capital-intensive ($5-10 billion for 10% population coverage) and time-consuming (5-10 years), with no credible pathway to achieve widespread access within a 20-year horizon using current AAV production technology [36].

This manufacturing constraint fundamentally alters the risk-benefit calculus for continued research investment in AAV-based approaches. If a therapeutic modality cannot achieve meaningful public health impact regardless of clinical efficacy due to insurmountable manufacturing limitations, continued concentration of research resources on optimization of this modality represents suboptimal allocation compared to alternative approaches with scalable manufacturing pathways.

Implications for research priority reallocation

The convergence of translational barriers efficiency gaps not addressable through dose escalation, species-specific glucose sensing requiring human tissue validation, and manufacturing economics precluding population-level access suggests that direct in vivo AAV-insulin gene delivery will serve only niche populations even if clinical efficacy is demonstrated. This conclusion necessitates strategic reallocation of research investments:

• (1)Prioritize non-viral delivery system development: Non-viral systems offer 10-100-fold lower manufacturing costs and scalable production timelines, representing the only viable pathway to equitable global access [35,36]. Current limitations in transduction efficiency (5-20% vs. 40-70% for viral vectors) warrant concentrated research investment in nanoparticle engineering, targeting ligand optimization, and endosomal escape mechanisms to achieve clinical competitiveness. Success in this domain would simultaneously address both efficacy and access barriers.
• (2)Accelerate ex vivo cell therapy approaches: The clinical development disparity favoring ex vivo approaches reflects fundamental advantages: validation of transduction efficiency and glucose-responsiveness pre-transplantation, circumvention of in vivo immune responses to vector capsids, and single manufacturing site serving multiple patients through allogeneic donor cell lines [62,63]. Research priorities include optimization of immune evasion strategies eliminating immunosuppression requirements, development of retrievable encapsulation devices enabling intervention if complications arise, and establishment of universal hypoimmune donor cell lines reducing manufacturing costs through economies of scale.
• (3)Develop predictive human tissue models: The species-specific barriers in glucose sensing and transduction efficiency necessitate validation platforms utilizing primary human pancreatic tissue, stem cell-derived organoids, or humanized mouse models for early-stage assessment of therapeutic strategies before committing to clinical trials. Investments in these platforms would reduce costly late-stage trial failures due to unpredicted species differences.

This critical analysis does not suggest abandonment of in vivo gene delivery research entirely, as subpopulations with contraindications to cell transplantation (obesity, cardiovascular disease) may benefit from hepatic insulin gene therapy if efficacy is demonstrated. Rather, it advocates for proportional research investment aligned with realistic impact projections: niche therapeutic applications warrant correspondingly limited resource allocation, whereas approaches with potential for widespread population-level benefit merit preferential investment to maximize public health impact per research dollar expended. Understanding these systematic barriers enables strategic research prioritization and realistic assessment of clinical implementation timelines, as examined in the following sections.

Part V: Implementation barriers and future directions

Economic barriers and implementation requirements

As of mid-2025, no gene therapy has received regulatory approval for diabetes. Clinical translation remains exploratory, with 83.2% of 143 registered trials in Phase I-II. Nevertheless, anticipating hypothetical implementation requires prospective analysis of economic and infrastructure barriers that will determine accessibility if regulatory approval is achieved in the projected 2028-2035 timeframe.

Manufacturing capacity: Fundamental barrier to population-level access

Current global AAV manufacturing capacity (10,000-50,000 therapeutic doses annually) creates mathematical impossibility for widespread diabetes treatment. The target population requiring intervention is 2-4 million patients worldwide with severe type 1 diabetes (hypoglycemia unawareness, >2 severe hypoglycemic events/year, or failed transplantation) would require 40-400 years to treat at maximum current capacity. Even focusing exclusively on highest-need patients (1-2 million worldwide), current capacity requires 20-200 years. Manufacturing capacity expansion to serve 10% of the target population (200,000-400,000 patients) necessitates estimated $5-10 billion capital investment for GMP AAV production facilities, requiring 5-10 years for construction, regulatory approval, and production to increase sharply [36].

Non-viral delivery systems offer 10-100-fold lower manufacturing costs ($500-$5,000 per dose vs. $50,000-$100,000 for AAV) with substantially faster production timelines, representing the only viable pathway to scalable access if clinical efficacy comparable to viral vectors can be demonstrated [35,36]. However, current non-viral systems achieve only 5-20% transduction efficiency versus 40-70% for optimized viral vectors in preclinical models, necessitating substantial optimization.

