A new class of recombinant human albumin with multiple surface thiols exhibits stable conjugation and enhanced FcRn binding and blood circulation

Human serum albumin is an endogenous ligand transport protein whose long circulatory half-life is facilitated by engagement with the human cellular recycling neonatal Fc receptor (hFcRn). The single free thiol located at Cys-34 in domain I of albumin has been exploited for monoconjugation of drugs. In this work, we increased the drug-to-albumin ratio potential by engineering recombinant human albumin (rHSA) variants with varying hFcRn affinity to contain three free, conjugation-competent cysteines. Structural analysis was used to identify positions for cysteine introduction to maximize rHSA stability and formation of the conjugated product without affecting hFcRn binding. The thiol rHSA variants exhibited up to 95% monomeric stability over 24 months and retained hFcRn engagement compared with a WT unconjugated control demonstrated by Biolayer Interferometry. The additional cysteines were further introduced into a panel of rHSA variants engineered with different affinities for hFcRn. After conjugation with three Alexa Fluor 680 (AF680) fluorophores, hFcRn binding was similar to that of the original triple-thiol nonconjugated rHSA variants (0.88 and 0.25 μm for WT albumin with or without 3xAF680 respectively, and 0.04 and 0.02 μm for a high hFcRn–binding variant with or without 3xAF680, respectively). We also observed a 1.3-fold increase in the blood circulatory half-life of a high hFcRn–binding triple-thiol variant conjugated with AF680 (t½ = 22.4 h) compared with its WT counterpart (t½ = 17.3 h) in mice. Potential high drug-to-albumin ratios combined with high hFcRn engagement are attractive features of this new class of albumins that offer a paradigm shift for albumin-based drug delivery.

Human serum albumin (HSA) 3 is the most abundant plasma protein in blood. It functions as a transport protein facilitated by multiple noncovalent binding sites for endogenous molecules such as fatty acids, steroids, and bilirubin (1) and a long circulatory half-life (ϳ19 days) predominately facilitated by engagement with the cellular recycling heterodimeric neonatal Fc receptor (FcRn) (2)(3)(4)(5). These properties promote albumin as an attractive drug delivery platform (6 -8) that has been utilized for long-acting drugs such as the marketed Levemir and Victoza (Novo Nordisk) that reversibly associate with hydrophobic binding pockets in endogenous HSA by inclusion of fatty acid chains in the drug design. An alternative approach to reversible association is covalent drug conjugation that offers more stable drug attachment and a defined drug-to-albumin ratio (DAR). HSA consists of three domains, with the major FcRn binding interface lying in domain III (DIII) (9). Drug attachment distal to this binding interface is thought to be a requirement to retain HSA's cellular recycling capacity and concomitant prolonged circulatory half-life. A single free cysteine at position 34 (Cys-34) located in DI has, therefore, been a focus for site-selective drug conjugation (10 -12). For example, Kratz et al. (13) reported maleimide-mediated drug conjugation to Cys-34 of endogenous albumin after intravenous injection of doxorubicin-maleimide. Although an elegant approach, the lack of available Cys-34 on endogenous albumins due to post-translational oxidation (cysteinylation or nitrosylation) (14) may limit reactivity toward the endogenous HSA pool. In vitro conjugation of drug to rHSA variants with Cys-34 in the reduced form (10) offers an alternative strategy to obtain a high conjugation efficiency and an opportunity to utilize albumins engineered for improved pharmacokinetic properties due to different hFcRn engagement (3,10). Availability of a single free thiol, however, still only permits a DAR of 1:1, limiting the drug-carrying potential.
This present work aims to increase the conjugation capacity by site-selective introduction of additional thiol groups into a range of rHSA variants with different FcRn-binding affinities described previously (3,15,16).
Analysis of the co-crystal structure of HSA engagement with hFcRn was used to identify and select desirable sites for the introduction of two additional conjugation-competent cysteine residues, Lys-93 and Glu-294, in DI and DII, respectively ( Fig. 1). This engineered triple-thiol variant (DAR3 variant) exhibited high conjugation efficiency and low aggregation and retained hFcRn engagement. When these mutations were introduced into a range of rHSA variants with different hFcRn affinities, tunable blood circulatory time in mice was observed. This work presents an albumin platform that allows enhanced DAR, albumin multifunctionality, and high hFcRn engagement.

