D. E. Otzen, Interdisciplinary Nanoscience Centre (iNANO), Centre for Insoluble Protein Structures (inSPIN), Department of Molecular Biology and Genetics, Aarhus University, Gustav Wieds Vej 14, DK – 8000 Aarhus C, Denmark
Thirteen years ago, α-lactalbumin (α-LA) was first reported to form a complex with oleic acid (OA). This complex, called HAMLET (human α-lactalbumin made lethal to tumour cells), was found to be cytotoxic to cancer cells. In HAMLET, α-LA assumes a partially unfolded conformation and can bind OA in various stoichiometries. Subsequently, different groups have been able to prepare HAMLET-like cytotoxic complexes in different ways, which all involve the destabilization of α-LA, and a number of different proteins have been able to form similar complexes. This suggests that the ability to form stable complexes with lipids may be a generic feature of the polypeptide chain, although the precise structural and functional details may vary from protein to protein. We review the biophysical and biochemical properties of this class of complexes, focusing on different methods of preparation, complex structure and the role of the protein and the lipid within these complexes. The cellular effects of these complexes are multifaceted and depend on the cell types. There are strong indications that OA has an essential role, whereas the protein component, rather than having a toxic effect on its own, functions as a vehicle for transporting the toxic OA to the cells and keeping the OA in solution. Fatty acids alone can affect numerous cellular signalling and metabolic pathways, in addition to playing important roles in immune responses and inflammatory processes. Further studies will aim to determine how the molecular properties of the different protein–lipid complexes correlate with their biological efficacy.
lipoprotein complex of equine lysozyme with oleic acid
human α-lactalbumin made lethal to tumour cells
Many of the compounds currently in use for treatment of cancer are not only toxic to cancer cells, but also to rapidly dividing healthy cells. As a result, treatment with chemotherapeutic agents can cause severe adverse effects, both in the short term (e.g. anaemia, mucocitis or neutropaenia), leaving patients at risk of severe infections and possibly lethal sepsis), and in the long term (e.g. chronic heart failure, neuropathy or mutagenic changes), leading to new cancers. Consequently, the search for tumour selective drugs that leave healthy cells unharmed is intense. Initiated by studies on a complex of the milk protein α-lactalbumin (α-LA) and oleic acid (OA) (named HAMLET for human α-lactalbumin made lethal to tumour cells), various protein–lipid complexes with a reported ability to kill tumour cells have been studied with great interest.
The discovery of HAMLET dates from 1995 where Håkansson et al.  found that an entire range of cell types was killed upon treatment with a cytotoxic complex derived from milk: ‘All transformed, embryonic and lymphoid cells tested were killed, while mature epithelial cells were spared’ . The cytotoxicity was initially assumed to arise solely from a multimeric form of the whey protein α-LA. However, the cytotoxic component was later identified as a complex between α-LA and a necessary cofactor (i.e. the fatty acid OA) (C18:1) [2, 3]. Svensson et al.  developed a chromatographic method to prepare this complex de novo from α-LA and oleic acid, and a large number of detailed investigations on the biochemical properties of HAMLET have been reported . In the present review, we use the term HAMLET (and its bovine equivalent BAMLET) specifically to describe complexes between α-LA and OA prepared by this chromatographic method . The discovery of HAMLET has sparked increased interest in the preparation and potentially cytotoxic properties of protein–lipid complexes. Recent observations suggest that neither the preparation method, nor the specific conformation of α-LA, nor even the exact protein sequence, are essential for the cytotoxic effect [5-10]. This highlights the possibility that OA may have an essential role [8, 11, 12], whereas the protein component, rather than having a toxic effect on its own, functions as a vehicle for transporting the toxic OA to the cells and keeping the OA in solution [8, 13].
In this review, we focus on our growing understanding of the biological properties of α.LA:OA complexes from the perspective of protein–lipid complexes in general. We start by describing the conformational properties of α-LA, followed by a description of the preparation of protein–lipid complexes (including other proteins in addition to α-LA) and their interaction with membranes, finishing with a brief overview of the biological effects of these complexes.
Folding properties of α-lactalbumin
α-LA is a small, acidic protein (14.2 kDa; pI 4–5) found in the milk of most mammals. In human and bovine milk, the concentration of α-LA is 2.5 mg·mL−1 and 1.3 mg·mL−1, respectively , making it the second-most common whey (noncasein) protein after β-lactoglobulin. The native protein folds into two major domains: a large α-helical domain and a smaller β-sheet domain connected by a calcium-binding loop (Fig. 1). The two domains are separated by a deep cleft but are held together by two disulfide bridges (amino acids 73–91; amino acids 61–77 in bovine α-LA). Two additional disulfide bridges (amino acids 6–120; amino acids 28–111 in bovine α-LA) contribute to the stabilization of the protein [15, 16].
