Leptin Resistance and Endoplasmic Reticulum Stress: ER Proteostasis as a Proposed Convergence Node—A Mechanistic Synthesis of Preclinical Evidence

Abstract

Most existing reviews treat endoplasmic reticulum (ER) stress as one mechanism among several in leptin resistance. This review advances a different claim: ER proteostasis may act as a central convergence node, translating metabolic, inflammatory, developmental, and nutrient-derived signals into a shared effector signature. In common obesity, leptin responsiveness is blunted despite hyperleptinaemia—the principal obstacle to leptin-based therapy. Leptin resistance is examined at the intersection of intracellular signalling networks (JAK2/STAT3, PI3K/Akt, MAPK, AMPK, mTOR, FoxO1, SOCS3, and PTP1B), hypothalamic ER stress, and the unfolded protein response (UPR). This narrative synthesis incorporates 22 primary studies, with PRISMA 2020 principles used to document study selection transparently. The ER shapes leptin signalling at three levels: receptor biogenesis (SEL1L–HRD1 ERAD), receptor bioavailability (CHOP-mediated soluble receptor), and post-receptor inhibition (SOCS3/PTP1B). Distinct stressors converge on a narrow effector set (CHOP, SOCS3, and PTP1B) whose direction of effect is context-dependent, and receptor quality control can operate without activation of the PERK arm, indicating that the ER’s role is not confined to the canonical stress response. The framework yields three testable hypotheses with an explicit falsification criterion and proposes that leptin resistance comprises phenotypically distinct subtypes. The evidence base is entirely preclinical (17 rodent in vivo; 5 in vitro studies); translation to humans remains untested.

1. Introduction

The signal linking adipose mass to feeding behaviour was predicted by Coleman’s parabiosis experiments [1] and identified molecularly with the cloning of the obese (ob) gene [2]. That recombinant leptin reduces food intake [3,4], that the receptor was subsequently cloned [5], and that loss-of-function mutations in the leptin gene cause early-onset obesity reversible by replacement therapy [6,7] together established the hormone’s role in energy balance. Leptin acts by binding LepRb on POMC and AgRP neurons of the arcuate nucleus, thereby suppressing food intake [8,9].

What is seen in common obesity is hyperleptinaemia rather than leptin deficiency: affected individuals fail to mount the expected anorexigenic response to leptin. This state, termed leptin resistance, distinguishes rare type 1 (leptin-deficient) obesity from the far more common type 2 (leptin-resistant) form [8]. Roughly one in eight people worldwide now lives with obesity, and the prevalence is projected to keep rising through 2050 [10,11]. Leptin resistance is both the principal barrier to the therapeutic use of leptin [12,13,14] and a contributor to cardiometabolic complications [9].

This review conceptualises ER proteostasis as a candidate regulatory hub at which metabolic, inflammatory, developmental, and nutritional inputs intersect and are routed towards a leptin-resistant phenotype. The framework rests on three testable hypotheses—input independence, node primacy, and the proteostasis–stress distinction—and further argues that leptin resistance may comprise several phenotypes differing in their dominant mechanism. The mechanisms are presented not as an established model but as an explicitly falsifiable framework.

2. Materials and Methods

2.1. Literature Search Strategy

This is a narrative review. The reporting principles of the PRISMA 2020 statement [15] were applied not to conduct a systematic review but to document study selection transparently. Searches of PubMed/MEDLINE and Scopus were completed on 19 July 2026. Records were screened independently by two authors, first on title and abstract and then at full text, and disagreements were resolved by discussion. Citation tracking of the reference lists of the included studies was not carried out, so the evidence base is limited to what the two database searches retrieved; this is noted among the limitations in Section 11.2. The design should be named precisely because the reporting conventions used here are those of a systematic review while the design is not one. No protocol was registered in advance, the search was confined to two databases, no meta-analysis was undertaken, and the synthesis is interpretive rather than quantitative. What the PRISMA conventions provide is an auditable account of how the twenty-two studies were arrived at, not a claim to the status of a systematic review.

The following search string was applied identically to both databases:

(“leptin resistance” OR “leptin signaling”) AND (JAK2 OR STAT3 OR PI3K OR Akt OR MAPK OR ERK OR AMPK OR mTOR OR FoxO1 OR SOCS3 OR PTP1B OR “ER stress” OR “endoplasmic reticulum stress” OR “unfolded protein response” OR UPR OR PERK OR IRE1 OR ATF6 OR CHOP OR XBP1).

Only studies published on or after 1 January 2015 were eligible for the core evidence table. The boundary was chosen deliberately rather than for convenience. That ER stress can cause hypothalamic leptin resistance was established before this date, principally by Hosoi et al. [16] and Ozcan et al. [17], and the present review does not set out to re-establish it; the question asked here is whether the evidence accumulated since then supports the stronger proposition that ER proteostasis functions as a convergence node. Confining the tabulated evidence to the more recent literature keeps the synthesis on that question and avoids counting the foundational studies twice, once as the premise of the framework and again as evidence for it. The restriction applies to the evidence table alone. Pre-2015 studies bearing directly on the question are appraised separately in Section 2.2 and in Table S1, and they are cited throughout the text wherever they bear on the argument; this includes Williams et al. [18], whose demonstration of Xbp1s function in POMC neurons is described in Section 6.1 as a foundation of the framework and is therefore discussed rather than tabulated. Two consequences follow and are stated here rather than left implicit: the evidence table is not a complete census of experimental work on this question, and no claim about the relative frequency or consistency of findings is made on the basis of its contents. Studies were included if they reported original research, addressed leptin resistance or leptin signal transduction as a primary topic, linked at least one ER stress or UPR component (PERK, IRE1α, ATF6, CHOP, XBP1, GRP78/BiP, eIF2α, and ERAD) mechanistically to leptin signalling, and were available in English as full text. Studies treating leptin only superficially were excluded, as were records without abstracts, conference proceedings, case reports, and retracted publications.

Of 421 records retrieved (PubMed/MEDLINE, n = 244; Scopus, n = 177), 83 duplicates were removed. Of the 338 unique records, 58 were excluded at title and abstract screening and a further 258 at full-text assessment (target mechanism not the primary focus, n = 189; signalling studies with no ER stress component, n = 54; reviews, n = 15). Twenty-two studies entered the synthesis and the evidence table, and the seventeen animal studies were assessed with SYRCLE’s risk of bias tool [19], as described in Section 2.3 (Figure 1; Table 1; Tables S2 and S3).

Figure 1. Study selection flow chart (PRISMA 2020 format). Flow of study selection for the core evidence base; the diagram follows the PRISMA 2020 format.

Table 1. Summary of the 22 primary studies addressing the relationship between ER stress and intracellular signal transduction in leptin resistance (in descending order of publication year). The set contains no human studies and no randomised controlled trials (animal in vivo n = 17; laboratory studies only n = 5). The study design category column is colour-coded by model system: yellow = animal study, red = laboratory (in vitro) study; because no human study is included, the green category is unrepresented. The effectors column states which of the three proposed effectors each study actually measured and whether it changed: ↑ significantly increased, ↓ significantly decreased, = measured without a significant change, ✓ measured, with the direction reported only within a particular comparison, – not measured. CHOP was measured in twelve of the twenty-two studies, SOCS3 in eight, and PTP1B in five; two studies measured all three. The model column names the tissue or cell type in which the reported outcomes were measured, so that findings in hypothalamic neurons can be distinguished from those in non-neuronal cells of the central nervous system and in peripheral tissues. That column describes how each study was designed and is not an assessment of its quality; risk of bias was assessed separately with SYRCLE’s tool (Section 2.3) and is reported in Tables S2 and S3. Each key finding states the direction of the reported effect, together with any feature of the study’s own reporting that limits what may be concluded from it.