Identification of positions suitable for thiol introduction
Positions for the introduction of cysteines with reactive thiol groups were based on structural analysis of co-crystal complexes of HSA and hFcRn (17,18). Positions were avoided whereby thiol introduction could potentially negatively impact folding of albumin e.g. due to the number of disulfide bonds required to maintain tertiary structure or its monomeric status. Accessibility of the thiol group for conjugation was also assessed. It was highly desirable to avoid a negative impact on hFcRn binding by direct interference with the intermolecular interface or by the disposition of subsequently conjugated molecules leading to steric hindrance with hFcRn. Therefore, positions in the principal binding interfaces were not selected for screening. Positions identified as meeting all of these criteria are shown in Fig. 1 (Cys-34, K93C, and E294C), Fig. S1, and Table S2 (full panel).

Free thiol profiles
Conjugation to Cys-34 was prevented by replacing the cysteine with an alanine (C34A), and conjugation profiles were assessed for 36 new introduced cysteine mutations in this C34A rHSA. To detect the number of free thiols, 5,5-dithio-bis-(2nitrobenzoic acid) (DTNB) was incubated with the variants, and mass increase (ϩ197 Da for 5-thio-2-nitrobenzoic acid (TNB)) was detected using MS. An increase in mass of 197 Ϯ 15 Da measured by MS upon DTNB incubation was indicative of the presence of one free thiol group on the protein in the sample. All variants exhibited successful binding of a TNB molecule, as shown in Table 1 (a sample selection) and Table S2 (all samples).

Aggregation characteristic profiles
The tendency of WT rHSA or the thiol rHSA variants to aggregate was determined by measurement of monomer, dimer, and higher-order oligomer quantities in solutions of the single-thiol rHSA variants. Gel-permeation high-performance liquid chromatography (GP-HPLC) and ultrahigh-performance liquid chromatography (UHPLC) were used to determine the aggregation of the single-thiol rHSA variants. Results are typically expressed as "percentage monomer," which is calculated as follows: amount of monomeric albumin by mass ϫ 100/(amount of monomeric albumin by mass ϩ amount of dimeric albumin by mass ϩ amount of higher-order oligomer by mass). The majority of shake flask-derived rHSA variants after purification by chromatography using AlbuPure matrix (9) had a monomer percentage equivalent to or higher than that of the WT rHSA control at t ϭ 0, which had a monomer percentage of 86 -87%. The variants maintained their monomeric protein stability during 6 months of storage at 2-8°C, with no significant evidence of aggregation observed (Table 2 and  Tables S3 and S4). However, it was evident that variants L66C, E277C, and E311C had a relatively low percentage monomer at t ϭ 0 and consequently had a propensity to form aggregates.

Conjugation efficiency profiles
The MS spectra for the single-thiol rHSA variants L66C, A92C, Q94C, A226C, D259C, L275C, and L284C indicated no conjugation following overnight incubation with maleimide-PEG2-biotin (Table S5). The MS spectra for the single-thiol rHSA variants E230C and I271C indicated that conjugation had occurred after overnight incubation, with both giving 72% monoconjugate when comparing the relative peak heights of conjugated and unconjugated species. The MS spectrum for single-thiol rHSA variant K93C exhibited a single species at 66908 Da (Table 3), indicating 100% conjugation and confirming that K93C has only one single free thiol group (data not shown). Also, the MS spectra for the WT rHSA control and the single-thiol rHSA variants L24C, V54C, H128C, E227C, K240C,  Engineered albumin with enhanced drug-carrier potential K262C, Q268C, E277C, E294C, K317C, A322C, E358C, K359C, and A362C all indicated that Ͼ90% conjugation was achieved with maleimide-PEG2-biotin (Table 3 and Table S5).

Variant stability over long-term storage
The double-thiol (DAR2) and triple-thiol (DAR3) rHSA variants and the DAR3 combined with a high-binding hFcRn (HBI) mutation (K573P) (3) were assessed with a single-thiol WT rHSA for stability during long-term storage at 5°C for 24 months in 25 mM sodium phosphate, 215 mM sodium chloride, pH 6.5. The aggregation profiles were determined at various time points by GP-HPLC (Table 4 and Fig. S3). All rHSA variants had a monomer percentage greater than 98% at t ϭ 0. The variants maintained a monomeric protein percentage Ն 95% after 24 months of storage at 5°C.

