Cancer biochemistry / Combination research

Phenylbutyrate and Natural Products

Doses, animal and human evidence, and the IL-8 question for 20 compounds.

Updated research synthesis · 18 October 2026 · In vivo evidence distinguished from cell culture and computational predictions

Two compounds can each inhibit cancer cells when tested alone and still work poorly together. One may remove a defence that the other needs to overcome. It may also remove the very stress that makes the other compound effective.

Phenylbutyrate makes this problem particularly clear. It can alter gene expression, change fuel use and support nitrogen disposal. It can also help cells cope with misfolded proteins. Whether that last action helps the host depends, in part, on which cells it protects and what the partner compound is trying to do.

The strongest claim should follow the strongest relevant experiment. Give tumour and survival outcomes in living systems more weight than cell markers, then ask whether the same compound, partner, dose and route were actually tested. Mechanistic alignment alone cannot settle a PB combination.

Phenylbutyrate’s mechanisms How doses and evidence are weighted Why the direction changes Compare all 20 compounds What the evidence supports The IL-8 comparison Melatonin reconsidered References

This review separates measured combination effects from pathway changes and mechanistic predictions. It identifies research leads and conflicts; it does not establish a clinical supplement regimen. “No opposing mechanism identified” is a limited observation, not proof of compatibility.

Read the dose together with the experiment

The comparison below puts in vivo evidence first: human studies and animal experiments are kept separate from concentrations applied to cultured cells. The doses are representative regimens verified in the cited sources, not recommended doses or an exhaustive catalogue. Where the primary text does not resolve a dose or route, that gap is stated.

1. For whether a PB pair works: controlled human outcomes would carry most weight, followed by well-controlled animal experiments of that exact pair, then cell-combination experiments. None of the retrieved human studies establishes efficacy for these PB pairs.
2. For an NP’s independent activity: an animal tumour result strengthens its rationale, but cannot establish what happens when PB is added. An animal trial with docetaxel, mitoxantrone, melatonin or chloroquine belongs to that combination.
3. For translation: human pharmacokinetics constrain exposure claims, even when no efficacy endpoint was studied. A single-dose plasma peak, sustained cell exposure and tumour concentration are different quantities.
4. For mechanism: gene expression, secreted protein, enzyme activity and functional rescue answer different questions. Docking ranks below experimental target engagement. A mouse study with missing controls or a misidentified graft remains limited evidence.

mg/day, mg/kg and µM are not interchangeable

mg/day describes an administered amount; mg/kg adjusts it for body mass; µM or mM describes concentration. Absorption, metabolism, protein binding and exposure duration connect these quantities. An injection, intranasal liposome or direct tumour injection cannot be treated as an ordinary oral dose. No human equivalent dose is inferred here from mouse regimens.

For PB itself, an oral phase I study tested 9–45 g/day and reached plasma concentrations above the preclinical 0.5 mM benchmark, but reported no complete or partial responses. Those findings neither establish sustained multi-millimolar tumour exposure nor validate a natural-product partner.^[1]

Part I / Several actions in one molecule

What phenylbutyrate changes

Here, PB or 4-PBA refers to 4-phenylbutyric acid and its phenylbutyrate moiety. Its effects cannot be reduced to a single target. Four actions are especially relevant to the compounds considered below.

01 / Gene regulation

Histone deacetylase inhibition

PB inhibits class I/II histone deacetylases and can change programmes governing differentiation, cell division and apoptosis. The direction is context-dependent: increased acetylation can also favour tumour-promoting signals. A partner acting on another regulatory mechanism might complement it; overlapping activity may add redundancy.^[2]

02 / Protein stress

Chemical chaperoning

PB can reduce endoplasmic-reticulum stress and modify the unfolded protein response. That response can sustain a stressed cancer cell or contribute to its death. Direct experiments with shikonin and luteolin illustrate opposite outcomes when PB is added.^{[3] [4]}

03 / Fuel use

PDK inhibition and active PDH

PB can inhibit pyruvate dehydrogenase kinases, relieving their inhibition of pyruvate dehydrogenase. A cancer-cell study found increased oxygen consumption and reduced extracellular acidification. This makes mitochondrial effects relevant when selecting a partner.^{[5] [6]}

04 / Nitrogen disposal

Phenylacetate–glutamine conjugation

PB is metabolized to phenylacetate, which forms a conjugate with glutamine for renal excretion. This provides an alternative route for nitrogen removal. It does not demonstrate selective glutamine starvation inside a tumour.^[7]

Phenylbutyrate → phenylacetate
Phenylacetate + glutamine → phenylacetylglutamine → urinary nitrogen excretion

The distinction matters for oxidative treatments. Ammonia lowering, glutamine depletion and glutaminase inhibition affect different biological processes. A proposed link from PB to lower tumour glutathione must be measured rather than inferred from its use as a nitrogen scavenger.

Gene regulation can also favour migration or stemness

One PB study linked increased IL-8-promoter acetylation to IL-8–CXCR1/2–ERK signalling and migration. Its reported gastric cell lines, MGC-803 and BGC-823, have HeLa identity problems, limiting gastric-specific conclusions.^{[8] [9] [10]}

Separately, PB promoted stemness and tumour development in hepatocellular-carcinoma cell and mouse models through PPAR-α–Wnt5b–Fzd5–β-catenin. That mechanism should not be automatically attributed to HDAC inhibition. Both findings make counter-regulatory partners worth investigating, without establishing the same effect in patients.^[11]

Part II / The same intervention, different outcomes

Is the stress response a defence or a route to death?

The unfolded protein response is a set of signalling pathways activated when the endoplasmic reticulum struggles to manage its protein load. PERK, IRE1 and ATF6 help organize the response. Its consequences depend on the cell, the intensity of the disturbance and its duration. A rise in an ER-stress marker does not, by itself, tell us whether the response is useful or harmful to the cancer cell.

When stress supports survival

PB can increase killing

The partner stresses the cancer cell.

↓

The cell activates a protective ER response.

↓

PB blunts that adaptation.

The partner becomes more effective.

Example: shikonin in glioblastoma stem-cell models.

When stress drives cell death

PB can reduce killing

The partner overwhelms cellular stress handling.

↓

ER signalling contributes to apoptosis.

↓

PB relieves part of the lethal stress.

More cancer cells may survive.

Example: luteolin in Neuro-2a neuroblastoma cells.

A conceptual comparison grounded in separate experiments, not a universal rule for every tumour. These examples concern the tested cell types and conditions.^{[3] [4]}

Curcumin shows why the tumour model matters

In lung cancer cells, PB enhanced curcumin-associated killing by interfering with an adaptive ER response. The PB experiment appeared in a paper mainly examining fenretinide plus curcumin: its animal results for that main combination should not be assigned to PB.^[12]

In colorectal and thyroid cancer models, the direction was different. PB reduced curcumin-associated effects where ER signalling contributed to death. These observations do not justify calling curcumin universally compatible or incompatible with PB. They make the role of the stress response a question that must be answered in the intended model.^{[13] [14]}

Part III / The 20-compound comparison

All 20 products: doses, living-system evidence and IL-8

Direct functional evidence measures a change in death, survival or tumour growth with the pair. Direct pathway evidence measures a molecular response, which may not settle the final outcome. Inferred interactions are predictions from separately demonstrated mechanisms.