Studies of complexes between α-LA and OA have been aided by the large body of work available on the folding and stability of α-LA [16-22]. The protein displays a very rich and complex (not to say confusing) folding behaviour. It remains folded at moderate temperatures (below 50–70 °C) between pH 4.2 and 9.5, where it remains monomeric and does not bind to zwitterionic or anionic lipids. The key to its structural versatility lies in the fact that it is stabilized by an exogenous agent: the Ca2+ co-factor. Take away this factor and a Pandora's box of alternate conformations appear. Removal of Ca2+ (either by the chelator EDTA or by a low pH) leads to the apo-form, which is more flexible than the native state and is able to bind anionic lipids , as well as many different classes of detergents , at neutral pH. However, at moderately acidic conditions (pH 4.2–3) and at alkaline conditions (> pH 9.5), α-LA forms the A-state, which is sufficiently ‘sticky’ to have a tendency to aggregate and interact with phospholipid vesicles [25, 26]. This state can also be induced by heat and organic solvents, particularly from the apo-form. The A-state is only partially folded. This is demonstrated by the fact that it contains native-like levels of secondary structure but no persistent tertiary structure, exposes ‘sticky’ patches on its surface that bind hydrophobic dyes such as 1,8-anilino-naphthalene-sulfonate, lacks the ability to unfold cooperatively (i.e. loss of structure in a narrow range of temperatures or denaturant concentrations) and is not as tightly packed as the native state, although not so expanded as the denatured state. Below pH 3, α-LA forms a so-called molten globule , which can also be induced from the apo-state by mild heating, the addition of organic solvents such as trifluoroethanol or the addition of 7.5 equivalents of OA . Interestingly, Ca2+ may also be displaced by polycations such as basic polyamino acids, which can reduce the affinity for Ca2+ by four orders of magnitude, leading to a state similar to the original apo-form . Thus α-LA may adopt a large number of different characteristic conformations .
Preparation of cytotoxic α-LA:OA complexes by different approaches
In retrospect, it may not be unexpected that the addition of an amphiphilic and yet sticky and only sparingly soluble molecule such as OA could have profound effects on the conformational properties of α-LA. Such complexes between α-LA and OA may be prepared in numerous ways. The overall strategy is to destabilize the protein followed by exposure to OA. In the chromatographic approach leading to HAMLET, α-LA is destabilized by removing bound Ca2+ with EDTA, followed by addition to an anion exchange column conditioned with OA and elution from the column by a salt gradient . Remarkably, this exposure to an immobilized hydrophobic molecule converts α-LA to a complex with cytotoxic properties. HAMLET was shown to have very fluctuating structure: it has a near-UV CD spectrum similar to the molten globule , a melting temperature 15 °C lower than the apo-form  and a greater flexibility than the apo- and holo-states of α-LA in terms of hydrogen/deuterium exchange and proteolysis . Proteolysis experiments suggest that contacts between OA and α-LA mainly occur via the α-helical domain , whereas chemical modification studies reveal that α-LA:OA interactions, and thus the ability to form cytotoxic complexes, involve positive amino acid side chains . Equine lysozyme, a homologue to α-LA can form complexes with OA (termed ELOA) by the same chromatographic approach .
A weakness with the chromatographic approach is that the method does not allow us to control the protein–lipid ratio and, indeed, it often leads to complexes with highly variable stoichiometries [7, 30-32, 37]. Initial efforts to produce HAMLET complexes by other approaches such as simple mixing under different solvent conditions were not successful [2, 33, 38]. However, over the last decade, and in particular within the last 3–4 years, several alternative methods to prepare cytotoxic α-LA:OA complexes have appeared (Table 1). In the simplest approach (direct titration), apo-α-LA is mixed at room temperature and neutral pH with OA, which may be ‘activated’ for incorporation into the protein complex by dissolution in ethanol or extrusion to form vesicles of defined sizes. The oligomeric complex may then be separated from free α-LA by gel permeation chromatography [5, 35]. Slightly rougher approaches involve the mixing of α-LA and OA under heat treatment (50–60 °C) [39, 40], acid treatment (pH 4.3)  or alkaline treatment (pH 12)  or the mixing of α-LA and sodium oleate followed by heat treatment (6 °C) [43, 44]. All of these complexes show a significant ability to induce cell death.
Table 1. Different ways to prepare cytotoxic protein–OA complexes
Ratio of OA to protein
Apo-form of protein (made by removing Ca2+ with EDTA) bound to column conditioned with OA and eluted in a salt gradient
α-LA, equine lysozyme
~ 1–5 OA (α-LA), 5–35 (EL; concentration-dependent)
The exact molecular structure of HAMLET and other α-LA:OA complexes has not been determined because attempts to solve the three-dimensional structure of HAMLET by X-ray crystallography have been unsuccessful . This is not unexpected because the α-LA in HAMLET and HAMLET-like complexes has a fluctuating structure with a strongly reduced stability, a slight reduction of secondary structure compared to the native state and a complete loss of tertiary structure. This dynamic conformation is quite incompatible with the regular packing of a crystal lattice. Furthermore, the complexes differ in terms of their size distribution (which may vary from dimer to hexamer and, in some cases, be even larger) . The protein–lipid stoichiometry also depends on the method of preparation, ranging from 1 : 3 to 1 : 48 [5, 9, 46] (Table 1), and may differ by at least a factor two from batch to batch for a given method [37, 47].