Conjugation with Alexa Fluor 680
The K93C and E294C mutations were combined with null FcRn-binding (NB) (3,9) and a high FcRn-binding hFcRn HBII variant described previously (16). The thiol variants were conjugated with Alexa Fluor 680 (AF680) dye, and the labeling efficiency was determined by the ratio of absorbance of the fluorophore at 680 nm and albumin at 280 nm. The 680 nm/280 nm ratio for the single-thiol WT rHSA variant was 1.3. The ratios for the triple-thiol rHSA variants NB, WT, and HBII were shown to be 4.3, 3.7, and 3.7, respectively, which indicated that the triple-thiol rHSA variants were ϳ3 times more fluorescent than the single-thiol WT conjugate (Table 5).

hFcRn affinity of albumin variants
The hFcRn-binding affinities of the selected triple-thiol NB, WT, and HBII variants were determined by biolayer interferometry (Table 6 and Fig. 2). Triple-thiol rHSA variants conjugated to three AF680 fluorophores showed slightly lower affinity compared with the unconjugated counterpart. For the triple-thiol WT rHSA, the conjugation of 3xAF680 increased the dissociation constant (K D ) from 0.25 to 0.88 M. For the triple-thiol HBII rHSA, the K D increased from 0.02 to 0.04 M after conjugation of 3xAF680. The triple-thiol NB rHSA did not show any hFcRn binding with or without 3xAF680 conjugated. The triple-thiol rHSA variants, conjugated or unconjugated, all showed the same tendency of K D value following the ranking WT Ͼ HBII, with NB not detectable. These findings show that the hFcRn affinity is not diminished due to the K93C and E294C mutations and is only marginally affected when fluorophores are conjugated to the three free thiols.

Pharmacokinetics of albumin variants in mice
The triple-thiol rHSA 3xAF680 fluorescence-labeled variants (NB, WT, and HBII) were administered by intravenous injection into BALB/cAnNRj mice (n ϭ 7); blood samples were taken at 1 min, 30 min, 4 h, 24 h, 48 h, 72 h, and 96 h; and the level of fluorescence in blood detected by In Vivo Image System (IVIS) (PerkinElmer Life Sciences) bioimaging (Table 7, Fig. 3, and Fig. S4). The blood circulatory half-life (t1 ⁄ 2 ) for the variants showed an increase from 17.3 h for the triple-thiol WT rHSA Table 2 Selection of samples for GP-HPLC aggregation screening indicates monomeric stability over 6 months at 2-8°C