Representative research exposures. In vivo results come first; each column identifies what was actually tested. ↓ = decrease; ↑ = increase; i.p. = intraperitoneal injection.

Compound	In vivo: dose, route and result	In vitro: concentration and model	IL-8 / CXCL8 evidence	Interpretation for PB
ShikoninNaphthoquinone	PB pair, mice: enhanced activity in glioblastoma stem-cell xenografts. Exact pair doses/route were not recoverable from the accessible primary abstract, limiting reproducibility. [3] NP alone, mice: 2.5 or 5 mg/kg intraperitoneally (i.p.), every other day for 24 days reduced A549 xenograft growth. This separate regimen cannot fill the PB-pair dose gap. [15]	Separate lung-cell study: 0.25–1.25 µM for 24 h in A549/H1299 cells. Do not transplant these concentrations into the GBM combination. [15]	Unresolved in the selected cancer evidence. The lung study demonstrated IL-6/STAT3 changes; it discussed IL-8 as a possible mechanism requiring investigation. IL-6 suppression is not an IL-8 result. [15]	EnhancementDirect functional · cells + micePB strengthened shikonin activity in glioblastoma stem-cell models by suppressing protective ER signalling involving JNK/c-Jun. This is the strongest direct positive lead identified in this comparison.[3]
HonokiolBiphenolic neolignan	NP alone, mice: liposomal honokiol 5 mg/kg intranasally, three times weekly, from one week after NNK exposure to week 16, reduced lung tumour burden. This is chemoprevention with a special formulation, not a PB-pair treatment trial. [16]	Lung-cell experiments used 5–7.5 µM over 24–72 h. [16] The separate liposomal-honokiol/PB study supplies pathway-conflict evidence; its precise pair exposure was not recovered here. [17]	↓, inflammatory cells only: 10 µM reduced P. acnes-induced IL-8 in THP-1 cells. A leukaemia-derived line used to model inflammation does not establish tumour control or reversal of PB-induced IL-8. [18]	Pathway conflictDirect pathway · liposomal formulationPB reduced liposomal-honokiol-induced degradation of HSP90 client proteins, potentially preserving EGFR/AKT/c-RAF signalling. Protective autophagy in the same system complicates the net prediction.[17]
LuteolinFlavone	NP formulation comparison, mice: 3.3 mg/kg i.p. every other day, starting seven days before Tu212 implantation and continuing for 30 days. Nanoparticle delivery outperformed free luteolin. No PB arm; pretreatment limits established-tumour inference. [19]	PB pair: reduced killing in cultured Neuro-2a cells; a mouse-derived cell line is still in vitro evidence. Exact pair regimen was not verified from accessible text. [4] IL-8 experiments used 30 µM in gastric cells and 20–80 µM in lung cells. [20] [21]	↑ / concentration-dependent: increased gastric-cell IL-8; lung-cell secretion rose at 20–40 µM. At 80 µM, lower total IL-8 did not translate into lower secretion per surviving cell. Do not rank luteolin as a dependable IL-8 blocker. [20] [21]	Reduced killingDirect functional · cell culturePB reduced luteolin-induced death in Neuro-2a neuroblastoma cells. CHOP knockdown supported a causal role for ER stress in the apoptotic response.[4]
EGCGGallated catechin	NP alone/curcumin pair, mice: colorectal PDX study used 50 mg/kg every other day for four weeks, with separate EGCG, curcumin and combination arms. Tumour growth fell; route is not specified in the accessible methods. No PB arm. [22]	5 µM EGCG, alone or with 50 µM curcumin, in endothelial cells exposed to HT-29-conditioned medium. Reduced JAK/STAT3/IL-8 signalling and angiogenic behaviour. [22]	↓, tumour-conditioned endothelial cells: mRNA and protein measurements support the pathway. The PDX growth result strengthens the overall antitumour rationale, but does not itself prove that IL-8 caused the in vivo response or that EGCG counters PB. [22]	Possible complementarityInferred · epigenetic mechanismEGCG’s experimentally reported DNA methyltransferase inhibition could complement PB’s HDAC inhibition. Removing two transcriptional barriers is a rationale to test, not a demonstrated cancer-combination benefit.[23]
ArtemisininEndoperoxide sesquiterpene lactone	Parent NP alone, mice: 100 mg/kg/day in drinking water, beginning one week after LNCaP implantation, inhibited xenograft growth. This is the authors’ stated administered regimen, not measured tumour exposure. [24]	LNCaP: half-maximal DNA-synthesis inhibition near 25 µM; mechanistic experiments generally used 300 µM for 48 h. Growth arrest and cell killing are different endpoints. [24]	Indirect / dose-resolved endpoint unverified: a parent-artemisinin/derivative study supports NF-κB suppression in H. pylori-associated gastric models, but the accessible abstract does not resolve the parent compound’s IL-8 dose and response. Artesunate/DHA results are not assigned to artemisinin. [25]	Unresolved for the parentParent and derivative evidence differParent artemisinin affects Sp1/CDK4-associated cell-cycle control. PB opposition reported with artesunate raises a family-level concern, but cannot establish the interaction with artemisinin itself.[24] [26]
SilibininFlavonolignan	Human exposure study: 360, 720 or 1,440 mg/day orally for seven days, as phosphatidylcholine-formulated silipide before surgery. Plasma concentrations were approximately 0.3–4 µM; colorectal tissue exposure exceeded liver exposure. This was pharmacokinetics, not clinical tumour-response evidence or a PB trial. [27]	The IL-8 study used 50–200 µM in stimulated HMC-1 mast cells. These concentrations exceed the plasma range in the cited oral study; tissue accumulation does not establish equivalent free intracellular exposure. [28] [27]	↓, non-oncology inflammatory model: IL-8 mRNA and secreted protein fell, alongside NF-κB effects. STAT3 anticancer evidence offers a separate rationale, but neither establishes in vivo tumour IL-8 suppression with PB. [28] [29]	Possible complementarityInferred · STAT3 mechanismExperimentally supported STAT3 inhibition could complement PB-dependent transcriptional changes. No specific conflict is established on this axis; the exact pair’s benefit remains unproven.[29]
MagnololBiphenolic neolignan	NP alone, mice: 40 or 60 mg/kg by daily oral gavage for 22 days in established orthotopic MOC1 oral tumours. Tumour reduction and immune changes carry more weight than isolated pathway predictions, while leaving the PB pair unresolved. [30]	10 µM in the directly verified IL-8 experiment. [18] The separate HCC ER/CHOP work supports a possible stress-relief conflict with PB, not a measured binary-combination outcome. [31]	↓, inflammatory cells only: reduced P. acnes-induced secretion in THP-1 cells. The oral mouse cancer study does not establish human CXCL8 suppression; its tumour/immune endpoints should retain their own labels. [18] [30]	Possible loss of activityInferred · ER/CHOP mechanismMagnolol-induced apoptosis in a hepatocellular-cancer model depended partly on ROS and CHOP. PB could weaken that lethal-stress component, but a binary combination outcome is not established by those experiments.[31]
AndrographolideDiterpene lactone; Andrographis constituent	NP alone/melatonin pair, mice: methods report 20 mg/kg i.p., five times weekly for two weeks; the melatonin arm used 40 mg/kg on the same schedule. This supports in vivo activity but tests a different partner from PB. [32]	T84 cells: 45 µM for 48 h was the selected growth-inhibitory exposure. PB reduced ER-response markers in this cellular work. [33] A separate IL-8 study used 20 µM in TNF-α-stimulated HCT116 cells. [34]	↓, cancer cells: reduced IL-8 transcription/protein with ROS, NF-κB and MAPK/AP-1 pathway involvement. This is direct cellular IL-8 evidence, not a demonstrated in vivo PB rescue strategy. [34]	Pathway conflictDirect pathway · colon cancer cellsPB reduced GRP78, IRE1 and BAX responses associated with andrographolide. This interferes with a relevant apoptotic pathway without proving that every anticancer action is lost.[33]
CurcuminPolyphenolic diarylheptanoid	NP alone/docetaxel pair, mice: ovarian dose-finding selected 500 mg/kg by oral gavage every 24 h; tumour IL-8 and other angiogenic signals fell. Growth effects varied by model and were stronger with docetaxel. No PB arm. [35]	Tumour-conditioned endothelial study: 50 µM curcumin with/without 5 µM EGCG. [22] The separate PB-pair literature reports opposite effects in lung versus colorectal/thyroid cells; its doses are not replaced by this unrelated regimen. [12] [13] [14]	↓, including tumour tissue in mice: one of the better in vivo IL-8 leads in this set. The high oral mouse dose and absence of a PB arm prevent a claim that oral human curcumin offsets PB-induced IL-8. [35]	Both directions reportedDirect functional · different cell modelsPB enhanced killing in lung cancer cells but reduced curcumin activity in colorectal and thyroid models. Protective versus lethal ER signalling helps explain the difference.[12] [13] [14]
Gallic acidTrihydroxybenzoic acid	NP alone, mice: 0.3% or 1% w/v in drinking water reduced prostate xenograft growth; the 22Rv1 experiment lasted six weeks. These are 3 or 10 mg/mL in water, not fixed mg/kg doses without intake/body-weight data. No PB arm. [36]	HepG2 IL-8 study: 550 µM for 48 h, with approximately 56% loss of viability. [37] Separate oral-cancer ER/calcium experiments do not establish PB neutrality. [38]	↓, cancer cells at a strongly cytotoxic exposure: reduced IL-8-associated angiogenic signalling. Loss of viable producer cells complicates interpreting this as selective IL-8 inhibition. No matched animal IL-8/PB result was verified. [37]	Conditional compatibilityInferred · calcium/ROS mechanismER calcium release and mitochondrial ROS contributed to death, but PERK knockdown did not prevent it. Independence from one UPR branch does not demonstrate neutrality of PB, which has several actions.[38]