Cost-effectiveness and insurance coverage considerations

Gene therapy pricing based on approved therapies for other genetic diseases (Luxturna: $850,000; Zolgensma: $2,100,000) suggests $500,000-$2,000,000 per patient costs [36,78]. Break-even analysis compared to lifetime traditional management costs ($300,000-$400,000) requires sustained efficacy exceeding 10 years for most scenarios, with economic viability contingent on preventing costly complications including end-stage renal disease requiring dialysis ($90,000 annually) [36,79].

Hypothetical insurance coverage would likely restrict eligibility to: age <50 years, disease duration <20 years, documented treatment failure (HbA1c >8.0% or ≥2 severe hypoglycemic events annually), absence of advanced complications, and BMI 18-30 kg/m². These criteria would restrict coverage to approximately 15-25% of the type 1 diabetes population (300,000-500,000 individuals in the United States), exceeding manufacturing capacity by 6-50-fold and creating a second rationing bottleneck independent of payer decisions [36].

Geographic disparities and global health equity

Low- and middle-income countries bear 75% of global diabetes burden but possess <10% of gene therapy manufacturing infrastructure. Systematic barriers include absence of ultra-cold storage (-80°C), limited specialized interventional capabilities for portal vein catheterization, insufficient immunological screening laboratories for HLA typing and anti-AAV antibody testing, and inadequate long-term monitoring infrastructure. A stratified implementation model acknowledges high-income countries utilizing complete therapeutic spectrum with projected capacity of 500-2,000 patients annually per country, middle-income countries concentrating on simplified non-viral approaches serving 100-500 patients per region, and low-income countries unlikely to access gene therapy until 2040-2050 contingent on substantial cost reductions (<$10,000 per patient) and local manufacturing capacity development.

Strategic research priorities for clinical translation

The progression of diabetes gene therapy from preclinical proof-of-concept to clinical application requires strategic prioritization of research goals that align with translational readiness and scientific feasibility. Based on the comparative analysis and translation gap assessment presented in Section 6, we propose a tiered framework that arranges research priorities according to anticipated clinical deployment timelines while considering technical complexity and regulatory pathways.

Tier 1 Priorities: Immediate translational barriers (3-5 year horizon)

These priorities address bottlenecks in existing clinical pipelines and focus on technologies with established proof-of-concept requiring optimization for human application: (1) AAV capsid engineering for immune evasion and enhanced tropism: The 58-78% seroprevalence of anti-AAV neutralizing antibodies (Section 5.1.1) excludes the majority of potential candidates from current AAV-based approaches. Rational engineering of capsid variants through directed evolution, ancestral sequence reconstruction, or structure-guided design must achieve: (a) evasion of pre-existing neutralizing antibodies while maintaining transduction efficiency; (b) enhanced human β-cell or hepatocyte tropism compared to serotypes optimized in rodent models; and (c) reduced immunogenicity upon repeat administration. This objective directly addresses the translational barrier preventing advancement of promising preclinical AAV-insulin approaches (Section 6) to efficacy-stage clinical trials.

(2) Standardization of immune evasion protocols for allogeneic cell therapies: The discontinuation of Vertex's VX-264 encapsulated islet program and the lifelong immunosuppression requirement for zimislecel (Section 5.2.2) highlight the need for validated immune evasion strategies. Research priorities include: (a) optimization of B2M/CIITA knockout combinations in stem cell-derived β-cells to minimize MHC class I/II expression while preserving function; (b) development of localized immunomodulation strategies (e.g., PD-L1 overexpression, FasL expression) that create immune privilege without systemic immunosuppression; and (c) engineering of retrievable encapsulation devices with improved biocompatibility and vascularization compared to first-generation devices.

(3) Predictive biomarker validation for patient stratification: Current clinical trials enroll heterogeneous patient populations, contributing to variable outcomes and difficulty interpreting efficacy signals. Validation of biomarker panels including residual C-peptide levels (>0.1 pmol/mL indicating remaining β-cell function), autoantibody profiles (GAD65, IA-2, ZnT8), metabolic responsiveness indices, and immune profiling would enable: (a) precision enrollment criteria identifying patients most likely to benefit; (b) stratification for personalized treatment selection (e.g., patients with residual β-cells may benefit from protective strategies, while those with complete β-cell loss require replacement approaches); and (c) reduced trial failure rates through enrichment for responder populations.