Study (Author, Reference)	Model/System	Study Design Category	ER Stress/UPR and Associated Pathways	Effectors Measured (CHOP/SOCS3/PTP1B)	Key Finding
Kim [20]	Mouse (ApoE−/−, in vivo); liver and aortic tissue	Animal study—dietary intervention	ER stress/UPR	CHOP ↓; SOCS3–; PTP1B–	Indole-3-carbinol (0.05% of a Western diet, 12 weeks) reduced fat-pad mass by 16–27% and plasma leptin by 47% without a change in body-weight gain, and lowered GRP78, p-eIF2α, XBP1, and CHOP relative to Western diet alone.
Xu [21]	Mouse (C57BL/6J, in vivo); hypothalamus	Animal study—dietary intervention	JAK2/STAT3, ER stress/UPR	CHOP–; SOCS3 ↓; PTP1B ↓ (expression and activity)	PEITC dose-dependently reduced body-weight gain and visceral fat; restored leptin sensitivity by increasing Ob-Rb expression, and JAK2 and STAT3 phosphorylation; suppressed hypothalamic neuroinflammation; and attenuated ER stress across all three UPR branches, reactivating POMC and inhibiting AgRP neurons.
El-Sehrawy [22]	Mouse (ovariectomised, in vivo); hypothalamus	Animal study—pharmacological	JAK2/STAT3, ER stress/UPR	CHOP–; SOCS3–; PTP1B–	Ovariectomy increased body weight, energy intake, hypothalamic NPY, and ER stress markers and reduced αMSH and leptin signalling in high-fat-fed mice; the ERα agonist PPT reversed these changes, lowering GRP78 and ATF6 and raising p-JAK2 and p-STAT3.
Hua [23]	Rat (DIO, in vivo); hypothalamus, with BV2 microglia and GT1-7 neurons in vitro	Animal study + laboratory (in vitro)	JAK2/STAT3, SOCS3, PTP1B, ER stress/UPR	CHOP–; SOCS3 ↓; PTP1B ↓	Fufang Zhenzhu Tiaozhi capsule and berberine reduced body weight and fat percentage; downregulated hypothalamic ATF6, eIF2α, GRP78, IRE1α, PERK, and XBP1; inhibited p-IRE1α/sXBP1 signalling; lowered PTP1B and SOCS3 while raising LepRb mRNA and p-STAT3; and increased autophagy markers. The hypothalamic outcomes rest on n = 3 per group.
Mao [24]	Mouse (POMC-specific SEL1L knockout, in vivo); hypothalamic POMC neurons	Animal study—conditional genetics	ER stress/UPR	CHOP–; SOCS3–; PTP1B–	Loss of SEL1L-HRD1 ERAD in POMC neurons caused newly synthesised LepRb to be retained in the ER and reduced the mature receptor reaching the cell surface, producing leptin resistance, and diet-induced obesity; receptor retention and the loss of leptin signalling occurred without detectable activation of the PERK arm, although BiP and spliced XBP1 were moderately elevated under high-fat feeding, so the entire ER stress response was not absent.
Kim & Kim [25]	Mouse (ApoE−/−, in vivo); liver and aortic tissue	Animal study—dietary intervention	ER stress/UPR	CHOP ↓; SOCS3–; PTP1B–	Diallyl disulfide lowered plasma cholesterol and leptin and reduced Western-diet-induced ER stress markers in ApoE-deficient mice.
Preninka [26]	SH-SY5Y cells expressing Ob-Rb (in vitro); neuronal cell line, not of hypothalamic origin	Laboratory only (in vitro)	JAK2/STAT3, ER stress/UPR	CHOP–; SOCS3–; PTP1B– (HERP was the only ER stress marker measured)	Homocysteine induced ER stress and inhibited leptin-induced STAT3 phosphorylation in Ob-Rb-expressing cells; methionine and cysteine had weak effects and IL-6-induced STAT3 activation was unaffected, indicating relative selectivity for leptin signalling.
Baba [27]	Mouse (ob/ob); isolated pancreatic islets (ex vivo)	Animal study—pharmacological	SOCS3, ER stress/UPR	CHOP ↑, normalised by PBA; SOCS3 ✓ (leptin-induced response restored by PBA); PTP1B–	CHOP and XBP1, elevated in ob/ob pancreatic islets, approached lean levels after phenylbutyric acid; leptin-stimulated SOCS3, which did not rise in untreated ob/ob islets, was restored to the lean pattern. Islets from two animals were pooled per sample, and protein samples were pooled again before immunoblotting.
Chen [28]	Rat (in vivo, RYGB surgery); hypothalamus, with BV2 microglia and a POMC cell line in vitro	Animal study—weight-matched control	ER stress/UPR	CHOP ↓ after RYGB only; SOCS3 ↓; PTP1B ↓	RYGB reduced hypothalamic gliosis, TLR4-MyD88 inflammatory signalling, and ER stress and restored leptin-induced suppression of feeding; the improvement was absent in a body-weight-matched calorie-restricted control and depended on the post-surgical gut microbiota.
Gong [29]	Mouse (whole-body Selenom knockout, in vivo); hypothalamus, with mHypoE-44 neurons in vitro	Animal study—causal genetic	JAK2/STAT3, ER stress/UPR	CHOP–; SOCS3–; PTP1B–	Leptin increased hypothalamic SELENOM expression, and leptin-induced STAT3 phosphorylation was attenuated in Selenom-null mice matched to controls for body weight; SELENOM deficiency also reduced thioredoxin activity and increased susceptibility to ER-stress-mediated cell death in hypothalamic neurons.
Horiuchi [30]	Neuronal cell line (in vitro); not of hypothalamic origin	Laboratory only (in vitro)	ER stress/UPR	CHOP–; SOCS3–; PTP1B– (GRP78 only)	Biochanin A attenuated ER-stress-induced cell death in neuronal cells, improved ER-stress-impaired leptin signalling, and suppressed ER-stress-induced GRP78 expression.
Drori [31]	Mouse (global CHOP-null, in vivo); liver, with hepatocytes in vitro	Animal study—causal genetic	ER stress/UPR	CHOP ✓ (genetic loss of function and promoter binding); SOCS3–; PTP1B–	CHOP binds the Lepr promoter and drives sOb-R expression; peripheral CB1R blockade restored circulating sOb-R and hepatic leptin sensitivity in wild-type but not in globally CHOP-null mice, in which lean animals already failed to raise hepatic pSTAT3 in response to leptin.
Park [32]	Mouse (maternal obesity, in vivo); hypothalamus of the offspring	Animal study—pharmacological	ER stress/UPR	CHOP ↑, reversed by TUDCA; SOCS3–; PTP1B–	Maternal obesity produced neonatal hyperleptinaemia and leptin resistance, upregulated hypothalamic ER stress genes, and disrupted the development of melanocortin projections in the offspring; neonatal TUDCA reversed both the metabolic and the neurodevelopmental deficits.
Aiceles [33]	Rat (congenital hypothyroidism, in vivo); hypothalamus	Animal study—observational/model	JAK2/STAT3, SOCS3, ER stress/UPR	CHOP–; SOCS3 ↑; PTP1B–	In adult male offspring of methimazole-treated dams, hypothalamic OB-R fell by 45–63%, pSTAT3 by 51–56%, pERK by 48–50%, and POMC by 41–46%, while SOCS3 rose by 52–170% and serum leptin by 20–25%.
Hosoi [34]	SH-SY5Y cells expressing Ob-Rb (in vitro); neuronal cell line, not of hypothalamic origin	Laboratory only (in vitro)	JAK2/STAT3, ER stress/UPR	CHOP ↑; SOCS3–; PTP1B–	4-HNE reduced leptin-induced STAT3 phosphorylation dose-dependently through the eIF2α-CHOP pathway, while global tyrosine phosphorylation was unaffected.
Gan [35]	Mouse (in vivo); white and brown adipose tissue, with preadipocytes in vitro	Animal study + laboratory (in vitro)	JAK2/STAT3, SOCS3, ER stress/UPR	CHOP ↓ with αMSH; SOCS3 ↓ with αMSH; PTP1B–	αMSH lowered GRP78, CHOP, and SOCS3 and raised pSTAT3 in adipose tissue of high-fat-fed and ob/ob mice, and promoted preadipocyte proliferation through Notch1; ER stress induced by tunicamycin or thapsigargin reduced leptin sensitivity in preadipocytes.
Ibars [36]	Rat (Wistar, in vivo); hypothalamus, liver, mesenteric adipose tissue and skeletal muscle	Animal study—dietary intervention	JAK2/STAT3, SOCS3, PTP1B, ER stress/UPR	CHOP = (unchanged by GSPE); SOCS3 ↓ in the hypothalamus and skeletal muscle on cafeteria diet, significantly ↑ with GSPE in both; PTP1B ↓ in both tissues on cafeteria diet, significantly ↑ with GSPE in skeletal muscle only (the hypothalamic change was not significant)	Grape-seed proanthocyanidin extract reduced food intake without reversing hyperleptinaemia or weight gain; it increased hypothalamic STAT3 activation and Pomc mRNA, lowered local inflammation, and raised Sirt1. The cafeteria diet reduced hypothalamic p-STAT3 and hypothalamic Socs3 and Ptp1b expression; GSPE significantly restored p-STAT3 and Socs3, whereas the increase in hypothalamic Ptp1b did not reach statistical significance (Figure 1d of that study). In skeletal muscle, both Socs3 and Ptp1b were significantly restored, while liver and mesenteric adipose tissue showed no significant change. Hypothalamic ER stress markers (spliced XBP1, ATF4, CHOP) and Obrb expression were unaffected by GSPE.
Hosoi [37]	SH-SY5Y cells expressing Ob-Rb (in vitro); neuronal cell line, not of hypothalamic origin	Laboratory only (in vitro)	JAK2/STAT3, ER stress/UPR	CHOP ↑; SOCS3–; PTP1B–	Glucose deprivation increased GRP78 and CHOP and inhibited leptin-induced STAT3 and STAT5 phosphorylation in neuronal cells; flurbiprofen restored pSTAT3 without altering UPR activation.
Sai [38]	Mouse (CNS-specific Tak1 knockout, in vivo); hypothalamus, with cell experiments in vitro	Animal study—conditional genetics	ER stress/UPR	CHOP ↓ in Tak1-deficient cells; SOCS3–; PTP1B–	CNS-specific Tak1 deletion upregulated SREBP-target lipogenic genes, expanded ER capacity, and blocked hypothalamic ER stress, preventing high-fat-diet-induced leptin resistance and hyperphagic obesity; the knockout mice were smaller than littermate controls at baseline.
Thon [39]	SH-SY5Y cells expressing Ob-Rb (in vitro); neuronal cell line, not of hypothalamic origin	Laboratory only (in vitro)	JAK2/STAT3, ER stress/UPR	CHOP ↑; SOCS3–; PTP1B–	Dehydroascorbic acid raised GRP78, CHOP, and spliced XBP1 and reduced leptin-induced STAT3 phosphorylation in neuronal cells.
Camargo [40]	Mouse (in vivo); hypothalamus	Animal study—dietary intervention	ER stress/UPR	CHOP–; SOCS3–; PTP1B–	Protein restriction combined with a high-fat diet produced hypothalamic leptin resistance and altered the expression of hypothalamic genes in ER stress, apoptosis, and cellular-defence pathways; taurine supplementation prevented both the leptin resistance and the excess calorie intake.
Hakim [41]	Mouse (CHOP heterozygous, in vivo); hypothalamus	Animal study—causal genetic	JAK2/STAT3, SOCS3, PTP1B, ER stress/UPR	CHOP– (genetic model only); SOCS3 = (unchanged); PTP1B ↑ (expression and activity)	Chronic sleep fragmentation increased food intake and body weight, activated hypothalamic UPR markers, left ObR expression unchanged, and reduced leptin-induced STAT3 phosphorylation through increased PTP1B activity; the effects were reversed by TUDCA and absent in CHOP-haploinsufficient mice.

Abbreviations: UPR, unfolded protein response; ER, endoplasmic reticulum; JAK2, Janus kinase 2; STAT3, signal transducer and activator of transcription 3; SOCS3, suppressor of cytokine signalling 3; PTP1B, protein tyrosine phosphatase 1B.

During the preparation of this manuscript, a large language model (Claude, Anthropic; model claude-opus-5) was used solely for language editing and academic style refinement of the English text. No scientific content, data, interpretations, or conclusions were generated by the tool. The corresponding author reviewed and approved all language-related edits and takes full responsibility for the content of the manuscript.

2.2. Appraisal of Foundational Pre-2015 Evidence

Four studies published before the eligibility window are appraised separately because the framework developed here depends on them. Hosoi et al. [16] showed that pharmacologically induced ER stress blocks leptin-induced STAT3 phosphorylation, that the effect is mediated by PTP1B and not by SOCS3, and that a chemical chaperone reverses it. Ozcan et al. [17] established in vivo that diet-induced obesity produces hypothalamic ER stress, that chemical chaperones restore leptin sensitivity, and that reduced ER folding capacity increases susceptibility to leptin resistance. Schneeberger et al. [42] showed that deletion of Mfn2 in POMC neurons reduces mitochondria–ER contacts; impairs POMC processing; and produces ER stress, leptin resistance, and obesity that pharmacological relief of ER stress reverses. Williams et al. [18] showed that Xbp1s in POMC neurons improves leptin sensitivity and glucose homeostasis. Table S1 sets out their models, findings, and the role each plays in the present argument, and they are discussed in Section 6.1 and Section 9.2 where they bear on it. Their absence from Table 1 is a reporting convention adopted for the reason given in Section 2.1; by design and by strength of inference, they are among the most informative studies in this field, and any appraisal of the framework should weigh them alongside the tabulated evidence.

2.3. Risk of Bias Assessment

The seventeen animal studies in the core evidence base were assessed against the ten items of SYRCLE’s risk of bias tool [19]. Each item was judged as low, high, or unclear risk from the full text of the primary report, including Supplementary Materials where available. Two conventions were applied so that the judgements are reproducible. First, item 1 was judged unclear wherever a report states that animals were allocated randomly without describing how the allocation sequence was generated; high risk was reserved for a described non-random method, which no report used. Second, where blinding or outcome assessment differed between outcomes, the judgement was made on the primary molecular outcome, that is, the measure of leptin signalling or ER stress on which this review draws. Item-level judgements for every study are reported in Table S2, and the basis for each judgement that departs from unclear is set out in Table S3. The five laboratory-only studies were not assessed because SYRCLE applies to animal experiments. The assessment is used in Section 11.2 to qualify the strength of the conclusions; it was not used to weight, rank, or exclude studies.

3. Leptin Physiology

Leptin is a 167-amino-acid polypeptide encoded by the ob gene and produced mainly by mature adipocytes of white adipose tissue; circulating concentrations correlate strongly with total fat mass [2,8]. Of the receptor isoforms, only LepRb carries a full-length intracellular domain and can therefore transmit JAK2/STAT3 signalling; it is densely expressed in POMC and AgRP neurons. The soluble isoform sOb-R determines the size of the free leptin pool [5,31,43]. Because canonical leptin physiology has been reviewed extensively elsewhere [8,9], only three features that intersect with the ER are emphasised here.