In addition, the complexes may interconvert or change to other conformations depending on handling after their initial production. Although HAMLET was first reported to be (partially) multimeric, based on SDS/PAGE analysis , chromatographically prepared HAMLET was subsequently claimed to be mostly monomeric although without reporting the underlying data . The same group has also reported minor amounts of oligomers according to NMR measurements  and highlights the role of oligomerization in allowing HAMLET to form annular pores on anionic phospholipid surfaces . Work conducted by numerous groups has now firmly established that the cytotoxic α-LA:OA complexes are oligomeric [13, 35]. It was specifically shown that the monomeric α-LA left after mixing α-LA with OA is biologically inactive; the cytotoxic species was shown to be multimeric according to gel filtration, dynamic light scattering and glutaraldehyde cross-linking . Heat treatment also increases the degree of aggregation of α-LA . If bovine α-LA is thermally denatured and aggregated (via covalent disulfide cross-links catalyzed by trace amounts of β-lactoglobulin), it is nevertheless still able to form complexes similar to BAMLET (the bovine version of HAMLET) by the original chromatographic treatment .
Importantly, the complex is not only oligomeric, but also dynamic in size and stoichiometry. Although early reports suggested that the different HAMLET preparations could withstand re-injection on a gel permeation column , more recent work has shown that a process as simple as dilution can lead to a reduction in the size of the aggregate . For ELOA, the binding of OA is reversible, and the dilution of ELOA results in less bound OA per equine lysozyme and more free OA (Fig. 2) , whereas irreversible binding of OA is suggested for α-LA:OA complexes prepared by mixing at 17 °C or 45 °C . Linoleic acid does not form an apoptosis-inducing complex with α-LA when prepared by the chromatographic method  but the α-LA–linoleic acid complex formed by the acid treatment method  is cytotoxic, highlighting that different approaches may expand/contract the range of lipids that can be used.
The role of the protein in the formation of cytotoxic protein–fatty acid complexes: a generic feature of the polypeptide chain
The large variation in α-LA:OA lipid stoichiometry, as well as α-LA:OA complex size, highlights an important fact: it may simply be necessary to accept that there are multiple different ways of forming active protein–fatty acid complexes and that the lipids may bind in different ways for different complexes. This versatility and adaptability to different conditions can also be seen as a consequence of the fluctuating structure of these complexes and the lack of rigidly maintained contacts. Thus, there is not one HAMLET complex of one defined size and stoichiometry, nor is α-LA the only protein that is able to form cytotoxic complexes with OA or other fatty acids.
Although the mechanisms are without doubt different, proteins involved in neurodegenerative diseases can also form complexes with long-chain unsaturated fatty acids such as arachidonic acid and OA. For example, mutations in superoxide dismutase can lead to amyotrophic lateral sclerosis. The more destabilized the mutants of superoxide dismutase are, the more easily they form cytotoxic granular aggregates with arachidonic acid and OA . A protein involved in Parkinson's disease, α-synuclein, forms high-molecular weight oligomers with unsaturated fatty acids in transgenic mouse brains , which are down-regulated when neurones are exposed to saturated fatty acids , whereas assembly of the Alzheimer-associated protein tau is induced by fatty acids . All of these proteins can form complexes with fatty acids that are toxic to neurones.
These phenomena are most likely part of a more general mechanism by which amphiphilic molecules (i.e. detergents as well as fatty acids) can induce protein aggregation by forming shared micelles stabilized by the hydrophobic moieties of the sequestered amphiphiles . These compounds adopt a characteristic core-shell structure, in which (hydrophilic) proteins decorate the rims of micelles  (Fig. 3). Structural and biophysical properties of protein–surfactant and protein–fatty acid complexes have recently been extensively reviewed . It remains to be determined whether this type of structure is a general feature of HAMLET complexes. However, it is an alluringly simple model for protein–lipid complexes in which the protein primarily acts as a protective shield to keep the OA fraction solubilized and ‘packaged’ for subsequent targeting to appropriate cellular destinations.
There are no stringent demands on the sequence, structure or integrity of α-LA in the formation of α-LA:OA complexes. α-LA from other organisms (cow, horse, pig and goat) , as well as the α-LA homologue equine lysozyme , can also form HAMLET-like complexes despite sequence variability, although only protein from human milk was shown to form HAMLET naturally, probably because of the higher OA content in human milk . Not unexpectedly, the removal of the ability of α-LA to bind Ca2+ does not adversely affect HAMLET formation . More dramatically, if the eight Cys residues of α-LA are replaced with Ala, leading to a molten globule state under physiological conditions, the overall yield of HAMLET, as well as the OA:α-LA stoichiometry, actually increases . Limited proteolysis of α-LA leads to disulfide-linked fragments that can form apoptosis-inducing complexes either by simple mixing with OA or by the chromatographic approach .
