Many clinical studies have been reported, involving ESRD patients, on albumin permeating into the dialysate. Only recent relevant publications are reviewed chronologically here.

Tsuchida and Minakuchi [24] in a clinical study involving 118 patients, (treated with a highly permeable HC filter), found that albumin leakage was on average 7.7 ± 1.0 g/session, whereas 314 patients using conventional high-flux HC filters exhibited lower albumin loss. They also reported that in the recommended Japanese standard for classification of HC membranes, “the desirable albumin leakage per treatment is less than 4 g”. In a clinical study by Fournier et al. [25], 8 patients underwent OL-post HDF and only albumin losses in dialysate were determined (i.e. 3134 ± 2450 mg/session); it was also reported that such losses did not lead to hypoalbuminemia. Although mass of albumin adsorption/deposition on the membrane was not determined, it was considered substantial depending on HC filter type. Vega et al. [26] in a cross-sectional study, involving 20 patients receiving OL-post HDF, analysed albumin leakage during the first hour of HC session. Moreover, ‘protein cake’ formation was considered responsible for the gradual reduction of permeability and albumin losses during this period. Potier et al. [7] collected data on albumin loss (to dialysate only) from sessions involving 37 patients and 19 different dialyzers; among other results obtained, they concluded that 4/19 dialyzers lose more than 5 g/session albumin and should not be used in OL-post HDF. Gayrard et al. [23], in a study involving 12 ESRD patients, determined total protein removal to dialysate ~ 2.3 g per session, in maximum convection OL-post HDF. They also employed a protein elution protocol ex situ to identify (and determine the mass of) particular plasma proteins (including albumin) adsorbed on the used HC filters. They reported that the total amount of adsorbed proteins in the membranes was only 6.1% of the respective amount of proteins removed through the dialysate. However, the accuracy/reliability of determining the total mass of the deposited proteins with their elution protocol is unclear, as discussed in “Discussion”.

Cuvelier et al. [27] reported on the case of a woman treated with high convective volume OL-post HDF, who developed severe hypoalbuminemia, attributed to massive albumin loss into dialysate, i.e. 23.6 g albumin loss in one session, whereas she only lost 4.6 g in a regular HD (haemodialysis) treatment. Such loss per session (23.6 g) is by far the greatest ever reported “leakage”, raising questions regarding its reliability/representativeness. Finally, in a recent study [28] involving 52 patients undergoing high-volume OL-post HDF, the temporal variation of albumin removal only through leaking was determined for three types of HC filters. Modest cumulative removal (~ 1.0 g to ~ 1.5 g per 4 h session) was reported. Moreover, “secondary membrane” formation was considered to interpret these data, although no attempt was made to quantify it. It was also concluded (with insufficient justification) that the albumin sieving-kinetics data point to reduced formation of ‘secondary membrane’.

Several recent review papers (presented chronologically) also deal with these issues. Boschetti-de-Fierro et al. [6] assessed clinical studies on the performance of HC filters. Although the significance of ‘secondary membrane’ was discussed, no such albumin loss data were provided. Furthermore, they suggested, based only on albumin leakage data [24], that “7.7 ± 1.0 g/session is an estimate for a threshold of albumin removal (that could impact serum albumin levels), which should not be exceeded…”. Similarly, Van Gelder et al. [8] concluded that with convective therapies (OL-post HDF), albumin loss (through leakage only) is significant (range: 0.08–7 g per 4 h treatment); however, they also noted that the acceptable upper limit of dialysis-related albumin loss remains unknown. Ward et al. [4] reviewed studies regarding the possible effect of albumin losses (only due to leakage) on hypoalbuminemia and the importance of concomitant inflammation on outcomes in ESKD patients. They cautioned on use of membranes causing albumin loss of 20 g/week, whereas the use of HC filters resulting in weekly loss of 12 g (i.e. ~ 4 g per session?) appeared to pose little risk to patients. Kalantar-Zadeh et al. [9] considered that albumin loss into the dialysate was (a potentially modifiable) cause of hypoalbuminemia; however, they also remarked that protein adsorption to the membrane and tubing can occur and that patients tend to lose approx. 6–8 g of total amino acids per session. In addition, it was noted that no definition of “excessive” albumin loss during dialysis has been proposed or accepted.