First, the functional competence of the receptor depends not only on its expression level but also on correct folding within the ER. Newly synthesised LepRb is prone to misfolding and is a substrate of the SEL1L–HRD1 ERAD complex, and loss of SEL1L in POMC neurons attenuates leptin signalling; notably, retention of the receptor within the ER occurs without activation of the PERK arm; BiP and spliced XBP1 were moderately elevated in that model, so the absence of PERK signalling should not be read as the absence of the entire ER stress response (Section 9.1) [24]. Second, receptor expression can fall under pathological conditions: in a model of congenital hypothyroidism, hypothalamic OB-R expression decreased by 45–63%, pSTAT3 by 51–56% and pERK by 48–50%, while SOCS3 rose [33]. Third, leptin resistance is not exclusively a central phenomenon: mice fed a high-fat diet develop local ER stress and leptin resistance within adipose tissue, and α-melanocyte-stimulating hormone (αMSH) reverses this picture by lowering GRP78 and CHOP [35].

4. Canonical Leptin Signalling

Binding of leptin to LepRb activates JAK2, which phosphorylates tyrosine residues on the cytoplasmic tail of the receptor and thereby creates docking sites for partially dissociable parallel arms (Figure 2).

Figure 2. Leptin signal transduction under physiological conditions. Parallel signalling arms emerging from the leptin–LepRb–JAK2 axis and their net effect on hypothalamic POMC/AgRP neurons. Arrows indicate the direction of signal transmission; ↑, increase; ↓, decrease; the dashed line represents the blood–brain barrier.

The JAK2/STAT3 pathway, one of the principal mediators of leptin’s anorexigenic effects, is frequently used as a readout of leptin resistance. Yet, stimuli as different as selenoprotein M deficiency, homocysteine, 4-HNE, and glucose deprivation all lower pSTAT3, which shows that the marker is sensitive but not mechanism-specific. In diet-induced obese mice, moreover, STAT3 DNA-binding activity increases, and pharmacological reduction in pSTAT3 facilitates leptin-dependent POMC expression [44]—evidence that reduced pSTAT3 does not correspond to leptin resistance under all conditions.

PI3K/Akt is among the principal pathways at which leptin and insulin signalling intersect within the hypothalamus: mice carrying a kinase-domain mutation in PI3K-C2α develop leptin resistance, age-dependent insulin resistance, and obesity together [45]. Perinatal nutritional interventions have also been reported to alter hypothalamic IRS1 and AMPK phosphorylation and thereby to affect leptin signalling permanently [46].

Evidence for a role of the MAPK/ERK arm in leptin resistance is limited and largely observational [33]. Hypothalamic AMPK is suppressed by leptin; in a glucose-deprivation model, flurbiprofen reversed the suppression of pSTAT3 without altering UPR activation, which suggests that UPR activation alone may not fully account for leptin resistance [37].

mTOR is linked to ER proteostasis through its role as a negative regulator of autophagy, but the supporting studies are confined to cardiac and hepatic tissue [47,48,49]. FoxO1, for its part, competes with STAT3 at the POMC promoter and so shapes the transcriptional balance [44,50], raising the possibility that leptin resistance also reflects a shift in the balance between opposing transcriptional regulators. No primary study has examined the relationship between PI3K/Akt, mTOR, or FoxO1 and ER stress directly in the context of leptin resistance; these links remain inferential (Table 2).

Table 2. Knowledge gaps and proposed research approaches. The principal gaps identified by this review and the methodological approach required for each.

Gap Area	Current Situation	Approach Required
Human intervention studies	Agents targeting ER stress have been tested in humans against insulin endpoints only	Randomised controlled trial with leptin sensitivity as the primary endpoint
PI3K/Akt—ER stress link	No primary study links the two processes directly	Simultaneous UPR and PI3K measurement in conditional genetic models
mTOR—autophagy—ER stress	Current evidence rests on extrapolation from peripheral tissues	Tissue-specific studies with hypothalamic endpoints
FoxO1—UPR interaction	No primary study addresses the question directly	Examination of the STAT3/FoxO1 promoter balance under ER stress
Sex differences	The great majority of studies were conducted in male animals	Inclusion of both sexes; sex-stratified analysis
Intestinal permeability and microbiota	RYGB data point to gut-derived signals, but the mechanism is unresolved	Gnotobiotic models; hypothalamic UPR measurement after microbiota transfer
Epigenetic mediators	Correlational human data exist for DNA methylation and miRNA; no link to ER stress has been established	Integration of the ER stress and epigenome layers within the same model
Cellular resolution	Measurements rest on tissue homogenates	Single-cell RNA sequencing; spatial transcriptomics
Absence of biomarkers	Hypothalamic ER stress cannot be measured in humans	Identification of circulating biomarkers through multi-omics
Nutrigenomics/precision nutrition	Individual response variability is unknown	Nutrition studies stratified by genotype and microbiota

5. Molecular Basis of Leptin Resistance

For leptin to exert its central effects, it must cross from the circulation into the brain, and the saturable nature of this transport may create an early bottleneck under hyperleptinaemic conditions [12]. Tanycytes surrounding the mediobasal hypothalamus have been shown to form an ERK-dependent gate for leptin, and this mechanism is disrupted in diet-induced obesity [51].

Upstream of this step lies the soluble receptor (sOb-R), which regulates the free leptin pool. Pharmacological and genetic modulation of hepatic CB1R alters sOb-R levels and hepatic leptin resistance, and CHOP binds the Lepr promoter and drives sOb-R expression [31]. The genetic evidence in that study requires careful reading: in globally CHOP-deficient mice, basal hepatic sOb-R was reduced, lean CHOP-null mice failed to raise the hepatic pSTAT3/STAT3 ratio in response to exogenous leptin, and peripheral CB1R blockade no longer restored circulating sOb-R. Deleting CHOP therefore reduced, rather than improved, hepatic leptin responsiveness. Two features limit how far this observation can be generalised: the model is a constitutive, whole-body null in which basal hepatic endocannabinoid tone is already markedly elevated, so a developmental or compensatory phenotype cannot be separated from the acute signalling role of CHOP, and the hypothalamus was not examined. Read in this way, the study identifies a basal transcriptional function of CHOP in setting hepatic sOb-R production, distinct from the stress-induced hypothalamic function discussed in the Section Context-Dependence of the Shared Effectors. ER stress can therefore interfere with leptin signalling through the circulating leptin pool, before the target cell is reached at all.

Impairment at the level of the receptor can arise in two ways. Direct evidence for reduced receptor expression comes from the congenital hypothyroidism model [33]. The more fundamental mechanism concerns receptor folding and maturation: LepRb has been identified as a substrate of the SEL1L–HRD1 ERAD complex, and loss of SEL1L in POMC neurons leads to accumulation of the receptor within the ER and to leptin resistance [24]. That this effect arises without activation of the PERK arm (Section 9.1) suggests that the ER acts as a basal regulator of receptor biogenesis and quality control, and it constitutes the strongest support for the framework proposed here.

SOCS3 is among the principal negative-feedback regulators of leptin signalling: induced by STAT3, it binds Tyr985 of LepRb and suppresses JAK2 activity [52]. Sustained activation of this loop may contribute to leptin resistance, and elevated SOCS3 has been demonstrated in the hypothalamus in congenital hypothyroidism [33] and in adipose tissue under high-fat feeding [35]. The cafeteria-diet model runs in the opposite direction: hypothalamic Socs3 expression was lower than in control rats and rose with an intervention that improved hypothalamic leptin signalling [36], and this study is therefore not evidence of elevated SOCS3 in resistance (Section Context-Dependence of the Shared Effectors). In the only study to address the relationship with ER stress directly, CHOP and XBP1, both elevated in the pancreatic islets of ob/ob mice, approached lean control levels after phenylbutyric acid (PBA); SOCS3, by contrast, was higher in leptin-stimulated islets from PBA-treated animals than in untreated ob/ob islets, and the PBA-treated group no longer differed from lean controls either with or without leptin [27]. The direction of this change deserves emphasis, since at first sight, it appears to contradict the role assigned to SOCS3 here. SOCS3 is a STAT3-induced negative-feedback gene, so its induction by leptin is a readout of an intact signalling pathway rather than of a blocked one; in untreated ob/ob islets, leptin failed to induce it, and PBA restored the lean-like inducible response. The finding is therefore consistent with relief of ER stress restoring leptin responsiveness, and it illustrates a distinction applied throughout this review: an acute, ligand-driven rise in SOCS3 reflects physiological feedback, whereas a chronically elevated basal level in the unstimulated state is the pattern associated with resistance. The evidence is nonetheless confined to a single tissue in a leptin-deficient rather than leptin-resistant model.

PTP1B is a tyrosine phosphatase that terminates leptin signalling by dephosphorylating JAK2 and is localised to the ER membrane—a position that makes it one of the most direct links between ER proteostasis and leptin signalling. In a model of chronic sleep fragmentation, increased food intake was accompanied by hypothalamic UPR activation, unchanged ObR expression, and reduced STAT3 phosphorylation; the picture was attributed to PTP1B, and both TUDCA and CHOP haploinsufficiency were protective [41]. PTP1B has also been shown to be modifiable by diet [23,36].

These four mechanisms do not rest on evidence of equal strength. The strongest causal findings come from receptor quality control and from mechanisms upstream of the UPR, where conditional genetic models link molecular changes to physiological endpoints [24,38,41]. Much of the evidence for SOCS3 and PTP1B, by contrast, is correlative: hypothalamic SOCS3 knockdown experiments are lacking [33], the ER stress–PTP1B relationship has not been tested with neuron-specific deletion, and the effect of PBA on SOCS3 is better explained by restoration of inducible feedback than by inhibition, which is itself a reason for caution in treating SOCS3 as a uniform marker of resistance (Section Context-Dependence of the Shared Effectors) [27]. No mechanism, including tanycyte-mediated transport, is supported by human data [51].

6. Endoplasmic Reticulum Stress

6.1. UPR Activation

This review distinguishes ER proteostasis, ER stress, and the UPR. ER proteostasis denotes the basal capacity of the ER to sustain protein folding, quality control, and degradation; ER stress denotes the point at which the unfolded protein load exceeds that capacity; and the UPR denotes the cellular programme that responds through the PERK–eIF2α–ATF4, IRE1α–XBP1, and ATF6 arms—initially adaptive but capable of impairing leptin signalling once it becomes chronic. The distinction matters because retention of LepRb following loss of ERAD can produce leptin resistance without activation of the PERK arm, showing that disrupted proteostasis can affect signalling without the stress response accounting for it [24]; the precise form of this claim is given in Section 9.1. Because the term ER proteostasis is used here in a specific sense, its components are set out explicitly in Table S4. Three layers are distinguished. The first is basal protein quality control: the folding capacity of the ER, the SEL1L–HRD1 ERAD complex, and the substrates it handles, among them LepRb itself; these operate continuously and are not switched on by stress. The second is stress sensing and the UPR: the PERK–eIF2α–ATF4, IRE1α–XBP1, and ATF6 arms together with their transcriptional outputs, of which CHOP is the one that recurs throughout this review. The third is the set of signalling effectors on which the first two layers are proposed to converge: CHOP, SOCS3, and PTP1B, of which only PTP1B is itself ER-resident, anchored to the cytosolic face of the ER membrane, which is what makes it the most direct physical link between the organelle and the termination of leptin signalling.