All of these observations are good indications that the ability to form these active protein–OA complexes are a generic property of the polypeptide chain, much like the ability to form amyloid aggregates , and that it is merely a question of finding the right conditions to encourage their formation and thus ‘turn virtually any natively folded protein into the carrier’ . Strong evidence to support this generic hypothesis has been provided in a study usingthe alkaline method to prepare complexes between OA and three completely unrelated proteins, namely α-LA, β-lactoglobulin (β-LG) and pike parvalbumin . All complexes were active against both bacteria (Streptococcus pneumonia) and mammalian Hep-2 cells, using the same mechanism (i.e. membrane depolarization and membrane damage). Strikingly, all three proteins show the same level of potency if their different OA binding stoichiometries are taken into account (a similar proportionality between OA levels and cytotoxicity has been reported for β-LG:OA complexes prepared by heat treatment) , and this potency is only two-to five-fold higher than that of free OA. Moreover, β-LG appears to adopt a native-like structure according to Trp fluorescence, indicating that a flexible structure is not critical for biological activity  (although this may also reflect the native high propensity of β-LG to bind many different types of hydrophobic molecules with relatively minor structural changes) [58, 59]. Rather, the it is suggested that the proteins act as ‘solubilizing vessels’ for OA to facilitate their transfer to the cell, much like the suggested cargo off-loading mechanism that we have proposed for ELOA .
The solubilization of OA by these proteins is demonstrated by a simple but elegant turbidity assay, which was also used to determine the protein–OA stoichiometry [42, 57]. Both turbidity and the fluorescence of the hydrophobic dye 6-(p-toluidino)-2-napthalenesulfonate have been used to follow the micelle formation of OA and it was found that this micellization is strongly reduced by cleaved , as well as intact  α-LA, leading to a strong suppression of turbidity. The nature of the contacts between OA and protein may vary from protein to protein. Working with complexes formed by direct titration, Spolaore et al.  suggested there were electrostatic attractions between the negatively-charged OA and the positively-charged clusters on (overall negatively-charged) α-LA (e.g. via the ten basic residues in the α-helices A, C and D) . However, the working under similar conditions, Knyazeva et al.  reported that the stoichiometry increases strongly with salt, indicating electrostatic screening and thus hydrophobic interactions. In practice, the two effects are probably difficult to separate, and there are likely to be multiple modes of interaction whose relative contributions will have to be deconvoluted by careful biophysical analysis.
The role of the lipid in the formation and activity of α-LA: lipid complexes
Understandably, there is much interest in determining how lipids impact on α-LA conformation and properties. Early studies highlighted a critical role of OA. NMR studies showed a well-defined interaction between OA and a predominantly (> 95%) monomeric α-LA state, in which the lipid is bound in a compact conformation . No cytotoxic complexes were formed if OA on the conditioning column was replaced by saturated fatty acids with 14–18 carbon atoms . A more systematic study showed that the two cis-unsaturated C18:1 fatty acids OA (C18:1, cis-Δ9) and vaccenic acid (C18:1, cis-Δ11) were the only fatty acids to form HAMLET-like states with a reasonable yield and a high level of activity ; other cis-containing fatty acids form complexes that are not apoptosis-inducing. Rather frustratingly, the non-apoptotic complexes were structurally identical to the active complexes in terms of secondary structure and the ability to bind 1,8-anilino-naphthalene-sulfonate. A simple explanation for this apparent paradox is that the cytotoxicity of these complexes is related to the cytotoxic properties of the bound lipid and the ability of the protein to take up the lipids in sufficient amounts. Somewhat confusingly, subsequent studies have vindicated the role of other fatty acids. For example, α-LA:linoleic acid complexes are reported to induce cell death, whereas stearic acid (a saturated fatty acid) hardly forms complexes . Yet other studies find that α-LA:lipid complexes are cytotoxic regardless of whether the fatty acids are cis or trans, or saturated or unsaturated, although there is variation in the level of toxicity; complexes with unsaturated fatty acids [OA, vaccenic acid, palmitoleic acid (C16:1, cis-Δ9), linoleic acid (C18:2, cis-Δ9, cis-Δ12) and elaidic acid (C18:1, trans-Δ9)] are generally far more cytotoxic than complexes with the saturated stearic acid (C18:0) .
The difference in the results obtained could be related to the preparation methods used in the three studies. Svensson et al.  suggested that the interaction between the fatty acids and α-LA was stereo-specific, favouring the cis conformation . However, Brinkmann et al.  successfully formed complexes with elaidic acid and stearic acid, which are both transfatty acids . The suggested stereo-specific interactions may play a role in how the fatty acids bind to the column matrix and how they are ‘presented’ on the matrix to be available for the protein. Because an ion-exchange resin was used by Svensson et al. , the pKa of the fatty acid may affect its interaction with the column. Because both the chain length and the conformation of the fatty acids influence the pKa, the interaction with the column and possible association with α-LA may differ between various fatty acids. A possible explanation for the difference in results could thus be the preparation methods used in the studies conducted by Svensson et al. , Zhang et al.  and Brinkmann et al. . The former study used the column based preparation method, whereas the latter two studies prepared the complexes by mixing α-LA and the lipids followed by acid treatment or heat treatment, respectively.