The interaction of human plasma proteins with membrane materials has been extensively studied (in vitro) in the general context of biocompatibility of materials for medical applications including HC [12]. In early seminal papers by Vroman and Adams [29, 30], interesting phenomena of competitive protein exchange on artificial surfaces have been observed, in which proteins already adsorbed on a surface (from a protein-mixture solution) are displaced by others, subsequently arriving. Significant research has followed because such exchanges, commonly referred to as the “Vroman effect”, seem to be related to blood platelet adhesion to surfaces and clotting (e.g. [31]). Of particular interest to this review are observations that, during the initial adsorption on surfaces, unfolding occurs of albumin and fibrinogen, under high concentrations as in HC [32]. Moreover, there is evidence that these proteins tend to competitively displace other adsorbed proteins [19, 33]. Soderquist and Walton [34] investigated the interrelation of adsorption/desorption processes (on co-polypeptide and silicone surfaces) with the structural changes of adsorbed albumin, y-globulin and fibrinogen. They suggested a three-stage process, including an initial reversible adsorption, a second phase where the adsorbed proteins undergo slow conformational change (with proteins essentially irreversibly adsorbed) and a final stage where the denatured material is slowly desorbed. The observed rather long timeframe of the last stage appears to be irrelevant to the shorter 4 h period of a HC session. It was also noted that denaturated albumin desorption (through such mechanism) or detachment by shear forces have been inadequately studied. Sivaraman and Latour [35] found that platelets bind to adsorbed albumin (through receptor-mediated processes), whose binding sites are formed by adsorption-induced protein unfolding. Importantly, a high degree of such unfolding, was correlated strongly with increased level of platelet adhesion. Moreover, greater albumin adsorption occurred with increasing albumin solution concentration. This was attributed to the fact that the transport rate of protein molecules to the surface increases as their concentration increases; thus, the molecules that adsorb from higher concentration have less time to unfold and spread before the surface becomes saturated with protein.

Pieniazek et al. [36] investigated changes in albumin structural characteristics during HD (haemodialysis). They evaluated the susceptibility of plasma albumin to oxidation in ESRD patients, before and after a HC session, in comparison to healthy persons. They also assessed the conformational state of albumin under such conditions, employing EPR (electron paramagnetic resonance) spectroscopy. Significantly, their data showed that during HC the level of thiols (± SH groups) was significantly affected, decreasing by ~ 15%. They concluded that the significant conformational changes, occurring in vivo during HC, negatively affect the albumin antioxidant function. Finally, Sishi et al. [37] recently investigated interactions between proteins and membrane material made of PES (polyethersulfone), PAN (polyacrylonitrile) and PVDF (polyvinylidene fluoride). In particular, they examined adsorption of main human serum proteins (albumin, fibrinogen, transferrin), at realistic concentrations, across the membrane thickness (i.e. into the pores), using an in-situ SR-μCT (Synchrotron-based X-ray micro-tomography) imaging technique. Albumin was preferentially adsorbed to all three membranes. PES membrane, possessing comparatively larger pores, adsorbs albumin within its whole thickness, whereas PAN and PVDF membranes tend to absorb it only at the top and in middle layers. SEM (scanning electron microscope) image analysis was employed to identify changes in the deposited proteins morphology, depending on membrane properties.

There is a significant amount of in vitro work on protein adsorption and deposition to membranes, where HC conditions are simulated to various degrees, aiming to clarify the complicated phenomena involved, which will be briefly reviewed. On the contrary, there is hardly any definitive in vivo study regarding albumin mass adsorbed/deposited in HC filters.

Interpreting data on membrane performance, in an early clinical study, Rockel et al. [13] recognised the secondary membrane formation and its significant effects on permeability and species rejection, but did not determine the deposited-protein mass/loss. The latter was neglected and account was taken only of albumin leaking into the dialysate, i.e. ~ 1.4 g per session. Later Gachon et al. [38], using an elution protocol, determined the adsorbed proteins on used HC filters after a session. The reported amount of adsorbed proteins, for ~ 1 m2 membrane surface area, was extremely small, i.e. < 10 μg total. The applied protocol involved extensive preliminary flushing (by recirculating saline solution), followed by sequential treatment with elution solutions; finally, reverse transmembrane pressure/flushing was applied to recover proteins adsorbed within the membrane pores. However, one can express reservations on the fitness of such protocol, to determine total mass of deposited proteins, particularly because the fouling layer (above the ‘tightly’ adsorbed proteins on the inner membrane surface) could be removed by flushing and be unaccounted. Significantly, the authors [38] express concern that protein may still remain adsorbed in the HC filters, even after the latter have been subjected to this intensive treatment protocol.