Two consequences of this classification should be stated plainly. First, GRP78/BiP, CHOP, and spliced XBP1 are read throughout this review as markers of UPR activation rather than as components of the basal quality-control machinery, so a study reporting them has measured the second layer and not the first. Second, the evidence assembled in this review is distributed very unevenly across the three layers: almost every included study measures the UPR layer, whereas the basal proteostasis layer, which is the most original element of the framework, rests on a single study [24]. The expression ER proteostasis node therefore denotes the union of these layers as a proposed functional unit and not a third entity distinct from them, and the proposition that the first layer belongs within the node is at present supported by one experiment.

Findings that all three UPR arms are activated together in the context of leptin resistance are consistent across models: sleep fragmentation [41], diet-induced obese rats [23], activation reversible by phenethyl isothiocyanate [21], and increased hypothalamic ER stress gene expression in neonates in a maternal obesity model [32]. The foundation of this framework is the study establishing the role of the IRE1α–XBP1 pathway in POMC neurons [18]. That study belongs to a small group of earlier reports on which the present argument rests: pharmacologically induced ER stress blocks leptin-induced STAT3 phosphorylation, an effect transmitted by PTP1B rather than SOCS3 and reversed by a chemical chaperone [16]; diet-induced obesity produces hypothalamic ER stress that chemical chaperones reverse, with reduced ER folding capacity increasing susceptibility [17]; and loss of Mfn2 in POMC neurons couples disrupted mitochondria–ER contacts to ER stress and leptin resistance [42]. These four studies fall outside the eligibility window of the evidence table and are appraised separately in Section 2.2 and Table S1.

Regulators upstream of the UPR indicate that the mechanism can be situated within a broader proteostasis network: deletion of TAK1 (MAP3K7) increases SREBP-mediated lipogenesis and ER capacity and thereby prevents ER-stress-driven leptin resistance and hyperphagic obesity under a high-fat diet [38].

6.1.1. Lipid Overload

Western-type diets rich in saturated fat are thought to trigger the UPR by altering ER membrane composition and protein-folding capacity. In apolipoprotein E-deficient mice, a Western diet raised markers of ER stress, whereas diallyl disulfide lowered these markers together with hypercholesterolaemia [25]; indole-3-carbinol has been reported to reduce plasma leptin and thereby to attenuate ER-stress-associated hyperleptinaemia [20].

These findings appear at first sight to conflict with the TAK1 study: in ApoE−/− models, reducing lipid load was associated with improvement in ER stress markers, whereas in TAK1 deletion, the increase in lipogenesis was interpreted as an expansion of ER capacity [38]. The discrepancy suggests that what matters is less the absolute lipid load than the balance between that load and the processing capacity of the ER—an interpretation that has not been tested directly.

6.1.2. Oxidative ER Stress

Because the ER lumen is an oxidising environment, shifts in redox balance can affect protein-folding capacity. The lipid peroxidation product 4-HNE has been reported to reduce leptin-stimulated STAT3 phosphorylation in a dose-dependent manner through the eIF2α–CHOP pathway; global tyrosine phosphorylation was unaffected, suggesting relative selectivity for leptin signalling [34]. Dehydroascorbic acid similarly raised GRP78, CHOP, and spliced XBP1 while lowering pSTAT3 [39], and redox disturbance was shown to affect leptin signalling in selenoprotein M deficiency as well [29].

6.1.3. Nutrient-Derived ER Stress

Nutrient excess is not the only relevant input; nutrient deprivation and specific metabolites also affect ER stress and leptin signalling. Glucose deprivation activates the UPR by increasing GRP78 and CHOP expression while reducing pSTAT3 and pSTAT5 [37], and elevated homocysteine selectively suppresses STAT3 activity [26]. From a developmental perspective, protein restriction combined with a high-fat diet has been associated with hypothalamic leptin resistance, a response attenuated by taurine supplementation [40].

Findings that several UPR arms are activated together across independent models are consistent, and CNS-specific Tak1 deletion offers causal support [38]; the load-versus-capacity interpretation, by contrast, is an inference drawn from the available data.

6.2. Molecular Convergence of Distinct Stressors on Leptin Resistance Effectors

One of the patterns that stands out in this review is that stressors as varied as sleep fragmentation, homocysteine, 4-HNE, glucose deprivation, Western-type diet, and maternal obesity all appear to act through a limited set of shared effectors—SOCS3, PTP1B, and CHOP. Of these, CHOP is the strongest candidate integrating mechanism: CHOP haploinsufficiency protects against hypothalamic leptin resistance [41], CHOP participates directly in regulation of the sOb-R promoter [31], and the effects of 4-HNE are mediated through the eIF2α–CHOP axis [34]. The term “shared effector” should nevertheless be understood as referring to the identity of the molecules involved rather than to a uniform direction of change: as the Section Context-Dependence of the Shared Effectors sets out, the sign of the effect depends on the tissue examined, on the stimulation state at the time of measurement, and on the genetic design used.

The localisation of PTP1B to the ER membrane provides a structural basis for considering SOCS3, PTP1B, and ER-stress-dependent effectors within a common proteostatic context. That αMSH lowers GRP78 and CHOP while suppressing SOCS3 and raising pSTAT3 further suggests that these mechanisms interact within a connected regulatory network rather than operating as independent processes [35]. The therapeutic hypothesis that follows is that modulating the mechanisms at which stressors converge might achieve broader efficacy than targeting each stressor separately; this approach has not yet been tested directly.

This shared pattern suggests that particular regulatory mechanisms may be common to distinct pathophysiological inputs. For CHOP, genetic findings support a causal role; for SOCS3 and PTP1B, the evidence is weaker. Because the stressors have not been assessed within the same model under comparable conditions, it remains difficult to determine whether the observed convergence is a general biological property or an apparent pattern arising from similarities between models (Table 3). Molecular convergence is therefore a hypothesis requiring direct comparative testing.

Table 3. Testable hypotheses of the proposed framework. Each row gives a hypothesis, the outcome expected if it holds, an experimental design capable of testing it, the finding that would falsify it, and the extent to which it has been tested. The final column is excluded from the colour coding because it reports the state of testing rather than the model system from which the evidence derives.

Hypothesis	Expected Outcome	Proposed Experiment	Falsifying Finding	State of Testing
Hypothesis 1—Input independence	In the hypothalamus, and at matched basal and post-leptin-challenge time points, the effector signature (CHOP ↑; basal SOCS3 ↑ with blunted leptin-induced SOCS3; PTP1B ↑) is similar whatever the stressor.	Parallel administration of five stressors (saturated lipid load, 4-HNE, homocysteine, glucose deprivation, sleep fragmentation) in the same strain, at the same age and time point, with both basal and leptin-stimulated states sampled in each animal and effector signatures compared by hypothalamic single-cell RNA-seq.	Stressors producing discrete, non-overlapping effector signatures within the same tissue, genetic background and stimulation state.	Indirectly consistent; no direct comparative study
Hypothesis 2—Node primacy	Intervening at the shared effector node is more effective than targeting upstream stressors individually.	Direct comparison, within the same DIO model, of CHOP silencing against three upstream interventions (antioxidant, anti-inflammatory, lipid reduction); primary endpoint weight-matched leptin sensitivity.	Any upstream intervention matching or exceeding the node intervention.	Partially supported (CHOP+/−; chemical chaperones)
Hypothesis 3—Proteostasis–stress distinction	Disrupted ER proteostasis can produce leptin resistance without activation of the PERK arm and without UPR activation sufficient to account for the signalling defect.	Measurement of leptin sensitivity in POMC-specific SEL1L/HRD1 deletion while UPR markers (BiP, sXBP1, CHOP) remain unchanged; proteostatic capacity monitored independently of the UPR.	Leptin resistance arising only when it accompanies UPR activation.	Partly supported [24]: receptor retention and loss of signalling occurred without PERK activation, although BiP and spliced XBP1 were moderately elevated under high-fat feeding
Hypothesis 4—Load-to-capacity ratio	What matters is not absolute lipid load but its ratio to ER folding capacity.	Measurement of the effect of the same high-fat load on leptin sensitivity in mice with genetically increased ER folding capacity (POMC-XBP1s).	Capacity-enhanced animals developing the same degree of leptin resistance at the same dietary load.	Indirect support [18,38]
Hypothesis 5—Translatability to humans	Agents targeting ER proteostasis improve leptin sensitivity in humans.	Randomised, placebo-controlled, weight-matched trial of TUDCA or 4-PBA; primary endpoint a standardised measure of leptin sensitivity.	No difference in leptin sensitivity versus placebo once weight is matched.	Untested—Kars et al. [53] and Xiao et al. [54] measured no leptin endpoint

Context-Dependence of the Shared Effectors

Two findings in the core evidence base run counter to a simple reading in which CHOP and SOCS3 act as uniformly inhibitory components of a single pathway, and they are best addressed directly.

For CHOP, the apparent conflict concerns tissue and genetic design rather than the results themselves. In the hypothalamus, CHOP is stress-induced, and partial loss of function is protective: CHOP haploinsufficiency prevented the leptin resistance produced by chronic sleep fragmentation [41]. In the liver, CHOP performs a basal transcriptional function, binding the Lepr promoter and sustaining sOb-R expression, and its complete, constitutive, whole-body deletion abolished the hepatic pSTAT3 response to leptin [31]. A partial reduction that blunts stress-induced expression and a null allele that also removes a constitutive function are not equivalent manipulations, and the two studies examined different organs. This review therefore treats CHOP as a context-dependent node carrying both a basal transcriptional role and a stress-induced inhibitory role, not as a uniformly inhibitory effector.

For SOCS3, the distinction is between feedback and inhibition. SOCS3 is transcribed in response to STAT3 activation, so a leptin-stimulated rise in SOCS3 indicates that the LepRb–JAK2–STAT3 axis is transmitting a signal; this is what was observed when PBA restored the inducible SOCS3 response of ob/ob islets to the lean pattern [27]. Resistance is instead associated with a chronically elevated basal level in the unstimulated state, as reported for the hypothalamus in congenital hypothyroidism [33] and for adipose tissue under high-fat feeding [35], accompanied by a blunted response to leptin challenge; hypothalamic SOCS3 in the cafeteria-diet model [36] runs in the opposite direction and is considered below. Studies reporting a single unstimulated SOCS3 measurement cannot distinguish the two situations, and this is a limitation of much of the evidence summarised here. Consistent with this distinction, the earliest demonstration that ER stress blocks leptin-induced STAT3 phosphorylation attributed the effect to PTP1B rather than to SOCS3 [16].

Both observations narrow the claim made in this section. The evidence supports the proposition that a small number of molecules recur across otherwise unrelated stressors; it does not yet support the stronger proposition that they move in the same direction, in the same tissue, under every input. Hypothesis 1 (Table 3) is restated accordingly, so that it applies to the hypothalamus, to a defined stimulation state, and to comparable genetic designs.

How far this convergence is a measured phenomenon rather than an inferred one can now be stated exactly. Across the twenty-two studies in the core evidence base, CHOP was measured in twelve, SOCS3 in eight, and PTP1B in five, and only two measured all three (Table 1) [28,36]. The pattern described in this section is therefore assembled across studies that each examined a different subset of the effector set; it is not a convergence that any single experiment has observed. Several of the measurements also run against the expected direction. In cafeteria-diet-fed rats, both hypothalamic SOCS3 and hypothalamic PTP1B were lower, not higher, than in control animals; hypothalamic SOCS3 rose significantly with the treatment that improved leptin signalling, whereas the parallel rise in hypothalamic PTP1B did not reach significance, so that significant restoration of both effectors was confined to skeletal muscle [36]; hypothalamic SOCS3 did not change at all under chronic sleep fragmentation, in the same study in which PTP1B expression and activity both increased [41]. Taken with the tissue and stimulation-state distinctions above, this is the strongest reason to treat the effector set as a recurring group of candidates rather than as a demonstrated common pathway.