BaicaleinTrihydroxyflavone	NP alone, mice: 50 mg/kg/day by oral gavage in lean/obese 4T1-bearing mice, from 50–100 mm³ tumours until the humane endpoint. Leptin/immune and tumour effects were studied; no PB arm. [39]	HeLa NF-κB work tested 20–100 µM; the IL-8 protocol used 12 h pretreatment followed by 12 h TNF-α exposure. [40] Ferroptosis inhibition is a separate functional concern. [41]	↓, cancer-cell transcription: IL-8 mRNA fell in TNF-α-stimulated HeLa cells. This is not a measurement of secreted IL-8 or proof of an in vivo CXCL8 effect. Baicalein must be distinguished from baicalin. [40]	Ferroptosis concernFunctional evidence for baicalein; PB pair unresolvedBaicalein protected experimental cancer cells from erastin-associated ferroptosis. It could oppose a lipid-peroxidation death strategy without necessarily opposing every epigenetic or metabolic effect of PB.[41]
BerberineIsoquinolinium alkaloid	Human prevention trial: 300 mg orally twice daily after polypectomy reduced adenoma recurrence during follow-up to two years. This deserves weight for that prevention endpoint; it does not show treatment of established cancer, tumour IL-8 target engagement or benefit with PB. [42]	15 µM berberine with 45 µM evodiamine was a reported AGS combination regimen. [43] Separate TNBC experiments measured IL-8 mRNA/protein and invasion; their exact exposure was not recovered from the abstract. [44]	↓, cancer cells: EGFR/MEK/ERK-associated suppression in TNBC; also counteracted evodiamine-induced IL-8 and endothelial adhesion in AGS cells. That direct rescue belongs to berberine–evodiamine, not berberine–PB. [44] [43]	Metabolic conflict; net effect uncertainInferred from respiration effectsComplex I-associated respiration inhibition may counter PB-driven mitochondrial oxidation—or create a lethal energy bottleneck. Opposite metabolic effects do not determine the combined survival outcome.[45]
ApigeninFlavone	NP/chemotherapy, mice: 10 mg/kg i.p., seven days after each mitoxantrone/vehicle administration in weeks 3, 5 and 7; evaluation at week 8. Human tumour/stroma grafts support a senescent-stroma rationale. No PB arm. [46]	10 µM suppressed the senescence-associated secretory phenotype in human stromal cells. [46] A separate positive PB experiment used problematic SGC-7901 cells; the identity issue lowers its disease-specific weight. [47] [48]	↓, tumour-relevant stroma: CXCL8 transcription/protein and secretome results support suppression of the broader secretory programme. Mouse chemotherapy benefit strengthens the stromal rationale; it does not isolate IL-8 as its sole cause or validate PB cotreatment. [46]	Preliminary enhancementDirect functional · problematic cell modelPB increased apoptosis in a study implicating protective ATF6/PERK signalling. Its SGC-7901 line is listed as HeLa-derived, limiting the claimed gastric-cancer interpretation.[47] [48]
ChrysinDihydroxyflavone	Human pharmacokinetics: a single 400 mg oral dose in seven volunteers produced parent-chrysin peaks of 3–16 ng/mL (≈0.012–0.063 µM). This exposure finding weighs against directly transferring micromolar culture results to ordinary oral dosing. No anticancer or PB efficacy endpoint. [49]	HeLa protein-array experiments used 10–15 µM; approximately 160–1,250 times the cited parent-plasma peak range. This calculated comparison is not a tumour-concentration measurement. [50] [49]	↓, cell-associated protein: CXCL8 fell on the protein array. Secreted IL-8 function, in vivo tumour target engagement and PB benefit remain unresolved. Epigenetic overlap alone does not establish synergy. [50]	Overlapping mechanismsInferred · acetylation/p21 mechanismChrysin affected HDAC activity, histone acetylation and p21-associated transcription. These actions align with PB, but shared mechanisms can be redundant rather than synergistic.[51]
AllicinThiol-reactive thiosulfinate	NP alone, mice: neuroblastoma PDX received 0.05 or 0.2 mg per intratumoral injection on days 0 and 7; 0.2 mg reduced growth. A 0.5 mg initial dose caused excessive bleeding and was reduced. This local-delivery result does not establish oral efficacy. [52]	Patient-derived oral-cancer/CD133+ cultures: 20–100 ng/mL for 48 h, as reported (≈0.12–0.62 µM). [53] Separate neuroblastoma assays had 19–50 µM growth IC50 values at 24 h; these are different experiments. [52]	↓, patient-derived cultures: IL-8 mRNA and secreted protein declined. The experiment was performed on removed cells, not by treating patients; the separate PDX experiment does not verify an IL-8 mechanism. [53]	Conditional complementarityInferred · glutathione/thiol mechanismAllicin perturbs thiol biology. Synergy would be plausible if PB actually limited tumour glutathione supply; that premise needs measurement. PB-mediated stress relief could also work against killing.[54]
PiperlongumineElectrophilic alkaloid	PB pair, mice: PL 10 or 20 mg/kg + PB 20 mg/kg, i.p. every other day for 21 days, in LLC-bearing mice. PB weakened tumour control and macrophage/ER effects. The design omitted PB alone, so this is an opposition signal, not a complete factorial interaction test. [55]	The lung paper contains inconsistent concentration units, so no clean cell-pair dose is assigned. [55] The separate IL-8 paper also differs between its cytokine methods (1 µM PL) and general treatment description (10 µM). [56]	↑ transiently, endothelial cells: LPS-stimulated HUVECs showed increased IL-8 at 24 h; the difference was not significant at 72 h. These are endothelial, not tumour-cell, data and do not support classifying PL as an IL-8 suppressor. [56]	Reduced antitumour activityDirect functional · cells + micePB weakened reported activity and reversed relevant ER/macrophage-polarization responses in a lung-cancer study. Inconsistent concentration notation in the methods limits confidence in precise replication conditions.[55]
EvodiamineIndole-containing alkaloid	NP alone, mice: 5 or 10 mg/kg i.p. daily for 14 days in LLC tumours reduced growth. The animal arm did not include PB; it cannot establish in vivo PB antagonism. [57]	Selected exposures: 11.22/22.44 µM for 24 h in A549 and 3.43/6.86 µM for 48 h in LLC cells; PB attenuated apoptosis/ER signalling in the cellular experiments. [57]	↑, gastric cancer cells: evodiamine increased AGS IL-8 secretion and adhesion to HUVECs. The study’s 45 µM evodiamine / 15 µM berberine combination counteracted this response. Antiproliferative activity can coexist with a potentially adverse cytokine response. [43]	Reduced apoptosisDirect functional · lung cancer cellsPB reduced evodiamine-induced apoptosis and ER-associated signalling. The paper’s separate animal work with evodiamine should not be treated as an animal test of the PB pair.[57]
MyricetinHighly hydroxylated flavonol	NP alone/chloroquine pair, mice: 30 mg/kg i.p. daily for 15 days in an orthotopic graft model; chloroquine was 60 mg/kg. No PB arm. The SMMC-7721 graft is HeLa-derived, substantially weakening the claimed HCC-specific conclusion. [58] [59]	25–200 µM over 12–48 h; many ER/autophagy experiments used 100 µM for 24 h. Hep3B and SMMC-7721 should not be treated as interchangeable authenticated HCC models. [58] [59]	↓, non-oncology cells: myricetin at 100 µM suppressed inflammatory cytokine release, including IL-8, from human mast cells. This does not establish tumour IL-8 suppression or the net PB interaction. [60]	Competing effects possibleInferred · apoptosis and protective autophagyMyricetin induced CHOP-associated death and protective autophagy in the same experimental system. PB might suppress either or both; the net effect cannot be read from ER markers alone.[58]
MelatoninIndoleamine	Human research context: 15–20 mg/day orally, with 20 mg nightly best documented. The 709-patient AMPLCaRe trial found no overall disease-free-survival benefit. [61] [62] Breast stromal work includes animals, but its exact dose/route was not recoverable here. No PB benefit was tested. [63]	GBM SIRT1 paper: 0.25–1 mM, about 1,100–4,300× a cited 20 mg oral mean plasma peak. AGS ER/PB experiment: 2 mM melatonin. These do not resolve activity at 15–20 mg/day. [64] [65] [66]	↓ in breast fibroblast models; mixed in HCC. Breast NF-κB/IL-8 evidence is more direct than separate colon/glioma STAT3 or NF-κB findings. HCC-cell IL-8 moved in opposite directions at 1 and 100 µM. See the detailed section below. [63] [67]	Context-dependent; plausible complementarityInferred benefit · direct ER-pathway conflictNF-κB/IL-8 and STAT3 suppression, plus SIRT1 inhibition in selected tumours, could complement PB. Effects vary by model; a high-exposure experiment also shows ER-pathway opposition. Exact-pair benefit is unproven. Human research context: 15–20 mg/day orally, with 20 mg nightly better documented. Read the mechanisms and dose evidence below.
Withaferin AElectrophilic steroidal lactone	NP/chemotherapy, mice: Panc-1 grafts received 4 mg/kg i.p. every other day for 24 days. Monotherapy tumour-volume effect was marginal (p=0.052); combinations with epirubicin or cisplatin were stronger. No PB arm. [68]	Purified withaferin A at 10–500 nM (0.01–0.5 µM) for 24 h in the H. pylori/AGS IL-8 protocol; 500 nM was used for signalling work. This is not a dose of ashwagandha extract. [69]	↓, infection-stimulated gastric cells: reduced secreted IL-8 and NF-κB activation. This relatively low culture exposure merits attention, but human tumour exposure and PB cotreatment remain untested in this evidence. [69]	Possible loss of activityInferred · protein-stress mechanismInterference with protein disposal and ER/CHOP-associated apoptosis can contribute to withaferin A activity. PB could ease that stress; the exact pair’s net outcome remains unestablished.[70] [68]