HAMLET and BAMLET complexes are not specific for cancer cells
According to numerous studies, HAMLET kills tumour cells and undifferentiated cells but not healthy differentiated cells [10, 32-34, 38, 46, 60-67]. However, this concept has been challenged by recent studies demonstrating that both cancer cells and normal primary cells are sensitive to killing induced by bovine α-LA:OA complexes (BAMLET) . Even though the most resistant cells tested were indeed normal primary cells (i.e. endothelial cells), the most sensitive cells were also normal primary cells, namely peripheral blood mononuclear cells freshly isolated from human blood (Fig. 4). Notably, BAMLET also caused lysis of erythrocytes . Although some normal cells may be more resistant to BAMLET than cancer cells, it is inappropriate to claim that BAMLET is tumour specific because some normal differentiated cells are as sensitive (or even more sensitive) to BAMLET than cancer cells. The cytotoxicity of BAMLET and HAMLET appears to be similar with respect to LC50 (i.e. the concentration needed to kill 50% of the cells) [37, 68], uptake, morphology and terminal deoxynucleotidyl transferase dUTP nick end labelling in A549 cells and the induction of cell death resembling necrosis in THP1 cells. In addition, HAMLET was shown to induce cell death in primary human vascular smooth muscle cells to the same extent as BAMLET . Interestingly, treatment with HAMLET or MAL (an early, crude preparation, directly obtained from breast milk, with activity similar to that of HAMLET) showed results similar to those obtained with BAMLET [1, 11]. In the original study, Håkansson et al.  reported that MAL is cytotoxic to lymphocytes, granulocytes and foreskin fibroblasts, whereas a subsequent study  stated that MAL treatment does not result in apoptosis in either of these cell types. Altogether, the similarities in the cytotoxicity of BAMLET and HAMLET, as well as the cell death of the primary human vascular smooth muscle cells induced by HAMLET, question the tumour specificity of HAMLET.
This does not imply that HAMLET is without effect for biological intervention. HAMLET can kill certain strains of bacteria (e.g. S. pneumonia and Haemophilus influenzae) [3, 69], and sublethal concentrations of HAMLET can potentiate the effect of antibacterial treatment of pneumococci . The combination of antibiotics and HAMLET reduces the required antibiotic dose and sensitizes even otherwise antibiotic resistant strains . In vivo studies with topical application of HAMLET have also been conducted  A two-phase clinical study on 40 patients with cutaneous papillomas showed that HAMLET treatment results in a decrease of lesion volume. However, the difference between the numbers of complete resolution of lesions in the two groups (HAMLET- or placebo-treated) was not significant. In a 2-year follow-up, complete resolution of all lesions were seen for 88% of the patients who had received HAMLET (n = 35). No adverse effects were reported . However, the study design and results received rather harsh criticism . HAMLET treatment was not compared to placebo- or conventional treatment because all patients received HAMLET treatment during the second phase of the trial.
HAMLET treatment has shown promising results on human glioblastoma xenografts in rats, with HAMLET treatment resulting in a reduction of tumour size and a delay of pressure symptoms compared to infusion with α-LA . HAMLET also has a beneficial effect on bladder cancer in mice and patients [65, 72]. Intravesical HAMLET instillations can delay tumour progression in a murine bladder cancer model but do not prevent tumour formation  and a preliminary clinical trial with nine patients showed that five daily intravesical installations for five consecutive days before scheduled transurethral resection resulted in increased cell shedding in the urine compared to treatment with NaCl/Pi (n = 1), saline (n = 3) or α-LA (n = 1) . In the previously cited in vivo studies, HAMLET was found to cause cell death in tumour cells but not in healthy surrounding tissue. This indicates that the difference in sensitivity of different cell types can be exploited, and that treatment with HAMLET-like protein–lipid complexes may well be promising in carefully chosen settings.
Cytotoxicity of different protein–lipid complexes
To be able to suggest which particular settings would be appropriate for treatment, a clearer understanding of what HAMLET and other protein–lipid complexes actually do is required. Despite numerous studies , the mechanism behind cell death induced by these complexes remains rather mysterious.
The cellular distribution of HAMLET and other protein–lipid complexes has been investigated by labelling α-LA, either alone or in complex with OA, with biotin [2, 40], I125  or fluorescent probes [40, 43, 63, 64, 74, 75]. HAMLET and other protein–lipid complexes, such as bovine and human α-LA:OA complexes and β-lactoglobulin:OA complexes prepared by mixing followed by heat treatment [40, 43], as well as ELOA prepared chromatographically , have all been shown to bind to the cell surface and to enter the cytoplasm of cancer cells [40, 43, 63, 75]. α-LA not in complex with OA binds weakly to the cell surface, and little or no α-LA is found in the cytoplasm in comparison to the protein–lipid complexes [2, 40, 63, 74]. After internalization, the protein–lipid complexes accumulate in the nucleus of the cancer cells [2, 40, 43, 63, 74]. Because OA is not labelled, it is not known whether α-LA and OA still exist in complex after internalization. By contrast to the other protein–lipid complexes, ELOA is not detected in the cytoplasm or nucleus of the treated cells until the integrity of the membrane is lost .