6.3. A Possible ER Stress-Mediated Link Between Leptin and Insulin Resistance

The metabolic consequences of ER stress may not be confined to leptin signalling; similar mechanisms may also contribute to insulin resistance, placing the two disorders on a shared cellular footing. The facts that loss of SEL1L in POMC neurons is associated with hepatic steatosis, glucose intolerance, and insulin resistance, as well as leptin resistance [24] and that TUDCA improved findings related to both forms of resistance in a maternal obesity model [32] lend support to this possibility. A plausible molecular basis is that PTP1B dephosphorylates both JAK2 and the insulin receptor and that PI3K/Akt is a signalling component shared by the two hormones. No direct human evidence shows that correcting ER proteostasis improves both forms of resistance together. Table 4 compares the mechanisms discussed in Section 5 and Section 6 by the UPR arm affected, the point at which each acts, and the strength of the evidence supporting it.

Table 4. Comparison of molecular mechanisms. Comparison of the mechanisms leading to leptin resistance by UPR arm affected, point of intervention, and level of evidence. The evidence column is colour-coded by the model system in which each mechanism was demonstrated: yellow = animal study, red = laboratory (in vitro) study.

Mechanism/Level	UPR Arm	Point of Action	Key Studies	Level of Evidence (Model System)
LepRb quality control (ERAD)	No PERK-arm activation; basal ERAD rather than a stress response (BiP and spliced XBP1 moderately elevated under high-fat feeding)	Receptor biogenesis; retention of LepRb in the ER	[24]	Animal study—tissue-specific conditional genetics
CHOP-mediated regulation of sOb-R	PERK–eIF2α–CHOP	Receptor bioavailability; circulating free leptin	[31]	Animal study—genetic (CHOP-KO) + pharmacological
PTP1B-mediated JAK2 dephosphorylation	All three arms	Post-receptor signal termination	[23,36,41]; in [36], hypothalamic PTP1B was reduced, not increased, by the obesogenic diet and was not significantly altered by the treatment—see Section Context-Dependence of the Shared Effectors	Animal study—genetic (CHOP+/−) + correlative
SOCS3-mediated feedback inhibition	PERK and IRE1α	LepRb Tyr985; suppression of JAK2	[27,33,35]	Animal study—largely correlative
Redox/oxidative ER stress	PERK–eIF2α–CHOP	Selective inhibition of STAT3 phosphorylation	[29,34,39]	Predominantly laboratory (in vitro)—limited animal support
Lipid load/ER capacity balance	IRE1α–XBP1; ATF6	ER membrane capacity; SREBP-mediated expansion	[20,25,38]	Animal study—conditional genetics + diet
Neuroinflammation–autophagy axis	All three arms	Microglia–neuron interaction; autophagic flux	[23,28]	Animal study—weight-matched control
Developmental programming	PERK–eIF2α–CHOP	Melanocortin circuit development; sustained SOCS3 elevation	[32,40,55]	Animal study—reversible by intervention

6.4. The Causality Problem: Separating Direct Effects from Weight Loss

Most of the intervention studies summarised here share a common interpretive difficulty: nearly all report reductions in body weight and/or adiposity alongside the intervention (Table 5). Because weight loss by itself reduces hypothalamic ER stress and leptin resistance, it cannot be determined whether the observed improvement reflects a direct effect of the intervention or the accompanying weight loss.

Table 5. Molecular target–dietary/pharmacological modulator–level of evidence. Molecular targets identified in leptin resistance and their corresponding dietary or pharmacological modulators. The evidence column indicates the model system in which each pairing was demonstrated and is colour-coded accordingly: green = human data, yellow = animal study, red = laboratory (in vitro) study.

Molecular Target	Nutritional Strategy	Model/System	Level of Evidence
ER stress/UPR (general)	Reduction in saturated fat and excess caloric load	Mouse, ApoE−/− [20,25]	Animal study—dietary intervention
GRP78, CHOP, XBP1	Indole-3-carbinol (cruciferous vegetables)	Mouse, ApoE−/− [20]	Animal study—dietary intervention
ER stress + JAK2/STAT3	Phenethyl isothiocyanate (cruciferous vegetables)	Mouse, C57BL/6J [21]	Animal study—dietary intervention
ER stress + cholesterol	Diallyl disulfide (garlic)	Mouse, ApoE−/− [25]	Animal study—dietary intervention
STAT3/SOCS3/PTP1B	Proanthocyanidins (grape seed)	Rat, cafeteria diet; hypothalamus (STAT3, SOCS3) and skeletal muscle (SOCS3, PTP1B) [36]	Animal study—dietary intervention
ER stress + autophagy	Berberine (FTZ preparation)	Rat, DIO + BV2/GT1-7 [23]	Animal study + laboratory (in vitro)
Hypothalamic ER stress	Taurine supplementation	Mouse, protein restriction + HFD [40]	Animal study—dietary intervention
GRP78/ER-derived cell death	Biochanin A (isoflavone)	Neuronal cell line [30]	Laboratory (in vitro) only
PTP1B + AMPK	Cannabisin A and B (hemp seed)	In vitro + computational [56]	Laboratory (in vitro) only
Neuronal leptin responsiveness	Rubisco-derived plant peptide	Rodent, oral administration [57]	Animal study—oral administration
Gut–hypothalamus axis	Microbiota-derived short-chain fatty acids	Human/animal review [58]; rat RYGB [28]	Animal study—indirect; mechanism untested
Developmental programming (SOCS3)	Modulation of maternal diet quality	Mouse/rat [32,46,55]	Animal study—critical-window intervention
ER proteostasis (chaperone)	Pharmacological: TUDCA/4-PBA	Human [53,54]	Human data—but insulin endpoint

The only study whose design overcomes this difficulty is the Roux-en-Y gastric bypass (RYGB) study of Chen et al. [28], which used a weight-matched caloric restriction control and showed that improvements in hypothalamic ER stress and leptin sensitivity were specific to RYGB. It is thus the sole direct basis for the claim that reducing ER stress can restore leptin sensitivity independently of weight loss, and it awaits independent replication; wider adoption of weight-matched controls in this field is needed.

A second difficulty concerns the target specificity of chemical chaperones: beyond reducing ER stress, PBA and TUDCA act on numerous targets, including histone deacetylase inhibition and bile acid signalling. Findings obtained with these agents therefore provide supportive but limited evidence, whereas findings based on CHOP genetics constitute a firmer basis.

Third, findings obtained solely in vitro warrant additional caution: four of these studies used SH-SY5Y neuroblastoma cells transfected to overexpress Ob-Rb, a tumour line that does not carry hypothalamic neuronal identity, and both receptor density and stressor concentrations lie outside the physiological range. Such studies are valuable for generating hypotheses but are not sufficient for inference about leptin resistance in vivo.

7. Intercellular Interactions

Leptin resistance is less a cell-autonomous event within an isolated neuron than the product of continuous interaction among hypothalamic neurons, microglia, astrocytes, and peripheral tissues. The low-grade inflammatory response that overnutrition induces in the hypothalamus is accompanied by ER stress induction and defective autophagy, and the process is mediated through interaction with non-neuronal cells [23,28]. A comparable relationship has been described at the level of sympathetic output: silencing Sirt1 or FoxO1 in the arcuate nucleus reduces leptin-mediated inflammation and sympathetic nerve activity [50].

The strongest evidence that this interaction can be reversed comes from bariatric surgery: RYGB improved hypothalamic gliosis, inflammatory signalling, and ER stress in diet-induced obese rats; the improvement was associated with increased central leptin responsiveness; and it was not observed in the weight-matched control [28] (Section 6.4).

Autophagy represents a further node in this network: a berberine-containing herbal preparation attenuated hypothalamic ER stress and leptin resistance through modulation of autophagy, with the findings confirmed in both microglial and neuronal cell models [23]. In the periphery, αMSH attenuates ER-stress-induced leptin resistance in adipose tissue, whereas ER stress induced by tunicamycin or thapsigargin reduces leptin sensitivity [35]—pointing to a link between central melanocortin signalling and peripheral adipose expansion that operates through ER stress.

Tissue and Cell-Type Specificity of the Evidence

The framework proposed here is formulated for the hypothalamus, but the evidence supporting it is not drawn from the hypothalamus alone, and the distinction matters because the same molecules do not behave identically across tissues. Of the twenty-two studies in the core evidence base, twelve measured outcomes in the hypothalamus, five in peripheral tissues alone (two in liver and aorta [20,25], one in adipose tissue [35], one in pancreatic islet [27], and one in liver [31]), and five in neuronal cell lines that do not carry hypothalamic identity. The model column of Table 1 names the tissue in each case.

Within the hypothalamic group, a second distinction is needed between neurons and the non-neuronal cells around them. Two studies attribute their effect to microglia rather than to neurons: the improvement in hypothalamic ER stress after gastric bypass was transferred through a circulating factor acting on microglia, which then altered POMC neurons [28], and the autophagy-dependent effect of a herbal preparation was reproduced in microglial and neuronal lines separately [23]. A measurement made in hypothalamic homogenates cannot distinguish these cell populations, and most of the included studies report homogenates.

Two conclusions follow. The claim that ER stress impairs leptin signalling is best supported in hypothalamic tissue, but the direction of the effect of individual components has been established partly in peripheral tissues, where it differs: CHOP is inhibitory in the hypothalamus and required for basal sOb-R production in the liver (Section Context-Dependence of the Shared Effectors), and the adipose and islet studies describe ER stress acting on leptin production and local responsiveness rather than on central signalling. This is the concrete reason for the single-cell and spatial approaches proposed in Section 11.4: the resolution of the measurements in the included studies is lower than the resolution the framework requires.

8. Developmental Programming and Epigenetic Regulation

Evidence that leptin resistance can be permanently programmed by early-life conditions continues to accumulate.

8.1. Developmental Programming

Maternal obesity has been reported to disrupt the development of melanocortin circuits in offspring, producing neonatal hyperleptinaemia and leptin resistance; to upregulate ER-stress-related genes in the hypothalamus; and to be corrected—in both its metabolic and its neurodevelopmental components—by neonatal TUDCA [32]. That an intervention delivered within a critical developmental window can alter an otherwise persistent phenotype points to a causal role for early-life ER stress.

Programming has been shown to begin at the level of the placenta: maternal obesity activates placental insulin/IGF-I/mTOR and leptin pathways [59], and placental dysfunction is associated with increased SOCS3 and PTPN2 expression [60]. The same axis operates postnatally, where perinatal diet composition and gestational thyroid hormone insufficiency produce persistent and partly sex-specific hypothalamic leptin signalling defects in adulthood [33,40,46].

The most direct evidence that this programming is reversible comes from silencing maternal SOCS3: a Socs3-shRNA lentivirus administered in early gestation to obese dams attenuated postnatal obesity in the offspring [55]. This suggests that increased hypothalamic SOCS3 is not merely a marker of early-life programming but also a mediator of it.

8.2. Epigenetic and Post-Translational Mediators

The first layer of epigenetic regulation comprises histone modification and the acetylation balance. Hypothalamic leptin action has been shown to be mediated by histone deacetylase 5 (HDAC5), which shapes the transcriptional output of leptin signalling at the chromatin level [61]. In parallel, hypothalamic expression of the nuclear receptor Nur77 falls in obese mice, leptin fails to reduce body weight in Nur77 knockout animals, and Nur77 recruits p300 and dissociates HDAC1, thereby increasing STAT3 acetylation and transcriptional activity [62].