Colour identifies the reported or proposed direction, not clinical suitability. “Enhancement” does not necessarily mean a formal synergy analysis was performed. Unresolved pairs are not established as neutral.

Part IV / Reading the gaps carefully

Which combinations deserve further investigation?

Exact-pair animal evidence changes the priority: shikonin remains the clearest positive replication lead in the GBM context already studied, although the inaccessible dose details limit reproducibility. Piperlongumine provides an in vivo opposition signal, tempered by the missing PB-only group. Cell-only findings for luteolin, evodiamine, curcumin and apigenin receive less translational weight. An animal experiment of the NP alone does not upgrade its PB interaction to animal evidence.^{[3] [55]}

For an IL-8-focused follow-up, curcumin has direct tumour-tissue evidence in mice; apigenin has a stronger tumour-stroma rationale than its problematic cell-line PB result alone suggests. EGCG has relevant endothelial IL-8 evidence alongside a separate PDX result. These are reasons to test PB combinations, not established synergy. Silibinin’s human tissue-exposure study and melatonin’s human trials are more informative about translation than pathway predictions, but do not establish PB-pair efficacy. Chrysin’s low oral parent exposure lowers confidence in direct transfer of its micromolar cell findings.^{[35] [46] [22] [27] [62] [49]}

“More killing” and “synergy” are different claims

A combination can outperform either single agent simply because two active effects have been added together. Formal synergy requires comparison with an explicit additive reference across informative concentrations. Conversely, a PB rescue experiment shows opposition under its tested conditions; it does not map every possible dose or schedule.