A recurrent problem in the interpretation of the reported discrepancies between the complex types and their constituents comprises the differences in the experimental set-ups, including the investigated cell types, the incubation times and the concentration of OA in the different complexes, all of which can influence the results considerably.
The apparent difference in the sensitivity of cancer cells and primary cells to HAMLET treatment has been investigated. Surface binding and cytoplasmic uptake of HAMLET was found with both cancer cells and primary cells . However, other studies have reported less efficient surface binding and a lower uptake of HAMLET in primary cells compared to cancer cells . Translocation of HAMLET to the nucleus occurs in the three cancer cell lines tested but not in the two primary cell types tested [nontransformed human astrocytes (CC-2565) and human renal tubular epithelial cells] . However, it should be noted that differences in subcellular location and nuclear translocation have not been investigated using less sensitive cancer cells (e.g. CaCo-2 cells)  or primary cells that are highly sensitive to treatment with BAMLET (e.g. peripheral blood mononuclear cells) . It remains to be determined whether the difference in uptake and translocation to the nucleus of HAMLET between the sensitive and less sensitive cell types correlates with differences in the membrane composition of these cells.
A large body of work mainly on HAMLET shows how the α-LA:OA complexes influence the cells and cause a plethora of intracellular effects. Cells treated with HAMLET show morphological changes characteristic of apoptotic cells [63, 76], induce caspase activation  and cause mitochondrial permeability transition in isolated mitochondria, which results in mitochondrial swelling, a loss of mitochondrial membrane potential and cytochrome c release . HAMLET treatment also shows signs of induction of autophagic cell death  and has been suggested to result in changes in proteasome structure and function . In addition, HAMLET has been found to co-localize with histones and perturb the chromatin structure  and also to facilitate cell detachment through binding to α-actinin  (Fig. 5).
Experiments conducted with BAMLET have shown that BAMLET, similar to HAMLET, causes dose-dependent chromatin condensation and cell shrinkage in cancer cells. BAMLET treatment activates signalling pathways leading to the induction of autophagy. Inhibition of autophagy, however, does not affect the viability of BAMLET treated cells, indicating that autophagy does not play a critical role in BAMLET-induced killing , in contrast to the results seen for HAMLET . In addition, where HAMLET has been found to be associated with mitochondria and accumulate in the nucleus , no co-localization with mitochondria or nuclei was seen for BAMLET. However, BAMLET does co-localize with lysosomes and induce the leakage of lysosomal enzymes followed by activation of the pro-apoptotic protein Bax. Based on these results, BAMLET has been suggested to induce cell death via activation of a caspase-independent lysosomal death pathway . It remains to be determined whether these differences between the mode of action of the two protein complexes on cells are linked to different molecular properties or reflect different experimental conditions. BAMLET and HAMLET have very similar properties overall, although BAMLET has a slightly reduced tendency to form annular pores in membrane monolayers compared to HAMLET .
Bovine α-LA:OA complexes and β-lactoglobulin:OA complexes prepared by mixing followed by heat treatment induce apoptosis  and bovine α-LA:OA complexes prepared using the column based preparation method have been reported to induce both apoptotic-like cell death and necrosis-like cell death dependent on the cell type [8, 11]. HAMLET-induced cell death in bacteria (S. pneumoniae) shows morphological and mechanistic similarities to apoptotic cell death in mammalian cells .
In summary, the protein–lipid complexes appear to influence several cellular components and mechanisms, although the exact mechanism for the induction of cell death remains unclear. It is not known whether the apparent differences between cell death induced by the various complexes relate to differences in the method and set-up used, or whether there is an actual difference between the cytotoxicity of the complexes. However, this demonstrates that protein–lipid complexes may have complex biological effects and that there is not necessarily one universal mode of action.
Effects of protein–OA complexes on artificial lipid membranes and cells
Recent reports on protein–lipid complexes focus on their effects on lipid membranes and thus add membrane permeabilization to the possible explanations for the cytotoxicity of the complexes.
Binding of α-LA to lipid bilayers at neutral pH has been established in several studies [23, 81, 82]. Apo-α-LA (without Ca2+) has a higher affinity for lipid bilayers than holo-α-LA  and the addition of calcium slowly reverts apo-α-LA to holo-α-LA, resulting in the release of α-LA from the lipid bilayer [23, 82]. These results suggest that the partial unfolding of apo-α-LA, which exposes hydrophobic areas , allows the protein to interact better with the membrane. Because OA in HAMLET appears to prevent reversion to the native state of holo-α-LA, the exposure of hydrophobic areas is maintained, favouring surface binding of HAMLET over α-LA. In addition, the presence of OA in HAMLET and other protein–lipid complexes may also increase the membrane affinity as a result of the aliphatic chain of OA, provided that this is not completely sequestered by the protein.