Tier 2 Priorities: Emerging technologies requiring optimization (5-10 year horizon)

These priorities leverage technologies with demonstrated feasibility in research settings but requiring substantial development before clinical translation: (1) Base and prime editing for monogenic diabetes: Base editors (cytosine and adenine base editors) and prime editors enable precise single-nucleotide changes without double-strand DNA breaks, potentially addressing safety concerns of traditional CRISPR-Cas9 (Section 6.5.2). Applications for monogenic diabetes include correction of: (a) PDX1 mutations causing pancreatic agenesis; (b) INS gene mutations causing neonatal diabetes; (c) glucokinase (GCK) mutations in MODY2; and (d) HNF1A/HNF4A mutations in MODY1/3. Development priorities include improving editing efficiency in quiescent β-cells, minimizing off-target effects, and establishing delivery methods that achieve therapeutic editing rates (>30% of target cells) in human pancreatic tissue.

(2) Synthetic glucose-responsive promoter systems: The reliance on species-specific glucose-sensing elements (GLUT2/glucokinase in rodents versus GLUT1/GLUT3 in humans) creates uncertainty in translating glucose-responsive insulin expression from preclinical models (Section 6.6.2). Engineering synthetic promoters that respond to physiological glucose concentrations (4-10 mM) through glucose-sensing transcription factors or metabolite-responsive riboswitches, validated directly in human hepatocytes and β-cells, would eliminate dependence on endogenous regulatory elements and enable predictable glucose-responsiveness across target tissues.

(3) Combination gene therapy and immunomodulation: Mechanistic understanding of immune tolerance induction suggests that combining β-cell replacement or regeneration with targeted immunomodulation may yield synergistic efficacy. Research priorities include: (a) co-delivery of β-cell genes with immunomodulatory factors (e.g., IL-10, TGF-β, PD-L1) to create localized immune privilege; (b) sequential therapy regimens where transient immunosuppression during engraftment is followed by tolerance induction protocols; and (c) identification of combination therapy responders through immune profiling to enable personalized treatment algorithms.

Tier 3 Priorities: Transformative but high-risk objectives (>10 year horizon)

These priorities represent scientifically ambitious goals requiring substantial foundational research before translational pathways become evident: (1) Universal hypoimmune donor cells: Multiplex gene editing to delete B2M (MHC class I), CIITA (MHC class II), and overexpress CD47 ('don't eat me' signal) would create universal donor cells eliminating the need for patient-specific manufacturing or HLA matching. This approach could transform manufacturing economics by enabling off-the-shelf cellular therapies at 10-100× lower cost than autologous or HLA-matched allogeneic approaches. However, challenges include: (a) ensuring edited cells retain full β-cell function despite multiple genetic modifications; (b) preventing NK cell-mediated rejection in the absence of MHC class I; (c) addressing potential oncogenic risk from immune evasion; and (d) achieving regulatory acceptance for cells with extensive genetic modifications.

(2) Closed-loop synthetic gene circuits: Integration of glucose sensors, signal processing circuits, and insulin expression modules into autonomous synthetic genetic systems represents the theoretical ideal for β-cell replacement. Synthetic biology approaches using layered feedback loops, glucose-sensitive transcription factors, and post-translational regulatory modules could achieve proportional insulin secretion matching physiological β-cell responses. Realization requires: (a) comprehensive characterization of glucose-sensing mechanisms in human β-cells; (b) design and validation of synthetic circuits with appropriate gain, sensitivity, and dynamic range; (c) demonstration of long-term circuit stability without genetic drift; and (d) development of delivery methods that maintain circuit function in differentiated target cells.

(3) Epigenetic Editing for Autoimmune Protection: Epigenetic modifiers (dCas9 fused to DNA methyltransferases, histone acetyltransferases, or demethylases) enable reversible transcriptional regulation without permanent DNA sequence changes. Applications for diabetes include: (a) silencing of autoantigens (GAD65, IA-2, ZnT8) in β-cells to reduce autoimmune targeting while preserving function; (b) activation of immunoregulatory genes in T cells to promote tolerance; (c) metabolic reprogramming of insulin-resistant tissues through epigenetic modulation of metabolic pathways. Unlike permanent genomic modifications, epigenetic editing offers reversibility if adverse effects occur, but challenges include achieving durable epigenetic marks that persist through cell division and demonstrating therapeutic efficacy with transient modifications.