Evidence from a nearby compound can also mislead. Artemisinin, artesunate and dihydroartemisinin are not interchangeable. Neither are baicalein and baicalin, silibinin and a whole silymarin mixture, or allicin and other garlic sulfur compounds. A result with sodium butyrate should not be relabelled as a phenylbutyrate result.

One recent berberine study used PB as an ER-stress probe while investigating berberine plus 5-FU. Its accessible abstract did not settle the net PB–berberine interaction. The reported 5-FU combination benefit belongs to that pair, not to PB.^[71]

IL-8: a useful question, not a universal explanation

PB can induce IL-8-associated migration in particular cell models, but a compound’s general anti-inflammatory reputation does not show that it blocks that response. The table distinguishes direct CXCL8 measurements from NF-κB/STAT3 changes, cancer cells from stroma or inflammatory cells, and cytokine transcription from secretion. A fall in total cytokine after extensive cell death is weaker evidence of selective pathway inhibition.

• Higher-priority in vivo leads: curcumin reduced IL-8 staining in ovarian tumour tissue; apigenin’s cellular CXCL8/SASP work was accompanied by mouse evidence for a chemotherapy/stroma strategy. Neither tested PB-induced IL-8 rescue.^{[35] [46]}
• Direct cellular leads: EGCG, andrographolide, berberine, baicalein, chrysin, allicin and withaferin A have measured IL-8/CXCL8 effects in the specified cancer or tumour-associated models. Their exposure, stimulus and endpoint differ. Melatonin has direct breast-fibroblast evidence but conflicting HCC-cell results; gallic acid’s high cytotoxic exposure requires particular caution.
• Counterexamples: luteolin and evodiamine increased IL-8 in some cancer-cell settings. Piperlongumine produced a transient increase in endothelial cells. These findings cannot be erased by citing general NF-κB inhibition elsewhere.^{[20] [21] [43] [56]}
• Indirect or lower-relevance evidence: honokiol, magnolol, silibinin and myricetin have inflammatory-cell IL-8 findings; parent artemisinin’s dose-resolved endpoint and shikonin’s cancer IL-8 mechanism remain unresolved in the retrieved evidence. “Not verified” is an evidence gap, not proof of no effect.

An animal cytokine result needs a species label

Mice and rats lack a direct orthologue of human CXCL8. Mouse Cxcl1/Cxcl2 findings may support a related chemokine mechanism, but should not be relabelled as human IL-8 measurements. Human cells in a mouse xenograft can produce human CXCL8; the assay must identify the source and analyte. This distinction matters when weighing an “in vivo IL-8” claim.^[72]

The decisive experiment would first show that PB raises secreted human CXCL8 at a relevant exposure, then test whether the NP reverses it without merely removing viable producer cells. IL-8 add-back, CXCR1/2 blockade or genetic perturbation should connect the cytokine change to migration, stemness or durable tumour control. In vivo confirmation should include both single agents and the combination. No retrieved study established this complete chain for one of the PB pairs.

Melatonin: a broader, context-dependent partner

Melatonin deserves more than a label based on one ER-stress experiment. Its effects on inflammatory transcription, stem-like behaviour and selected deacetylase pathways offer plausible ways to complement PB. The evidence supports testing that proposition in defined tumours.

IL-8, NF-κB and STAT3: distinguish the measured pathways

• Breast cancer: melatonin reduced cancer-associated fibroblast IL-8 through NF-κB suppression and opposed AKT-associated epithelial–mesenchymal transition in cell/animal work. This supports an IL-8 rationale; prevention of PB-induced IL-8 was not tested.^[63]
• Colon cancer: cell and xenograft experiments linked enhancement of 5-FU activity to PI3K/AKT and NF-κB/iNOS suppression. That is evidence for those pathways and that drug pair, rather than proof of IL-8 suppression with PB.^[73]
• Glioblastoma: melatonin reduced self-renewal and proliferation in patient-derived stem-like cells through AKT–EZH2–STAT3 signalling. This supports an anti-stemness hypothesis without establishing the same IL-8 mechanism.^[74]
• Liver cancer: STAT3/HIF-1α/VEGF suppression was reported in hypoxia-modelled HepG2 cells. But IL-8 gene expression increased in HCC24/KMUH and decreased in HCC38/KMUH cells in another study. A universal IL-8 inhibition claim would miss this counterexample.^{[75] [67]}

SIRT1: complementary target coverage is plausible

SIRT1 is a class III, NAD⁺-dependent deacetylase, distinct from PB’s conventional class I/II HDAC targets. PB’s HDAC classification does not establish SIRT1 inhibition or equal activity against every class I/II isoform.^{[2] [76]}

In prostate-cancer cells and TRAMP mice, melatonin suppressed SIRT1; forced SIRT1 expression partially rescued cell proliferation. The cellular evidence includes 10 nM–10 µM exposure for 14 days, as well as acute millimolar experiments. In osteosarcoma cells, SIRT1 knockdown or sirtinol enhanced melatonin’s effects, while the activator SRT1720 weakened them. These are substantive reasons to investigate complementary deacetylase targeting.^{[77] [78]}

Adjacent evidence strengthens the idea: SIRT1 overexpression counteracted H4K16 acetylation and etoposide sensitization produced by trichostatin A or valproate. That experiment supports a possible compensating mechanism, but did not test PB or melatonin.^[79]

GBM: the SIRT1 route remains unresolved at oral-dose exposure

The GBM paper reported increased SIRT1 protein abundance in U251/U87 cells after 0.25–1 mM melatonin. It did not directly measure SIRT1 enzymatic activation or establish that SIRT1 was necessary for melatonin’s effects. Adhesion-marker findings do not establish tumour control.^[64]

That is 250–1,000 µM, approximately 1,100–4,300 times the 0.23 µM mean plasma peak in the cited oral 20 mg study: about 3.0–3.6 orders of magnitude. This is a plasma comparison, not a tumour-exposure measurement. Neither activation nor inhibition of tumour SIRT1 in GBM is established at 15–20 mg/day by these data. The paper carries little weight for either direction at oral-dose exposure.^[65]

A separate breast-cancer model reported increased SIRT1 alongside reduced STAT3 acetylation and restored paclitaxel sensitivity. That remains context-specific evidence against a universal tumour-versus-normal rule; it does not resolve SIRT1’s direction in GBM.^[80]

The defensible hypothesis is complementarity in SIRT1-dependent tumours where melatonin suppresses SIRT1. Calling it “the escape route that melatonin closes” would require demonstrating PB resistance through SIRT1 and reversal by the combination. Different substrates, including p53 and STAT3, can make deacetylation beneficial or harmful.