The second layer has been defined at the level of DNA methylation and non-coding RNAs, and it includes human data: adiposity-associated DNA methylation signatures in adolescents have been related to leptin concentrations and perinatal factors [63], and circulating microRNAs in children with obesity have been associated with leptin resistance [64]. Experimentally, a maternal high-fat diet impairs leptin signalling in the neonatal hypothalamus and upregulates the type-1 cannabinoid receptor through sex-specific epigenetic changes [65].

The two layers do not rest on evidence of equal strength. Findings on developmental programming carry relatively strong causal support because they are reversible by intervention within a critical window [32,55], whereas the DNA methylation and microRNA findings are cross-sectional and correlational. More importantly, no direct mechanistic link between ER stress and epigenetic marks has been demonstrated; the fact that these two lines of evidence remain unintegrated is the most conspicuous gap in the framework proposed here.

9. A Unified Mechanistic Model and the Proposed Conceptual Framework

9.1. Integrating the Findings into a Single Mechanistic Model

The findings summarised in this review can be read not as a series of independent observations but as a single mechanistic route (Figure 3). In the proposed model, stressors that appear unrelated—lipid overload, oxidative products, nutrient-derived stressors, maternal obesity, and sleep fragmentation—converge on hypothalamic ER stress; ER stress activates all three arms of the UPR; and that activation converges on a narrow set of effectors (CHOP, SOCS3, and PTP1B). These effectors can be argued to suppress LepRb–JAK2–STAT3 signalling and so to produce leptin resistance, which in turn drives hyperphagia and increased adiposity, feeds back onto ER stress, and closes a self-reinforcing loop. No single study has demonstrated the route in its entirety. The model also keeps the two arms separate by design: what is shown on the left of Figure 3 concerns receptor maturation and surface availability, and no evidence connects it to the CHOP–SOCS3–PTP1B node.

Figure 3. ER stress-mediated leptin resistance: a unified mechanistic model. Two mechanistically distinct arms are shown. In the ER stress-dependent arm (solid arrows), unrelated stressors converge on hypothalamic ER stress, which activates all three UPR branches and recruits a narrow effector set (CHOP, SOCS3, and PTP1B); the direction of effect within that set is tissue- and context-dependent (Section Context-Dependence of the Shared Effectors). In the proteostatic (ERAD) arm, labelled in the figure as the PERK-independent receptor quality-control arm (dashed arrows), which proceeds without activation of the PERK arm rather than in the absence of the entire ER stress response—BiP and spliced XBP1 are moderately elevated under high-fat feeding (Section 9.1)—loss of SEL1L–HRD1 ERAD causes LepRb misfolding and retention within the ER and so reduces the receptor available at the cell surface [24]; this arm acts directly on LepRb–JAK2–STAT3 signalling and does not pass through the effector set. The red loop denotes the self-reinforcing cycle in which leptin resistance, hyperphagia, and increased adiposity feed back onto hypothalamic ER stress. DHAA, dehydroascorbic acid; 4-HNE, 4-hydroxynonenal. No single study has demonstrated the complete route.

A second arm runs parallel to this route: LepRb quality control mediated by the SEL1L–HRD1 ERAD system, which operates without activation of the PERK arm [24]. Mao et al. showed that loss of SEL1L in POMC neurons causes the receptor to be retained within the ER and attenuates leptin signalling without detectable activation of the PERK arm; CHOP, SOCS3, and PTP1B were not examined in that study, and no work has tested whether this arm engages them. It is therefore represented in Figure 3 as acting directly on LepRb–JAK2–STAT3 signalling, bypassing the effector node. On this view, the ER can shape leptin signalling through both its stress response and its basal proteostatic function, by two routes that remain mechanistically distinct and should not be collapsed into one another. The independence from ER stress should be stated precisely rather than absolutely: in SEL1L-deficient POMC neurons, the receptor was retained without detectable activation of the PERK arm, while BiP and spliced XBP1 were moderately elevated under high-fat feeding [24]. What that study supports is therefore that receptor retention and the loss of leptin signalling occur without PERK activation, not that the ER stress response is wholly absent.

9.2. ER Proteostasis as a Convergence Node

On this basis, the following testable framework is proposed:

Endoplasmic reticulum proteostasis may function as a central convergence node that translates metabolic, inflammatory, developmental, and nutrient-derived signals into leptin resistance.

What distinguishes the framework is that it positions the ER not as one step in a chain leading to leptin resistance but as a proposed regulatory hub converting different classes of input into a common output. Three hypotheses follow, each corresponding to a directly testable experimental design.

The first is that the effector signature is independent of the input: whatever the stressor, a similar effector pattern encompassing CHOP, SOCS3, and PTP1B would be expected. The available data neither support nor refute this expectation with any precision: only two of the twenty-two studies measured all three effectors, several measured none of them, and two of the measurements that were made run against the predicted direction (Section Context-Dependence of the Shared Effectors). No study has tested different stressors within the same model under comparable conditions.

The second is that intervening at the shared effector node will prove more effective than targeting upstream stressors individually. Work on chemical chaperones and CHOP genetics offers partial support, but no study has compared node-level intervention with upstream intervention within a single model.

The third is that disrupted ER proteostasis can produce leptin resistance without activation of the PERK arm and without UPR activation sufficient to account for the signalling defect. The SEL1L–HRD1 ERAD finding of Mao et al. [24] is the only direct evidence supporting this hypothesis and represents the most original aspect of the framework, although the support is partial rather than complete, since receptor retention occurred without PERK activation while BiP and spliced XBP1 were moderately elevated (Section 9.1); were it confirmed, it would follow that studies measuring only UPR markers are missing part of the leptin resistance phenotype. Whether this arm ultimately converges on the same effectors is an open question rather than an assumption of the model, and Figure 3 represents it as unconnected to them for that reason.

The framework requires the ER stress and ER proteostasis literature to be brought together, and it carries an explicit falsification criterion: a demonstration that different stressors produce markedly different effector signatures would invalidate it.

Table 3 lists these three hypotheses, together with two further hypotheses concerning the load-to-capacity ratio and translatability to humans, in terms of the expected outcome, the proposed experimental design, the falsifying finding, and the current state of testing.

10. Clinical Implications

Details about the molecular mechanism acquire clinical meaning only to the extent that they can be translated into patient care. Figure 4 sets out the therapeutic approaches considered in this section by level of intervention, together with the nature of the evidence available for each.

Figure 4. Therapeutic targets and levels of evidence. Therapeutic approaches by level of intervention and the nature of the evidence available for each. Receptor quality control (SEL1L–HRD1 ERAD) is shown as a category separate from interventions directed at the UPR effectors (CHOP, SOCS3, and PTP1B) because ERAD is a basal proteostatic mechanism rather than a stress-signalling pathway (Section Context-Dependence of the Shared Effectors and Section 9.1).

10.1. Clinical Outcomes of Leptin Replacement Therapy

Recombinant leptin has been tested in common obesity in several randomised, placebo-controlled trials, and the results have been consistently modest (Table 6). The largest, reported by Zelissen et al. [66], randomised 284 overweight and obese adults, after a three-week diet lead-in, to placebo or to 10 mg of recombinant methionyl human leptin given each morning, each night, or twice daily for twelve weeks; weight loss of 2.6 to 3.4 kg did not differ between the arms, and no schedule was superior to any other. In the earlier dose-escalation study of Heymsfield et al. [67], 54 lean and 73 obese participants received daily subcutaneous leptin (0.01–0.30 mg/kg/day). The two cohorts were not studied in parallel: lean participants were treated for four weeks on a weight-maintaining diet, and only the obese participants continued into the 24-week phase alongside a 500 kcal/day energy deficit. A significant dose–response relationship for weight loss was reported in the obese cohort, with the largest mean loss at the highest dose of 0.30 mg/kg/day (p = 0.01 across doses among the 47 obese participants who completed 24 weeks), although comparisons between individual dose groups did not reach significance. The size of that effect depends on the analysis applied: mean weight change at 0.30 mg/kg/day was −7.1 (8.5) kg against −1.3 (4.9) kg on placebo among completers (n = 8 and n = 12), but −3.3 (6.7) kg against −1.0 (3.8) kg when the last observation was carried forward (n = 18 and n = 20), so that the placebo-adjusted difference at the highest dose falls from about 5.8 kg, roughly 6% of baseline body weight, to about 2.3 kg, roughly 2.5%. Pegylated leptin was tested in the same period: weekly PEG-OB produced no difference from placebo in obese men [68], and when combined with a very-low-energy diet (2.1 MJ/day), it produced significantly greater weight loss than placebo during treatment (p < 0.03), followed by faster and larger regain once treatment stopped [69]. Two trials in type 2 diabetes, in which leptin might plausibly have acted on glycaemia rather than on weight, were also negative for body weight: fourteen days of recombinant leptin at 30 or 80 mg per day changed neither body weight nor body composition nor insulin action in eighteen participants [70], and sixteen weeks of metreleptin in seventy-one participants left body weight unchanged while reducing HbA1c only marginally, from 8.01% to 7.96% [71]. The hormone has also been given to people in whom it might be expected to work better: in 27 women studied at least eighteen months after Roux-en-Y gastric bypass, when circulating leptin is low, metreleptin twice daily for sixteen weeks changed neither body weight, fat mass, nor resting energy expenditure [72].

Table 6. Randomised trials of leptin in common obesity. Randomised, placebo-controlled trials of recombinant or pegylated leptin in people with overweight or obesity who are not leptin-deficient, in order of publication. Trials in congenital leptin deficiency and lipodystrophy are not included because leptin is replacement therapy in those conditions. The weight outcome column reports the comparison with placebo rather than the within-group change, since all of these trials were conducted alongside an energy-restricted diet or after weight loss.

Trial (Author, Year, Reference)	Population	n	Intervention	Duration	Weight Outcome Versus Placebo
Heymsfield et al., 1999 [67]	Lean and obese adults	54 lean, 73 obese	Recombinant leptin 0.01–0.30 mg/kg/day subcutaneously; obese cohort with a 500 kcal/day energy deficit, lean cohort on a weight-maintaining diet	Lean 4 weeks; obese 24 weeks	Significant dose–response relationship for weight loss in the obese cohort, with the largest mean loss at 0.30 mg/kg/day (p = 0.01 across doses, 47 obese completers): −7.1 (8.5) kg versus −1.3 (4.9) kg on placebo among completers (n = 8 and n = 12), and −3.3 (6.7) kg versus −1.0 (3.8) kg on last-observation-carried-forward analysis (n = 18 and n = 20); placebo-adjusted difference about 5.8 kg (≈6% of baseline weight) among completers, about 2.3 kg (≈2.5%) under imputation; comparisons between individual dose groups were not significant, dose groups were small, and attrition was considerable
Hukshorn et al., 2000 [68]	Obese men	30	Pegylated recombinant leptin (PEG-OB) weekly subcutaneously, with an energy deficit	12 weeks	No difference from placebo
Lejeune et al., 2003 [69]	Overweight men	24	PEG-OB 80 mg weekly subcutaneously, with a 2.1 MJ/day very-low-energy diet	6 weeks of treatment, 8 weeks of follow-up	Significantly greater weight loss than placebo during treatment (p < 0.03), followed by faster and larger regain during follow-up
Zelissen et al., 2005 [66]	Overweight and obese adults after a 3-week diet lead-in	284	Recombinant methionyl human leptin 10 mg subcutaneously each morning, each night, or twice daily	12 weeks	Weight loss of 2.6–3.4 kg, with no difference between any leptin schedule and placebo
Mittendorfer et al., 2011 [70]	Adults with newly diagnosed type 2 diabetes	18	Recombinant methionyl human leptin 30 or 80 mg/day subcutaneously	14 days	No change in body weight or body composition, and no effect on insulin action
Moon et al., 2011 [71]	Adults with obesity and type 2 diabetes	71	Metreleptin subcutaneously, randomised 2:1 against placebo	16 weeks	No change in body weight; HbA1c reduced marginally (8.01% to 7.96%, p = 0.03)
Korner et al., 2013 [72]	Women at least 18 months after Roux-en-Y gastric bypass	27	Metreleptin 0.05 mg/kg twice daily subcutaneously, crossover design	16 weeks per arm	No difference from placebo in body weight, fat mass, or resting energy expenditure