HAMLET binds to neutral and negatively-charged large unilamellar vesicles and giant unilamellar vesicles at pH 7.0 . HAMLET shows no sign of internalization in giant unilamellar vesicles but disrupts negatively-charged large unilamellar vesicles at physiological pH. At concentrations from 4 μm, HAMLET causes 100% leakage of fluorescent dye from the vesicles after just 1 min . This may be caused by pores because atomic force microscopy studies reveal HAMLET to form large annular pores on membrane monolayers . The HAMLET-induced effect on membrane permeability is also seen for live cells. HAMLET treatment of A549 lung adenocarcinoma cells has a time-dependent effect on the cells and the integrity of the membrane. After 30 min, HAMLET (initially added to 21 μm) accumulates at the plasma membrane and is internalized; after 3 h, the HAMLET-treated cells round up and form apoptotic blebs; and, after a total of 6 h, the membrane integrity is lost and HAMLET fills the cells . In comparison, α-LA binds weakly to the cell surface, and little or no α-LA is found in the cytoplasm [2, 63, 74].
Both HAMLET and α-LA:OA complexes prepared by direct titration of OA to α-LA at 17 °C show an increased binding to neutral small unilamellar vesicles compared to α-LA and OA alone. The complexes also affect the permeability of plasmalemma ion channels of the green alga, Chara corallina. The α-LA:OA complexes block the Ca2+ transmembrane currents and Ca2+-dependent Cl– currents, as well as leading to irreversible nonspecific K+ leakage currents. This wide range of effects indicates nonselective membrane permeability .
ELOA complexes bind to phospholipid membranes of giant unilamellar vesicles and supported bilayers but do not result in significant membrane permeability . Upon interaction with lipids, ELOA regains some of the native conformation enzymatic activity of equine lysozyme. This could be suggestive of release of OA from ELOA to the membrane, thus allowing equine lysozyme to re-fold into a native-like conformation . Experiments examining single rat adrenal pheochromocytoma cells (PC12) using live microscopy have demonstrated that ELOA accumulates at the plasma membrane. After ~ 1 h, the membrane ruptures, and ELOA rapidly enters and fills the cell .
In summary, all of the investigated protein–OA complexes exert an effect on lipid membranes, thus highlighting membrane alterations as a plausible explanation for the induction of cell death.
Cytotoxicity of milk and α-LA
In several of the studies available in the literature on HAMLET and HAMLET-like complexes, α-LA and OA are not found to be cytotoxic by themselves at the concentrations at which HAMLET is cytotoxic [2, 8, 35, 41, 47]. However, as described below, other studies have investigated the cytotoxicity of α-LA and OA and obtained contradictory results.
Human milk has been found to have an inhibitory effect on cell growth . The inhibitory activity is both donor- and cell type-dependent but is not correlated with the α-LA content in the donor milk. Consistent with the observations made by Håkansson et al. , fractionation of the milk into whey and casein showed that the inhibitory activity resides in the casein fraction and that a relatively large amount of casein (5–10 mg·mL−1) is required for growth inhibition. Pocovi et al.  found no inhibition of cell growth in bovine milk regardless of the donor cow or the fractionation, in good agreement with Pettersson et al. .
Although there are reports showing that native α-LA can also bind histones and thus potentially exert an effect on DNA structure without OA if positioned inside the cell , there is overwhelming evidence that α-LA is not cytotoxic unless it forms a complex with OA. There are a few studies that point in another direction: α-LA purified from human and bovine milk has been found to inhibit cell growth, although the required concentrations (5 ng·mL−1 to 25 μg·mL−1) and incubation times (3–5 days) differ between studies [87, 88]. Commercial α-LA has also been shown to inhibit cell growth  and to induce apoptosis and necrosis in α-LA-treated cells . However, this may be related to a different phenomenon, namely the formation of different aggregation states of α-LA. Thus, Xu et al. [91, 92] found the cytotoxicity of commercial bovine α-LA to be batch-dependent and linked this cytotoxicity to an oligomeric form of α-LA (SDS-stable 20- and 30-kDa forms), whereas monomeric α-LA isolated from the cytotoxic commercial α-LA preparations or from bovine milk was not cytotoxic. However, incubation of monomeric α-LA or inert batches of commercial α-LA in 30% trifluorethanol/acetate buffer (pH 5.5) at 37 °C for 5 days resulted in the formation of cytotoxic oligomers [91, 92]. The toxicity of these oligomers may derive from an unrelated phenomenon. Under conditions promoting the formation of amyloid (such as organic solvents, elevated temperatures, acidic pH or intermediate denaturant concentrations) , several proteins are able to transiently form oligomers with cytotoxic properties; these oligomers have been shown to lead to membrane permeabilization, probably as a result of pore formation or general disruption of the membrane [93, 94]. Interestingly, the observation that HAMLET also can form annular oligomers on membrane monolayers  suggests overlapping functionality with nonlipid oligomers.
A recent study by Lin et al.  revealed that different batches of commercial α-LA were contaminated with endotoxin to various degrees. The endotoxin level was not tested in any of the above-mentioned studies investigating the cytotoxicity of α-LA.