Retain the direct ER-pathway conflict

In AGS gastric-cancer cells, PB reduced ER-stress, autophagy and BAX-associated responses to 2 mM melatonin. This is a real pathway-level conflict at the tested exposure, without establishing the net combination outcome across tumours or oral doses.^[66]

Human research dose context: 15–20 mg/day

15–20 mg/day orally, generally in the evening, is the adjunctive oncology-research range used here. 20 mg nightly is much better documented: a Cochrane review catalogued 20 mg/day in 18 studies and 15 mg/night in one small lung-cancer study. The 15 mg figure comes from that review’s trial summary.^[61]

This describes studied doses, not an established effective anticancer dose. A small advanced-lung-cancer trial reported an evening-dose survival signal; the larger, 709-patient AMPLCaRe trial found no overall disease-free-survival benefit after lung-cancer resection. Its advanced-stage subgroup signal needs confirmation. Neither trial tested PB–melatonin.^{[81] [62]}

A single oral 20 mg dose produced a mean peak near 0.23 µM in one healthy-volunteer study, far below 2 mM. Plasma peaks do not measure tumour exposure or reproduce prolonged lower-concentration experiments. Human tumour SIRT1 suppression at 15–20 mg/day remains unestablished.^[65]

Updated assessment: plausible complementarity with context-dependent opposition. No retrieved experiment demonstrated that melatonin prevents PB-induced IL-8/Wnt signalling or SIRT1-mediated PB resistance. The next step is an exact-pair experiment measuring migration, stemness, durable survival and the direction of SIRT1 activity.

Ferroptosis and immunity add another layer

Ferroptosis is a form of cell death associated with iron-dependent lipid peroxidation. It is tempting to connect PB-mediated nitrogen disposal with glutathione depletion and then assume stronger oxidative killing. The available biology does not support that shortcut.

A death programme can be suppressed

Baicalein inhibited ferroptosis in experimental cancer cells. PB itself also suppressed lipid peroxidation and cell death in a separate brucine-treated glioma system. Neither result proves every combination will fail, but both challenge a blanket prediction of ferroptosis sensitization.^{[41] [82]}

The immune compartment can respond differently

Experimental colorectal-cancer research linked ammonia to T-cell dysfunction and investigated ammonia-directed interventions including glycerol phenylbutyrate. That offers an immune rationale, but does not validate any natural-product pair in this article.^[83]

Protecting a T cell and protecting a cancer cell are different outcomes. A tumour-only culture cannot establish the immune benefit of systemic ammonia disposal, while a reduction in bulk tumour oxidative markers cannot reveal which cell population was protected.

Part V / From a mechanism to a useful result

Exposure and experimental design decide what a finding means

Some PB rescue experiments use millimolar concentrations, alongside high experimental exposures to the natural product. A historical phase I study in refractory solid tumours reached approximately 0.5 mM PB plasma concentrations but reported no complete or partial responses. It does not establish sustained multi-millimolar tumour exposure or validate these pairs.^[1]

Formulation and route also matter. Sodium and glycerol phenylbutyrate have different pharmacokinetics. A liposomal-honokiol experiment does not automatically describe ordinary honokiol, and an allicin neuroblastoma mouse study using intratumoral injection does not establish the effect of oral allicin.^{[84] [52]}

The most informative next studies would ask four connected questions:

1. Does the pair change durable tumour-cell survival? Compare both single agents with the combination, and distinguish a metabolic assay signal from actual death or longer-term regrowth.
2. Which PB action is responsible? Separate acetylation, PDH/respiration and ER-stress effects using independent pathway measurements and perturbations.
3. Does timing change the result? Pretreatment, simultaneous exposure and sequential treatment may alter whether adaptation or terminal stress dominates.
4. Which cells and compartments benefit? Measure tumour and immune effects separately. If nitrogen disposal is the proposed mechanism, use a system that can represent it and measure the relevant metabolites.

Computational docking and network analyses can nominate targets. In this literature, they contribute to hypotheses involving PB–PDK, silibinin–STAT3 and EGCG–DNMT interactions. Their value increases when supported by biochemical or cellular measurements; they cannot establish the direction of a two-drug interaction on their own.^{[5] [29] [23]}

What can be concluded now

There are specific positive leads, specific examples of opposition and several plausible but untested pairings. No human efficacy evidence for these exact natural-product–PB combinations was established by this targeted review. The uncertainty is most useful when it points to the mechanism and experiment that could resolve it.

The useful question is whether phenylbutyrate removes a defence the cancer needs—or relieves a stress the partner needs to kill it. The answer belongs to a particular compound, tumour and exposure.

References and evidence notes

Primary studies are linked below, with their role in the argument identified. The Cochrane evidence synthesis, drug information, enzyme classification and cell-line registries are labelled separately. Some conclusions rely on accessible abstracts or indexed article sections; unresolved endpoints are kept explicit.