Two conclusions follow from this body of work, and they should be kept apart. The first is that the clinical benefit of raising circulating leptin in common obesity is limited; inconsistent across doses, schedules, formulations, and degrees of preceding weight loss; and not reliably sustained after treatment ends. The qualification is deliberate because the evidence is not uniformly null: Heymsfield et al. [67] reported a significant dose–response relationship for weight loss in the obese cohort, which indicates that exogenous leptin can exert a measurable pharmacological effect on body weight and that a blanket statement of no effect would overstate what these trials show. Three features of that study nevertheless limit how far the finding can be taken. It was a dose-escalation design in which comparisons between individual dose groups did not reach significance, so the trend rests on the ordering of group means rather than on any demonstrated separation between doses; the dose groups were small and the between-individual variance in weight change wide, so the mean loss at the highest dose reflects a subset of responders rather than a consistent response; and attrition over the 24 weeks was considerable, which leaves the completer-based estimate open to bias: carrying the last observation forward reduces the mean loss at the highest dose from 7.1 to 3.3 kg and the placebo-adjusted difference from about 5.8 to about 2.3 kg. Read together with the larger and longer trials that followed [66,68,69,70,71,72], in which the only separation from placebo was the additional weight loss observed during PEG-OB treatment (p < 0.03), itself followed by greater regain after treatment ended [69], the body of work indicates a benefit that was limited; inconsistent across doses, schedules, and formulations; and not reliably sustained, rather than an effect of therapeutic magnitude. It does not support leptin administration as a treatment for common obesity. The second does not follow from these trials at all. That limited and inconsistent benefit is consistent with the framework proposed here, in which resistance is generated at the receptor, post-receptor, and bioavailability levels at once, so that adding ligand corrects none of them, but it is equally consistent with saturable transport of leptin into the brain [12], with the pharmacokinetics of the preparations used, and with the enrolment of unstratified populations in which any mechanism-specific subgroup would be diluted (Section 10.5). The second of these alternatives is not hypothetical: in the metreleptin trial in type 2 diabetes, total leptin, leptin-binding protein, and anti-leptin antibodies all rose during treatment, so that free leptin remained limited despite administration [71], which is a bioavailability-level explanation of the kind described in Section 5 and is independent of any hypothalamic mechanism. A trial of a ligand, whether negative or weakly positive, identifies neither the step at which the pathway fails nor the mechanism responsible, so these results cannot be read as support for an ER origin of leptin resistance, and they are not treated as such here. What they do establish is the clinical premise from which this review proceeds: in common obesity, the aim is not to administer leptin but to repair the response to it. States of genuine deficiency, such as congenital leptin deficiency and lipodystrophy, fall outside this assessment.

10.2. Human Data on Chemical Chaperones

Chemical chaperones have been tested in humans, but leptin endpoints were not assessed. Kars et al. [53] gave TUDCA (1750 mg/day) or placebo to 20 participants with obesity for four weeks and reported an approximately 30% increase in hepatic and muscle insulin sensitivity in the TUDCA arm. Xiao et al. [54] showed that two weeks of oral sodium phenylbutyrate (7.5 g/day) partly attenuated lipid infusion-induced insulin resistance.

These two trials point in two directions at once. Interventions targeting ER stress are feasible in humans, are tolerated, and can produce measurable metabolic benefit, and the interest is extending to other metabolic diseases [73,74]. Against this, both trials measured only insulin endpoints, the samples were small (n = 20 and n = 8), and the exposures were brief; the assumption that chemical chaperones correct leptin resistance in humans therefore remains untested and represents the most concrete gap in the field.

10.3. PTP1B as a Therapeutic Target

PTP1B is the member of the proposed effector node that appears most amenable to drug development [75]. The aminosterol trodusquemine was reported to suppress appetite and produce fat-specific weight loss in diet-induced obese mice [76] and was well tolerated in phase 1 trials, yet the development programme for the obesity indication was not pursued.

The reason is not inefficacy alone but the structural nature of the target: the catalytic site of PTP1B is highly polar and charged, molecules that bind it remain too polar to cross the cell membrane, and structural similarity to other members of the phosphatase family creates a selectivity problem. Current work has therefore turned to allosteric and natural product-derived inhibitors [56]. Correctly identifying an effector node, in other words, does not make that node druggable.

10.4. Bariatric Surgery and Incretin-Based Therapies

The mechanism of bariatric surgery bears directly on the proposed framework (Section 6.4 and Section 7): that RYGB improves hypothalamic ER stress and inflammation, that this is associated with increased central leptin sensitivity, and that it is not seen in a weight-matched control [28] offer a mechanistic explanation for why surgically induced weight loss is more durable than equivalent loss achieved by diet.

GLP-1 receptor agonists raise a different question: liraglutide has been reported to attenuate the fall in free plasma leptin during maintenance of weight loss [77]. Since weight loss normally lowers free leptin and thereby facilitates weight regain, this effect may contribute to the drug’s efficacy. The study did not, however, assess sensitivity, and the effect of GLP-1 receptor agonists on hypothalamic ER stress has not been examined in humans. Table 7 summarises the interventions that target this axis, together with the current state of human data for each.

Table 7. Therapeutic interventions and the state of human data. Interventions targeting the ER stress–leptin resistance axis; the human data column indicates the current limit of clinical translation. The column is colour-coded by the model system in which the intervention was evaluated: green = human data, yellow = animal study.

Agent/Approach	Class	Model	Key Endpoint	Human Data
TUDCA	Chemical chaperone	Mouse [32,41]; human [53]	Leptin sensitivity ↑ (rodent); hepatic/muscle insulin sensitivity +30% (human)	Yes—but no leptin endpoint (n = 20, 4 weeks)
4-PBA (sodium phenylbutyrate)	Chemical chaperone	Mouse [27]; human [54]	UPR markers ↓, SOCS3 modulation; lipid-induced insulin resistance ↓	Yes—no leptin endpoint (n = 8, 2 weeks)
Recombinant leptin	Hormone replacement	Human RCT [67]	Placebo-adjusted weight loss at the highest dose ≈6% of baseline weight among completers, ≈2.5% on last-observation-carried-forward analysis	Yes—benefit limited, inconsistent and not sustained
Trodusquemine (MSI-1436)	PTP1B inhibitor	Mouse [76]; phase 1/1b	Fat-specific weight loss; appetite suppression	Phase 1 completed; obesity programme not pursued
GLP-1RA (liraglutide)	Incretin analogue	Human RCT [77]	Prevention of the fall in free leptin during weight-loss maintenance	Yes—leptin level measured, sensitivity not
Roux-en-Y gastric bypass	Surgical	Rat [28]	Hypothalamic ER stress ↓, leptin sensitivity ↑ (independent of weight loss)	Indirect—mechanism untested in humans
PEITC, I3C, DADS, proanthocyanidins, berberine, taurine, biochanin A	Diet-derived phytochemical	Rodent in vivo/in vitro	ER stress markers ↓, pSTAT3 ↑, weight ↓	None—doses above the range attainable through diet
ERα agonism (PPT)	Hormonal	Mouse, ovariectomised [22]	Hypothalamic ER stress ↓, leptin signalling ↑	None

10.5. Molecular Heterogeneity of Leptin Resistance and a Phenotyping Framework

One of the most direct clinical implications of this review is that leptin resistance is not a single entity: the findings in Section 5 and Section 6 show that it can arise at different molecular levels and that a set of phenotypes distinguished by dominant mechanism exists.

Table 8 proposes a phenotyping framework on this basis, linking five questions: which patient, which dominant mechanism, which biomarker, which intervention, and which primary endpoint. Because an intervention effective in only one subgroup will appear ineffective overall in an unstratified trial, establishing this chain is a precondition for making the axis clinically testable. The history of the leptin field illustrates the risk: recombinant leptin produced only limited and inconsistent weight loss in an unselected obese population [67] but was markedly effective in the leptin-deficient phenotype [7].

Table 8. Proposed leptin resistance phenotypes and the chain of clinical testing. The chain: patient–dominant mechanism–candidate biomarker–intervention–primary endpoint. The phenotypes are derived from animal and laboratory studies; none has been confirmed in humans. The final column is colour-coded by the model system on which the phenotype rests: green = human data, yellow = animal study. The biomarkers and the predicted responses in this table are hypothetical. No phenotype listed here has been identified in a human population, no biomarker has been validated for this purpose, and no predicted response has been tested in a trial. The final column states the kind of study each row is extrapolated from and should be read as the basis of the proposal rather than as a level of support; rows resting on human observational data are distinguished there from those extrapolated from animal or cell models.

Proposed Phenotype	Dominant Mechanism	Candidate Biomarker (Hypothetical)	Predicted Response to Intervention (Hypothetical)	Proposed Primary Endpoint	Basis of the Proposal
Leptin-sensitive obesity (reference)	No dominant signalling defect; energy balance load predominates	Leptin within the range expected for fat mass; normal sOb-R	Expected response to caloric restriction and GLP-1 RA	Weight loss; hunger/satiety scores	Human data—clinical observation
ER-dominant (proteostatic) phenotype	Retention of LepRb in the ER, SEL1L–HRD1 ERAD defect; CHOP-mediated regulation of sOb-R	sOb-R ↑/free leptin index ↓; LEPR misfolding variants (e.g., Cys604Ser)	Group expected to benefit most from chemical chaperones (TUDCA, 4-PBA)	Weight-matched leptin sensitivity	Animal study (conditional genetics) + a single human variant report
Inflammatory phenotype	Hypothalamic low-grade inflammation; SOCS3 induction	hsCRP, IL-6, TNF-α; neuroinflammation imaging markers	Anti-inflammatory dietary pattern; RYGB (weight-independent effect)	hsCRP + weight-matched leptin sensitivity	Animal study; indirect in humans [28]
Oxidative phenotype	Lipid peroxidation products (4-HNE); loss of redox protection (selenoprotein M)	Plasma 4-HNE adducts; oxidised LDL; selenium status	Redox-targeted intervention; reduction in saturated fat load	4-HNE adduct level + food intake	Animal study + laboratory (in vitro)
Post-receptor inhibition (PTP1B-dominant)	PTP1B-mediated JAK2 dephosphorylation; localisation to the ER membrane	No direct marker; indirectly, indices of insulin resistance	PTP1B modulation—unconfirmed clinically	HOMA-IR + leptin sensitivity	Animal study; human trial inconclusive (trodusquemine)
Developmental/epigenetic phenotype	Melanocortin circuit programming driven by maternal obesity	Adiposity-associated DNA methylation signatures; circulating miRNA profiles	Intervention within the pregnancy and early childhood window	Childhood adiposity trajectory	Human data (correlational) + animal study [63,64]
Transcriptional balance phenotype	STAT3/FoxO1 balance and STAT3 acetylation status	No known circulating biomarker	No benefit expected from chemical chaperones	Not defined	Animal study; not directly tested

The status of this framework should be stated plainly: none of the proposed phenotypes has been confirmed in humans. The classification is a hypothesis-generating structure derived from animal and laboratory findings, and the biomarkers listed are candidates. Its purpose is not to offer a clinical algorithm but to define a testable stratification hypothesis.