Langsdorf and Zydney [17] have shown that the permeation characteristics of particular flat-sheet (Cuprophan and PAN) membranes can be described using a two-layer membrane model, i.e. that a layer of adsorbed plasma proteins provides an additional resistance to mass transfer in series with that of membrane itself. Later, Morti and Zydney [39], using PAN and CTA (cellulose triacetate) HC filters, performed in vitro tests with human plasma, under rather “mild” conditions (QUF = 0, Qblood = 200 and Qdial = 500 mL/min) and measured permeability as well as other characteristics of deposited secondary layer. They determined experimentally the developing thickness of protein layer for PAN and CTA HC filters at 1.9 ± 0.5 μm and 4.4 ± 0.5 μm, respectively. It is estimated that, for a typical HC filter of 2.0 m2 surface area and fibre inner diameter 200 μm, a layer thickness 1.0 μm amounts to a deposit volume of ~ 2.0 mL (or ~ 2 g, for deposit density ~ 1 g/mL); therefore, the total mass of deposited proteins corresponding to these data is roughly ~ 4 g to ~ 9 g. However, under conditions of large convective/ultrafiltration rates (QUF ≈ 100 mL/min), as in OL-post HDF, one would expect a significantly greater mass of deposited proteins, including albumin. Birk et al. [40] tested (in vitro) 12 commercially available HC filters (using 11 different membrane materials) and perfused them with human blood containing 1251-labelled plasma proteins. Under filtration conditions (not quite representative of those prevailing in high-convection HC modes), the total protein adsorption ranged from 338 to 2098 mg/m2 membrane surface, whereas the fraction of adsorbed low-molecular-weight-proteins (LMWP < 65 kDa) varied between 14 and 70% of total protein.

Yamamoto et al. [16] investigated the effects of internal filtration/ultrafiltration on membrane fouling based on the membrane’s pure-water permeability, diffusive permeability, and sieving coefficient. Membrane fouling caused by protein adhesion was shown to increase due to enhanced ultrafiltration, particularly at the early treatment stage. Although evidence of membrane fouling was clear, the albumin/protein mass deposited on the membranes was not quantified. Tomisawa and Yamashita [41] using HC filters made of PMMA (polymethyl methacrylate) and PEPA (polyester polymer alloy) membrane material simulated HC by employing dilute synthetic BSA (bovine serum albumin) solutions (i.e. 5.10 g BSA in 2000 mL batches). It was reported that fractional BSA adsorption exceeded 50% (i.e. ~ 2.3 g) at rather high QUF (> 60 mL/min) with PMMA membranes but smaller with PEPA ones, concluding that the significant amount of albumin adsorbed by the membrane should be taken into account when a clinical criterion of the total albumin loss is considered. In vitro tests by Kim et al. [42] confirmed previous results [17] that protein deposition occurs quickly, noting that the properties of the protein-deposit become nearly constant after ~ 20 min as also noted earlier [17]. Moreover, it was suggested that protein deposition is enhanced by increasing ultrafiltration rates, further affecting HC performance.

Gomez et al. [22] employing a novel in vitro uremic matrix, determined total albumin loss during simulated HDF sessions. Reported data with a PMMA-HDF mode were: total mass of albumin extracted/lost (Mext) > 15 g, albumin lost in dialysate (Mdial) ~ 5 g and albumin adsorbed only ~ 50 mg. The difference [Mext − Mdial] = > ~ 10 g may suggest that this albumin mass is absorbed/deposited within the module and possibly in the rest of HC circuit. This [Mext − Mdial] difference was ~ 4.5 g for CTA-HDF treatment. Although some sources of error or uncertainties should be considered, these relatively large albumin deposited/adsorbed mass cannot be overlooked. Kiguchi et al. [43] used dilute aqueous albumin solution (4 g/L), recirculated through three types of PEPA and one PSf (polysulfone) HC filter, to simulate fouling and study-related clearance effects. Although an albumin layer was developed and immobilised, the deposited albumin mass was not determined. In a clinical study, by Vanommeslaeghe et al. [44], involving 10 ESRD patients, data on albumin concentration in inlet and outlet blood/plasma streams were obtained (designated as Albinlet, Alboutlet). These data were apparently used to correct respective venous concentrations, and determine extraction ratios, but not to estimate total albumin losses during HC session. Finally, Abdelrasoul et al. [45] investigated the competitive adsorption (on PES membrane) of main proteins albumin, fibrinogen (FB) and transferrin (TRF), by employing synthetic single and multiple protein solutions. In general, the proportion of adsorbed FB and TRF in the deposit was significantly greater than that in the initial protein feed-solution, suggesting preferential adsorption of those proteins compared to albumin. In addition, using the special SR-μCT technique, the adsorbed albumin within the membrane pores appeared to be dominant and substantial. However, no specific quantitative data on total adsorbed/deposited mass of those proteins were obtained, although recognised as significant.

There are few recent reviews on protein/membrane–material interactions of relevance to the topic of this paper. Huang et al. [14] dealt with blood–membrane interactions that influence solute removal. The role of secondary membrane formation, and concentration polarisation on membrane performance was discussed. Attention was paid to the composition of fouling layer (comprised mostly of the dominant proteins, albumin, fibrinogen, immunoglobulinG) and its effect on inflammatory response and thrombogenicity. Westphalen et al. [46] assessed our understanding protein-adsorption phenomena during HC, including related mechanisms and blood activations as well as the associated consequences. It was concluded that there is no model available to correlate/estimate the rate (or mass) of protein adsorption or the total amount of protein adsorbed during hemodialysis as a function of main operating conditions.