1. Oral sodium phenylbutyrate in refractory solid tumours: phase I study (2001)Early human exposure and response evidence; no validation of these combinations. ↩ Back to text
2. Sodium phenylbutyrate — NCI Drug DictionaryAuthoritative drug-mechanism description. ↩ Back to text
3. Liu et al. Enhanced antitumor effect of shikonin by inhibiting ER stress in glioblastoma stem cells (2015)Exact PB combination; cell and mouse experiments. ↩ Back to text
4. Choi et al. Luteolin induces apoptosis through ER stress and mitochondrial dysfunction in Neuro-2a cells (2011)Exact PB rescue experiment in cultured neuroblastoma cells. ↩ Back to text
5. Differential inhibition of PDKs by phenylbutyrate and enhancement of pyruvate dehydrogenase complex activity (2015)Biochemistry, modelling and experimental validation. ↩ Back to text
6. Phenylbutyrate inhibits PDK1 and contributes to its anticancer effect (2017)Enzyme and cancer-cell experiments. ↩ Back to text
7. BUPHENYL (sodium phenylbutyrate): prescribing informationOfficial information on metabolism and nitrogen disposal. ↩ Back to text
8. 4-Phenylbutyric acid promotes migration through HDAC inhibition-mediated IL-8 upregulation (2020)PB-specific cellular mechanism; gastric-line identity limitations apply. ↩ Back to text
9. Cellosaurus: MGC-803, CVCL_5334Cell-line identity record; reports a HeLa hybrid. ↩ Back to text
10. Cellosaurus: BGC-823, CVCL_3360Cell-line identity record; reports a HeLa-derived line. ↩ Back to text
11. Chen et al. PB promotes hepatocellular carcinoma through PPAR-α and cancer stem cells (2021)Cell and mouse models; PPAR-α–Wnt5b–Fzd5–β-catenin mechanism. ↩ Back to text
12. Chen et al. Synergistic effect of fenretinide and curcumin treatment in non-small-cell lung cancer (2016)PB–curcumin evidence comes from cell experiments within this paper; its principal animal combination used fenretinide. ↩ Back to text
13. Curcumin affects colorectal cancer-cell apoptosis through ATF6-mediated ER stress (2024)Exact PB interaction in a colorectal cancer-cell model. ↩ Back to text
14. Curcumin, ER stress and death in BCPAP thyroid cancer cells (2018)Exact PB interaction in cultured thyroid cancer cells. ↩ Back to text
15. Shikonin, IL-6/STAT3 and lung-cancer progression (2020)Cells and mice; IL-6 was measured. An IL-8 mechanism was proposed, not demonstrated. ↩ Back to text
16. Liposomal honokiol in an NNK-induced lung-tumour prevention model (2017)Intranasal liposomal formulation; prevention model, without PB. ↩ Back to text
17. Liposomal honokiol, HSP90 client-protein degradation and protective autophagy in NSCLC (2017)Direct PB effects on protein degradation; formulation-specific pathway evidence. ↩ Back to text
18. Honokiol and magnolol suppress P. acnes-induced inflammation in THP-1 cells (2005)Direct IL-8 measurement in an inflammatory cell model; not tumour-control evidence. ↩ Back to text
19. Luteolin nanoparticle chemoprevention: in vitro and in vivo evaluation (2014)Free versus nanoparticle luteolin; treatment began before tumour implantation. ↩ Back to text
20. Luteolin enhances IL-8 production in H. pylori-exposed gastric cells (2020)Human gastric cancer-cell culture; increased IL-8, not a universal suppressive effect. ↩ Back to text
21. Luteolin in cisplatin-resistant lung cancer cells (2017)IL-8 secretion and viable-cell normalization reveal a concentration-dependent caveat. ↩ Back to text
22. Curcumin and EGCG suppress colorectal cancer-associated endothelial transition (2017)Cellular JAK/STAT3/IL-8 findings and a separate colorectal PDX experiment; no PB. ↩ Back to text
23. Fang et al. EGCG inhibits DNA methyltransferase and reactivates methylation-silenced genes (2003)Enzyme, modelling and cancer-cell evidence; not a PB cancer-combination trial. ↩ Back to text
24. Artemisinin, Sp1 and CDK4 regulation in prostate cancer (2009)Parent artemisinin; cellular and xenograft work. ↩ Back to text
25. Artemisinin and derivatives in H. pylori-induced gastric carcinogenesis (2019)Primary cell/animal report; accessible abstract verifies NF-kB and prevention findings, not a dose-resolved parent-ART IL-8 endpoint. ↩ Back to text
26. Artesunate induces ER-derived ROS-mediated death by altering iron handling in hepatocellular cancer (2021)Related-compound evidence; supplementary PB rescue result does not establish parent-artemisinin behaviour. ↩ Back to text
27. Hoh et al. Oral silibinin in colorectal cancer patients: tissue exposure (2006)Short presurgical human pharmacokinetic/pharmacodynamic study; not a tumour-response trial. ↩ Back to text
28. Silibinin suppresses inflammatory responses in HMC-1 cells (2013)IL-8 mRNA and secreted protein in a mast-cell inflammatory model, at 50–200 micromolar. ↩ Back to text
29. Verdura et al. Silibinin as a STAT3 inhibitor (2018)Computational and experimental STAT3 studies; no exact PB combination established. ↩ Back to text
30. Magnolol in orthotopic oral squamous cell carcinoma (2024)Oral gavage in immunocompetent mice; tumour and immune outcomes, without PB. ↩ Back to text
31. Magnolol and ER-stress-mediated apoptosis in hepatocellular carcinoma cells (2018)Cellular mechanism; PB interaction is inferred. ↩ Back to text
32. Melatonin and andrographolide in colorectal spheroids and mouse models (2022)Mouse dosing is specified as intraperitoneal in methods; combination partner was melatonin, not PB. ↩ Back to text
33. ER stress and IRE1 signalling cause colon cancer-cell apoptosis in response to andrographolide (2016)Direct PB pathway measurements and complementary genetic experiments. ↩ Back to text
34. Andrographolide inhibits TNF-alpha-induced IL-8 in HCT116 cells (2018)20 micromolar cellular experiment; IL-8 transcription, protein and angiogenesis-related assays. ↩ Back to text
35. Curcumin inhibits growth and angiogenesis in ovarian carcinoma (2007)Oral mouse dose finding and tumour IL-8 measurements; single-agent and docetaxel arms, not PB. ↩ Back to text
36. Gallic acid inhibits prostate carcinoma growth in culture and nude mice (2009)Drinking-water concentrations, not an independently fixed mg/kg dose. ↩ Back to text
37. Gallic acid, IL-8 and angiogenic signalling in HepG2 cells (2016)High-concentration cell experiment with substantial loss of viability. ↩ Back to text
38. Lu et al. ER-dependent calcium, ROS and mitochondrial apoptosis in gallic-acid-treated oral cancer cells (2016)PERK knockdown evidence; not a verified demonstration of PB neutrality. ↩ Back to text
39. Baicalein, adipocyte leptin and antitumour immunity in TNBC (2023)Oral treatment in lean/obese 4T1-bearing mice; no PB or direct human CXCL8 endpoint. ↩ Back to text
40. Baicalein inhibits TNF-alpha-induced NF-kB target genes (2016)HeLa-cell IL-8 mRNA measurement; distinguish transcription from secreted protein. ↩ Back to text
41. Identification of baicalein as a ferroptosis inhibitor by natural-product library screening (2016)Functional protection against ferroptosis in experimental cancer cells. ↩ Back to text
42. Chen et al. Berberine versus placebo for colorectal adenoma recurrence (2020)Randomized human chemoprevention trial after polypectomy; not treatment of established cancer. ↩ Back to text
43. Berberine counteracts evodiamine-induced IL-8 in AGS cells (2013)Direct IL-8 secretion and adhesion findings; berberine–evodiamine, not PB. ↩ Back to text
44. Berberine downregulates IL-8 through EGFR/MEK/ERK in TNBC cells (2018)Cell experiments; the patient-survival analysis is observational and does not show berberine efficacy. ↩ Back to text