Framework rationale and limitations

This tiered prioritization synthesizes: (1) technical feasibility assessments from translation gap analysis (Sections 6); (2) regulatory pathway timelines based on analogous gene therapy approvals (Luxturna: 7-year development timeline; Zolgensma: 5 years); (3) manufacturing scalability constraints (Section 7.1); and (4) clinical need assessment prioritizing patients with inadequate responses to current therapies.

The framework uniquely prioritizes research investments based on translational readiness rather than disease burden alone, recognizing that some high-impact approaches (Tier 3: universal donor cells) require >10 years of foundational science before clinical pathways emerge, while others (Tier 1: capsid engineering) represent immediately actionable objectives with clear paths to IND-enabling studies.

Important limitations must be acknowledged: (1) this framework was developed by the authors based on literature synthesis and does not incorporate formal stakeholder input from patient advocacy organizations, regulatory agencies, payers, or manufacturers; (2) no cost-effectiveness modeling or health economic analysis was performed to weigh research investments against projected quality-adjusted life years (QALYs) gained; (3) the framework does not compare to existing diabetes research priority frameworks established by the Juvenile Diabetes Research Foundation (JDRF) or National Institutes of Health (NIH). Future work should conduct Delphi consensus panels with multistakeholder participation, integrate health economic modeling, and prospectively validate predicted timelines as trials advance through 2030-2035.

Future directions and emerging technologies

Next-generation gene editing tools

Advanced gene-editing technologies, including base and prime editing, provide improved precision in rectifying specific genetic mutations linked to monogenic diabetes, thereby facilitating the treatment of neonatal diabetes and other rare genetic variants [80,81,82,83]. The development of next-generation delivery vectors is progressing with a focus on enhanced tissue specificity, reduced immunogenicity, and improved safety profiles. Engineered AAV variants and novel non-viral delivery systems have significant potential [84].

Artificial intelligence and personalized medicine

Applications of artificial intelligence can guide RNA design, forecast treatment responses, and tailor therapeutic protocols [85]. The integration of genomic profiling, immune monitoring, and metabolic assessment facilitates patient stratification according to genetic susceptibility, immune status, and residual β-cell function [86].

Combination therapeutic approaches

Future success will likely necessitate gene therapy with traditional treatments, immunomodulatory agents, and regenerative medicine [87]. Examples include the combination of gene therapy with GLP-1 analogs to improve β-cell function and survival, where the timing and sequencing of interventions are critical for optimizing efficacy and reducing adverse effects [88].

Conclusion

Gene therapy for diabetes has transitioned from an experimental concept to a clinical reality, with various approaches showing therapeutic potential in preclinical studies and initial clinical trials. Animal models demonstrate sustained insulin production, improved glucose homeostasis, and enhanced wound healing. In contrast, clinical investigations indicate significant advancements in glycemic control and complication management. Initial human trials of gene-edited cell therapies have shown both safety and preliminary efficacy, representing a significant milestone in clinical translation.

The integration of CRISPR-Cas9 gene editing, AAV vector delivery systems, and advancements in diabetes immunology has generated significant opportunities for the development of curative therapies. Technological advances position gene therapy as a transformative approach to diabetes management, providing targeted interventions that tackle fundamental pathophysiological mechanisms instead of merely managing symptoms.

Nonetheless, considerable challenges persist in optimizing delivery systems, ensuring long-term safety, and attaining equitable access. The intricate nature of diabetes pathophysiology necessitates ongoing research focused on enhancing delivery efficiency, understanding immune responses, and ensuring applicability to varied patient populations. Economic factors, such as production expenses and healthcare integration, will be essential for achieving broad accessibility.

The future is expected to encompass combinatorial approaches addressing various pathophysiological mechanisms, personalized treatment strategies tailored to individual genetic profiles, and the incorporation of conventional therapies. Successful translation of these promising approaches into widely available treatments necessitates ongoing collaboration among researchers, clinicians, industry, and regulatory agencies. The primary objective is to develop safe, effective, and accessible treatments that can significantly change the disease trajectory and enhance the quality of life for millions affected by diabetes mellitus.

Statements & declarations

Acknowledgment

The author acknowledges Editage (www.editage.com) for English language editing.

Disclosure statement for figures and tables

All figures (Fig. 1, Fig. 1-3) and tables (Table 1A and Table 1B) are original and generated by author for this review article. Fig. 1 and Fig. 2 was created using Microsoft PowerPoint and Adobe Illustrator. Figure 3 was created using Canva.

AI tools disclosure statement

No artificial intelligence (AI) tools were utilized in the writing of this article, creation of images, or collection and analysis of data.

Funding declaration

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.