The weakest link in the framework is measurement: hypothalamic ER stress cannot be measured directly in humans, so distinguishing an ER-dominant phenotype depends on unvalidated indirect markers. The second weak link is the endpoint, since no standard method exists for measuring leptin sensitivity in a clinical setting. Until both gaps are addressed, the pairings in Table 8 cannot be tested (Table 2).

10.6. Determinants of Individual Response Variability

The framework assumes that individuals do not respond identically to the same metabolic load—the founding premise of the personalised nutrition literature [78], which has empirical support [79]. A plausible molecular counterpart is that, where ER folding capacity differs between individuals, the same dietary lipid load will generate different degrees of ER stress [38] (Section 6.1.1); this inference remains conceptual.

The first source of variability is genotype. The most direct human evidence comes from the LEPR Q223R (rs1137101) variant, which reduces leptin binding and predicts annual weight gain in childhood obesity [80]. Rare variants illustrate the same principle at the level of proteostasis: the Cys604Ser variant of LepRb is prone to misfolding and is processed as an ERAD substrate [24]. Because no study has shown that a genotype-tailored intervention alters outcome, the evidence remains at the level of a hypothesis [81].

The second source is developmental windows; the determinant of response variability with the strongest evidence base is early-life exposure (Section 8). Its clinical implication is twofold: the timing of an intervention may determine its efficacy, and sex should be treated as an independent response modifier [32,46,55,65].

The third source is diet-derived modulators, and it is notable that most of the compounds found effective along this axis are dietary in origin: phenethyl isothiocyanate, indole-3-carbinol, diallyl disulfide, proanthocyanidins, biochanin A, berberine, taurine, and a Rubisco-derived plant peptide [20,21,23,25,30,36,40,57] (Table 5). Three caveats apply: the studies were conducted almost entirely in animal models and at doses above the range attainable through diet; most reported concurrent weight loss, so direct effects cannot be separated (Section 6.4); and differences in bioavailability and microbiota-mediated metabolism may alter efficacy [58].

10.7. Implications for Clinical Practice

As far as the current evidence allows, the clinical implications can be grouped under four points.

The first concerns the interpretation of hyperleptinaemia: a high leptin concentration in a patient with obesity reflects a defective response rather than an excess of hormone, and leptin measurement does not currently function as a decision tool in routine practice.

The second concerns diet composition: although most of the compounds of interest are dietary in origin, the findings remain at the level of animal models and do not support supplementation. They provide mechanistic support only for existing advice to reduce saturated fat and excess caloric load.

The third concerns the nature of weight loss: surgical data suggest that the same amount of weight loss may have different effects on the central leptin response depending on how it is achieved, so that the method as well as the magnitude of loss deserves attention.

The fourth concerns the prospect of phenotype-based stratification: adiposity-associated DNA methylation signatures and circulating microRNAs [63,64] may represent measurable molecular correlates of individual differences in response. These findings are correlational and unvalidated, however, and provide no basis for any clinical decision at present.

The clinical value of this review lies not in offering a treatment recommendation but in explaining mechanistically why current treatments do and do not work, and in defining a concrete research question ready for testing: can agents targeting ER proteostasis restore leptin sensitivity in humans?

11. Conclusions

This review has examined leptin resistance at the intersection of intracellular signalling networks and endoplasmic reticulum stress, synthesising 22 core primary studies together with the supporting literature. The aim was not to restate known mechanisms but to unify them under a single testable framework and to explicitly set out the points at which that framework could be falsified.

11.1. Principal Conclusions of the Synthesis

Three principal conclusions emerge. First, the ER appears to shape leptin signalling at the levels of receptor biogenesis, receptor bioavailability, and post-receptor inhibition; that the first of these layers operates without PERK activation and rests on a conditional genetic design makes it the best-supported claim in this review, although it rests on a single study whose risk-of-bias profile is set out in Section 11.2. Second, distinct stressors appear to recruit a narrow set of effectors, although convergence refers to the identity of these molecules rather than to a uniform direction of change: CHOP acts as a stress-induced inhibitor in the hypothalamus yet sustains basal sOb-R production in the liver, and leptin-induced SOCS3 marks intact feedback whereas a chronically raised basal level accompanies resistance. The causal role of CHOP is supported by genetic evidence, while the evidence for SOCS3 and PTP1B remains correlative. Third, this node appears pharmacologically accessible in experimental models, yet direct effects have been separated from weight-loss-dependent effects in only one weight-matched study [28].

11.2. Limitations of the Evidence Base

All of these conclusions must be read against a substantial constraint: seventeen of the twenty-two studies in the core evidence base rest on rodent in vivo models and five on cell culture alone, and no study has been conducted in human participants. Studies using the human-derived SH-SY5Y line provide data from a human cell line, not human data.

A second constraint concerns the internal validity of those studies. Applying SYRCLE’s tool to the seventeen animal studies produced 136 of 170 item-level judgements of unclear risk, 26 of high risk and 8 of low risk (Tables S2 and S3). No study described the method by which an allocation sequence was generated, reported allocation concealment, or reported random housing; only one study made an explicit statement about blinding, and that statement was that the investigators were not blinded [24]. The high-risk judgements cluster in two items. The first is incomplete outcome data, where eight studies report group sizes that fall between allocation and analysis without explanation or that differ between the methods and the results. The second is other sources of bias, where twelve studies carry a concern specific to the design, most often an analysis unit that does not correspond to the unit of allocation. These findings do not show that the reported effects are false. They show that, for most items in most studies, risk of bias cannot be determined from what has been published, which is a different and more constraining position than a study judged to be at low or high risk.

The point applies with most force to the findings on which this review leans hardest. The SEL1L–HRD1 result that supports the proteostasis–stress distinction comes from a study whose authors state that the investigators were not blinded and in which some comparisons rest on one or two animals per group [24]. The CHOP haploinsufficiency result that provides the causal evidence for the effector node comes from a study whose group sizes differ between the methods and the results and whose genetic model is described inconsistently [41]. The SOCS3 finding discussed in the section Context-Dependence of the Shared Effectors comes from a study in which islets from two animals were pooled per sample and protein samples were pooled again before immunoblotting, so it carries no biological replication and no estimate of between-animal variance [27]. Each remains the best available evidence for the claim it supports, and none should be treated as settled.

Together, these constraints determine which claims the evidence supports. Animal models support the proposition that ER stress can impair hypothalamic leptin signalling and that the impairment is reversible, and they do so consistently across unrelated models and interventions, but the proposition cannot be called strongly supported because the studies that establish it are, with few exceptions, not reported in enough detail to exclude selection, performance, and detection bias. Consistency across studies whose internal validity is largely unverifiable is weaker evidence than consistency across studies that have been shown to be at low risk of bias. The same evidence base does not support the proposition that targeting ER proteostasis should be used as an obesity treatment in humans; such a proposal is at best a hypothesis worth testing.

The distinction deserves emphasis because the history of the leptin field is itself instructive: recombinant leptin was markedly effective in rodent models yet failed in common human obesity, and there is no basis for assuming that strategies targeting ER stress will avoid a comparable discontinuity.

A further limitation concerns the construction of the evidence base itself. Studies were eligible only if they linked an ER stress or UPR component to leptin signalling, so the search was designed to retrieve evidence of ER involvement and could not retrieve evidence that some other mechanism predominates. The synthesis can therefore show that ER-related mechanisms are reported across a wide range of inputs; it cannot establish that they matter more than the inflammatory, transport-level, or transcriptional mechanisms examined in studies this search did not sample. Node primacy (Hypothesis 2, Table 3) is stated as a hypothesis for precisely this reason, and testing it requires head-to-head comparison within a single model rather than a larger review.

Further limitations are as follows: (i) the great majority of studies used male animals, and sex differences have not been examined systematically, with El-Sehrawy et al. [22] being an exception; (ii) phytochemical doses exceed the range attainable through diet in humans; (iii) effect sizes vary widely across models; (iv) study durations are generally limited to 6–12 weeks; (v) although leptin resistance manifests with different molecular signatures in different tissues, most studies treat it as a single entity; (vi) the search was confined to two databases and did not include citation tracking of the reference lists of the included studies, so relevant reports indexed elsewhere may have been missed; and (vii) the five laboratory-only studies all originate from a single research group, so the in vitro evidence is less independent than the number of reports suggests, and the recurrence of the same markers across them partly reflects a shared methodological tradition.

11.3. Research Gaps

The principal research gaps and the methodological approaches required to address them are summarised in Table 2; the most important is the absence of a randomised controlled trial testing whether an intervention targeting ER stress restores leptin sensitivity in humans.

11.4. Methodological Requirements and Future Directions

Closing these gaps will require several methodological advances: single-cell and spatial transcriptomics could resolve the cell-type-specific UPR activation that tissue homogenates obscure; human iPSC-derived hypothalamic organoids could overcome the limitations of the SH-SY5Y line; multi-omic integration could be used to identify circulating biomarkers; and computational drug discovery could contribute to the design of selective PTP1B and UPR modulators [56].

How the response varies between individuals, meanwhile, is unknown. Studies stratified by genotype, epigenetic profile, microbiota composition, and baseline metabolic state offer the most realistic route to testing these compounds in humans, but this is a research programme that first requires phenotyping tools to be developed.

The field should also adopt two design standards: the use of weight-matched control groups in intervention studies, and the inclusion of both sexes.

11.5. Concluding Assessment

In rodent models, the endoplasmic reticulum appears to act at more than one level of leptin resistance rather than at a single point in a linear chain, and this multi-layered role—extending from receptor quality control to induction of negative regulators—is supported by conditional genetic designs. It does not follow that the ER is the regulatory centre of leptin resistance. The evidence assembled here cannot establish that ER-related mechanisms matter more than the inflammatory, transport-level, or transcriptional mechanisms examined in studies this search was not designed to sample (Section 11.2), and node primacy is retained for that reason as a hypothesis awaiting head-to-head comparison within a single model (Hypothesis 2, Table 3). The endoplasmic reticulum is therefore advanced here as a candidate convergence node, and recasting that observation as a convergence-node framework remains at the level of a hypothesis.

The most important consequence of this multi-layered organisation is that leptin resistance is not a uniform disorder; the phenotypes arising at different levels (Table 8) offer a molecular basis for explaining why the same intervention produces different outcomes in different individuals. Resolving the molecular heterogeneity of leptin resistance may be necessary before interventions aimed at restoring leptin sensitivity can be tested in the appropriate patient subgroups.

The conditional phrasing is deliberate: the evidence base consists largely of animal and cell culture studies, and in humans, neither can leptin resistance phenotypes be distinguished nor can hypothalamic ER stress be measured. The stratification proposed here is therefore not a clinical strategy but a mechanistically justified research programme.

The priority for the field is not to generate further rodent studies but to take three concrete steps: (i) developing and validating circulating biomarkers capable of distinguishing leptin resistance phenotypes in humans; (ii) testing agents already administered in humans, such as TUDCA and 4-PBA, in weight-matched trials with leptin sensitivity as the primary endpoint; and (iii) designing studies in which interventions are stratified by genotype, epigenetic profile, and microbiota composition.