45. Berberine induces combined cell death in gastrointestinal cell lines (2023)Respiration and cell-death measurements; does not resolve the PB pair. ↩ Back to text
46. Apigenin targets senescent stroma and improves chemotherapy in mice (2025)10 micromolar cell experiments; sequential mouse treatment after mitoxantrone, without PB. ↩ Back to text
47. Apigenin-induced apoptosis, protective ER stress and ATF6/PERK signalling (2023)Exact PB experiment, with a major cell-line identity limitation. ↩ Back to text
48. Cellosaurus: SGC-7901, CVCL_0520Cell-line identity record; reports a problematic HeLa-derived line. ↩ Back to text
49. Disposition and metabolism of oral chrysin in humans (2001)400 mg single-dose pharmacokinetics in seven volunteers; not an anticancer efficacy study. ↩ Back to text
50. Chrysin modulates epigenetic regulators and cancer-related proteins in HeLa cells (2022)10–15 micromolar cell-lysate protein-array results include CXCL8; not an in vivo IL-8 test. ↩ Back to text
51. Chrysin, histone deacetylase activity and p21-associated chromatin in A375 cells (2012)Enzyme and cancer-cell experiments; no PB pair established. ↩ Back to text
52. In vivo antitumour activity of allicin in a neuroblastoma PDX mouse model (2025)Intratumoral allicin administration; not an oral PB-combination experiment. ↩ Back to text
53. Allicin and IL-8 in patient-derived oral squamous cancer cell cultures (2022)20–100 ng/mL as reported, for 48 hours; patient-derived cells are not an intervention in patients. ↩ Back to text
54. Purified allicin: effects on cancer-cell proliferation and glutathione (2001)Cellular thiol/redox and proliferation findings. ↩ Back to text
55. Zhou et al. Piperlongumine, ER stress and suppression of M2 macrophage polarization in lung cancer (2025)Exact PB interaction in cell/mouse models; methods contain inconsistent concentration notation. ↩ Back to text
56. Piperlongumine and an annexin A1 peptide in tumour and endothelial cells (2020)IL-8 assay was in HUVECs; transient increase and internally inconsistent concentration reporting. ↩ Back to text
57. Evodiamine and ER-stress-mediated apoptosis in non-small-cell lung cancer (2022)PB interaction tested in cells; do not transfer it to the separate animal arm. ↩ Back to text
58. Myricetin induces apoptosis and protective autophagy through ER stress in hepatocellular carcinoma (2022)Mechanistic perturbation of death and protective-autophagy branches; PB effect inferred. ↩ Back to text
59. Cellosaurus: SMMC-7721, CVCL_0534Cell-line identity registry; HeLa-derived, limiting the claimed HCC identity of the myricetin xenograft. ↩ Back to text
60. Flavonols inhibit mediator release in human mast cells (2005)Includes myricetin at 100 micromolar; non-oncology inflammatory evidence. ↩ Back to text
61. Cochrane: Melatonin in cancer treatment (2025)Systematic review, not a primary trial; source for the distribution of studied doses, including the 15 mg trial. ↩ Back to text
62. Seely et al. AMPLCaRe: melatonin after NSCLC resection (2021)709-patient randomized trial, 20 mg nightly; no overall disease-free-survival benefit. ↩ Back to text
63. Melatonin reduces fibroblast IL-8 and reverses EMT in breast-cancer models (2023)Primary cell/animal study; NF-κB–IL-8 and AKT-associated EMT, without PB cotreatment. ↩ Back to text
64. Melatonin modulates the glioblastoma microenvironment through SIRT1 (2019)Limited cell-culture evidence at 0.25–1 mM: SIRT1 protein abundance and adhesion endpoints; oral-dose SIRT1 action remains unresolved. ↩ Back to text
65. Jenjiratthitigarn et al. CYP1A2 induction and melatonin pharmacokinetics (2026)Healthy-volunteer plasma measurements after oral 20 mg; not tumour target engagement. ↩ Back to text
66. Melatonin, apoptosis, autophagy and ER stress in gastric cancer (2019)Direct PB pathway measurements at high experimental melatonin exposure. ↩ Back to text
67. Lin and Chuang: divergent angiogenic chemokine responses to melatonin in HCC cells (2010)IL-8 gene expression increased in one line and decreased in another. ↩ Back to text
68. Withaferin A, proteasome/autophagy disruption and pancreatic cancer (2016)Protein-disposal mechanism relevant to the proposed PB interaction. ↩ Back to text
69. Withaferin A inhibits H. pylori-induced IL-8 in AGS cells (2016)Purified compound at 10–500 nM; an infection-stimulated gastric-cell assay, without PB. ↩ Back to text
70. Withaferin A, ER stress and apoptosis in Caki renal cancer cells (2011)Causal pathway experiments; exact PB pair not established. ↩ Back to text
71. Berberine, ER-stress-mediated autophagic flux and 5-FU sensitivity in colorectal cancer (2025)Uses PB as a mechanistic probe; 5-FU synergy must not be attributed to PB. ↩ Back to text
72. Genome diversification of rodent and lagomorph chemokines (2013)Comparative genomic research: human CXCL8 has no direct mouse/rat orthologue. ↩ Back to text
73. Gao et al. Melatonin enhances 5-FU effects through PI3K/AKT and NF-κB/iNOS suppression (2017)Colon-cancer cells and xenografts; the tested partner was 5-FU. ↩ Back to text
74. Melatonin inhibits glioblastoma stem-like cells through AKT–EZH2–STAT3 (2016)Patient-derived stem-like cell experiments; does not establish an IL-8 or PB mechanism. ↩ Back to text
75. Melatonin inhibits hypoxia-associated STAT3/HIF-1α/VEGF signalling in HepG2 cells (2013)Cancer-cell and endothelial assays; not proof of IL-8 suppression. ↩ Back to text
76. Histone deacetylase families — IUPHAR/BPS Guide to PharmacologyCurated enzyme classification: zinc-dependent HDACs and NAD-dependent sirtuins. ↩ Back to text
77. Jung-Hynes et al. Melatonin, SIRT1 and prostate cancer in culture and TRAMP mice (2011)SIRT1 suppression, rescue experiments and prolonged lower-concentration exposure. ↩ Back to text
78. Cheng et al. SIRT1 inhibition by melatonin in osteosarcoma cells (2013)SIRT1 siRNA/inhibitor enhancement and activator attenuation of melatonin effects. ↩ Back to text
79. Hajji et al. Opposing hMOF/SIRT1 effects on H4K16 acetylation and etoposide sensitivity (2010)SIRT1 counteracted trichostatin A/valproate sensitization; neither PB nor melatonin was tested. ↩ Back to text
80. Circadian melatonin disruption, SIRT1/STAT3/ARHI and paclitaxel resistance in breast cancer (2019)Cell and rat-xenograft models; increased SIRT1 accompanied melatonin-associated chemosensitization. ↩ Back to text
81. Grutsch et al. Evening melatonin in advanced NSCLC (2021)Small randomized trial; adjusted survival signal with evening 20 mg. ↩ Back to text
82. ATF3 contributes to brucine-triggered glioma-cell ferroptosis (2021)Direct example of PB suppressing ferroptotic death; brucine is not one of the 20 screened products. ↩ Back to text
83. Bell et al. Microenvironmental ammonia enhances T-cell exhaustion in colorectal cancer (2023)Experimental tumour/immune mechanisms; not validation of a listed natural-product pair. ↩ Back to text
84. Pharmacokinetics of glycerol phenylbutyrate and sodium phenylbutyrate in urea-cycle disordersHuman formulation and exposure differences. ↩ Back to text

Scope: a targeted synthesis of mechanisms and interaction evidence for 20 named natural products, prepared on 18 October 2026. It includes preclinical work and distinguishes measured outcomes, pathway evidence and inference. It is not a systematic review or a clinical treatment protocol.