Cytotoxicity of OA
The effect of fatty acids is highly complex. Fatty acids alone can affect numerous cellular signalling and metabolic pathways in addition to playing important roles in immune responses and inflammatory processes, as reviewed previously [96, 97]. Furthermore, the accumulation of lipid in non-adipose tissues in hyper-lipidaemic states can cause cell dysfunction and cell death in a process known as lipotoxicity . The literature contains several reports on the cytotoxicity of fatty acids. Unsaturated fatty acids can induce apoptosis and necrosis in a range of different cell types and cause mitochondrial depolarization, the release of cytochrome c, caspase activation, accumulation of intracellular lipids and DNA fragmentation in treated cells [98-106]. In addition, unsaturated fatty acids can destabilize the plasma membranes of treated cells , leading to a loss of plasma membrane integrity and haemolysis of erythrocytes [98, 103, 108, 109]. However, the sensitivity of different cell lines to treatment with fatty acids varies with both concentration and the type of fatty acid. Studies investigating the cytotoxicity of protein–lipid complexes increasingly pay attention to the toxicity of OA alone [5, 11, 12, 47]. Knyazeva et al.  report a toxicity of OA at ≥ 300 μm , and Svensson et al.  report that ‘very high’ OA concentrations cause necrosis; however, OA does not induce apoptosis at any of the concentrations tested . Permyakov et al.  report that the content of OA in the complexes is critical for the cytotoxicity and that the cytotoxicity of OA alone is comparable to the cytotoxicity of the protein–lipid complexes (Fig. 6). This view is supported by other studies [11, 43, 44].
In summary, within the ‘HAMLET-field’, there is consensus regarding the lack of cytotoxiciy of α-LA, whereas the question of OA cytotoxicity remains controversial.
Perspectives on protein–lipid complexes
The cytotoxicity of a complex between two naturally occurring constituents in breast milk, as well as its extension to include other proteins without direct connection to milk, is intriguing. Despite the contradictory results reported above, the proposed selectivity of HAMLET-induced cell death and the promising in vivo studies of HAMLET make this area of research very interesting. Clearly, there are many questions that remain unanswered.
The precise mechanism of cell death induced by HAMLET and HAMLET-like complexes is not well understood. Several cellular targets and mechanisms have been suggested to influence HAMLET-induced cell death. However, it is not clear whether they are all equally important or whether some are merely secondary effects to a primary cell death-inducing effect. It remains to be determined whether there are two different effects of α-LA:OA complexes: (a) perturbation of cellular membranes and (b) effects of translocated α-LA:FA complexes in the cytoplasm and nucleus. Alternatively, membrane perturbation is the primary effect and intracellular effects such as caspase activation, proteasome and chromatin changes are secondary.
The ability to form cytotoxic protein–lipid complexes with other proteins [12, 43] suggests that OA is the key player in the induction of cell death. This is supported by the cytotoxic properties of OA noted earlier in this review and also by recent studies showing a cytotoxic effect of OA alone in concentrations present in cytotoxic levels of HAMLET [11, 12] (Fig. 6). Treatment with OA alone induced cell death similar to that induced by BAMLET, with Jurkat cell and THP1 cells going into cell death resembling apoptosis and necrosis, respectively . The interaction of HAMLET with α-actinin  may in part be similar to that described for OA alone . In addition, the reported inhibition of proteasome activity caused by HAMLET  may well be related to that of OA . Furthermore, the bactericidal action of HAMLET [3, 70] may be related to that of OA because OA has a bactericidal effect on group A Streptococci  and affects bacterial viability and biofilm production in Staphylococcus aureus . In addition, drug-resistant S. aureus is eradicated by liposomal oleic acids . The OA-mediated toxicity of the protein–lipid complexes may be supplemented by additional (possibly synergistic) effects from the protein component itself, which, as a partially unfolded oligomeric species, could also have intrinsic cytotoxic properties similar to those observed for many pre-fibrillar oligomeric species [93, 94]. Such side-effects will of course vary from one protein to the next.
The cytotoxic protein–fatty acid complexes can be prepared using different methods and their mechanism of action appears to be different. For example, the cellular uptake of HAMLET is reported in a number of studies, whereas the presence of ELOA inside the cells is not seen until the cell membrane ruptures and ELOA then fills the cells. In addition, there are differences between cell death induced by HAMLET and BAMLET with regard to co-localization with mitochondria, lysosomes and the nucleus, and the role of autophagic and lysosomal death pathways.
Ideally, we need to be able to determine whether the molecular properties of the different protein–lipid complexes correlate with their biological efficacy. This requires that researchers compare the cytotoxic properties of different protein–lipid complexes with systematically varying ratios of protein and lipid and a range of sizes, as well as investigate whether these structures are retained upon exposure to a cellular environment. We also need to be able to follow the fate of the OA that is presumably released when the protein–lipid complex is exposed to the membrane. Given that some proteins such as β-lactoglobulin do not need to unfold to complex with lipids, it also remains to be determined whether the protein truly needs to be in a dynamic and structurally fluctuating state to form these remarkable complexes. However, the emergence of this new class of protein–lipid complexes will undoubtedly stimulate further research into the range of conformational states that proteins can assume under different conditions, thus ultimately increasing our understanding of protein properties in general.
DEO is supported by the Danish Research Foundation (inSPIN) and the Lundbeck Foundation (BioNET 2). We are grateful to Dr L. Morozova-Roche and Dr J. Skov Pedersen for ongoing stimulating and enjoyable discussions.