Discussion
A high drug load, site-selectively conjugated to rHSA without compromising hFcRn binding, is important for harnessing albumin's physiological transport properties for effective drug delivery. We present, for the first time, rHSA variants   Table 6 Binding affinities for triple-thiol rHSA variants toward hFcRn, pH 5.5  Table 7 Elimination half-life (t1 ⁄2 ) calculated from the exponential fit curves (Fig. 3)  Engineered albumin with enhanced drug-carrier potential site-selectively engineered by a structural screen to contain three free cysteines for improved DAR without affecting hFcRn engagement. Candidate positions were selected for screening to avoid negatively impacting binding of albumin to hFcRn, based on cocrystal structures. Additional factors included accessibility for conjugation and predicted minimal impact on the monomeric or folded status of albumin. Surface exposure and flexibility of the backbone in the regions of introduced thiols were noted as promoting these latter features.
Following identification of positions with desirable conjugation features, the spatial disposition of the sites on rHSA was considered before production of variants containing multiple free thiol groups, to limit possible steric interactions from conjugated moieties. The spatial distribution of the natural and the two engineered thiols have been illustrated based on a previously published crystal structure of FcRn in complex with HSA ( Fig. 1). Positions selected for mutations were Lys-93 (to give K93C) and Glu-294 (to give E294C), which, when combined with the Cys-34, gives the triple-thiol rHSA a monomer stability Ͼ 95%. Further combining with K573P gives high hFcRn binding (HBI) (3) ( Table 4). The single-thiol K93C and E294C (C34A versions) rHSA variants each showed promising conjugation potential (100 and Ͼ95%, respectively) when conjugated to maleimide-PEG2-biotin.
These two additional thiol mutations, K93C and E294C, were introduced in DI and DII distal from the predominant hFcRn binding interface in DIII. However, DI has a contributory role in binding to FcRn, and Cys-34 is proximal to this interface (19). Conjugation to the free cysteine at Cys-34 in DI can, therefore, influence the hFcRn binding affinity as observed in earlier studies (11,20). In this work, the hFcRn-binding affinity of the triple-thiol WT-FcRn-binding rHSA variant was retained and in the same range as the WT HSA single-thiol control (K D was 0.25 M compared with 0.55 M, respectively). This was also observed for the fluorophore-conjugated triple-thiol WT rHSA version carrying three AF680 molecules (K D ϭ 0.88 M). The K93C and E294C mutations were further introduced into rHSA variants either engineered for null-FcRn binding (NB; H464Q/ K500A) or for high FcRn binding (HBII; E492G/K573P/K574H/ Q580K). The NB rHSA variant contains the K500A mutation combined with a key histidine residue known to reduce FcRn binding (9), where the HBII rHSA variant is a combination variant comprising K573P and additional mutated positions from domain III in regions that have previously been shown to impact FcRn binding (3,9) and exhibit a long circulatory halflife (16). Neither of the triple-thiol NB rHSA variants (with or without 3xAF680) exhibited hFcRn binding, whereas the triplethiol HBII rHSA variant (with or without 3xAF680) bound with Ͼ10-fold higher affinity to hFcRn compared with triple-thiol WT rHSA variants (with or without 3xAF680). There was a slight decrease in binding affinity after conjugation of three AF680 molecules observed as a K D shift from 0.02 to 0.04 M for the triple-thiol HBII rHSA variants and from 0.25 to 0.88 M for the triple-thiol WT rHSA variant. Conjugation of nucleic acids or PEG to rHSA has previously been shown to decrease hFcRn binding affinity, but hFcRn engagement could be rescued by using a high hFcRn-binding rHSA (HBI) variant (11,20). The high-binding (HBII) triple-thiol rHSA described here exhibited higher hFcRn affinity when bearing three AF680 molecules compared with the unconjugated single-thiol WT rHSA control; therefore, both the DAR and hFcRn engagement are improved overall.
Three-AF680 molecule conjugation creates a fluorescent detection system for analysis of the pharmacokinetic properties. Furthermore, conjugation of 3xAF680 (ϳ1000 Da/molecule) mimics transportation and distribution of small molecular drugs conjugated to HSA. The Cys-34 -conjugated nucleic acids and PEG molecules (presented in Refs. 11 and 20, respectively) are all considered macromolecular drugs due to their size (10 and 5-30 kDa, respectively) and are more likely to create steric hindrance in the binding of HSA to hFcRn, as observed in the decreased binding affinity toward hFcRn. Small molecular drugs, such as the anticancer therapeutics paclitaxel and doxorubicin (800 and 500 Da, respectively) may exhibit minimal alteration of HSA-hFcRn binding, as observed with the 3xAF680 conjugation. The application of fluorophore conjugation creates a platform for easy measurement of the circulatory half-life of rHSA, by scanning fluorescent signal in blood samples as described previously by our group (21). The relative fluorescent signal in heparinized plasma was detected, and halflives were calculated. The circulatory half-life of the NB and that of the WT triple-thiol rHSA were similar (t1 ⁄ 2 ϭ 15.4 and 17.3 h, respectively). This similarity for NB and WT rHSA was shown by Andersen et al. (3) in NMRI mice, where the half-lives of unconjugated NB and WT rHSA were 19.1 and 21.0 h, respectively. The lack of a larger difference in circulatory halflives of the NB and the WT is most likely due to the effect of competition from the endogenous MSA for mouse FcRn (mFcRn) binding. This is supported by work in MSA-deficient mice, where HSA showed a longer half-life compared with the half-life observed in WT mice (22). The cross-species difference in HSA/MSA mFcRn-binding affinity has been observed by Andersen et al. (3, 23), showing a ϳ25-fold increase of the K D

Engineered albumin with enhanced drug-carrier potential
of MSA binding to mFcRn (K D ϭ 0.8 M) compared with the K D of HSA binding to mFcRn (K D ϭ 25 M). This was further demonstrated in the work of Viuff et al. (15). The use of a humanised mouse model, such as the humanised albumin/ FcRn double transgenic mouse introduced in our previous work (15), has shown longer circulatory half-life of rHSA variants, however, the same tendency of a longer half-life for the rHSA variants engineered for higher FcRn affinity is also shown in the wild type model presented in the present work. Furthermore, the same binding trend HBII Ͼ HBI Ͼ WT of the singlethiol albumin variants toward hFcRn was observed for mFcRn at pH 5.5 (Table S6), which supports the use of the WT model to identify improved pharmacokinetics of HBII.
The circulatory half-life of the 3xAF680-conjugated triplethiol HBII rHSA variant was increased to 22.4 h in this present work, which is a 1.3-fold change compared with 3xAF680-conjugated triple-thiol WT rHSA. This is comparable with the 1.65-fold change observed between WT and HBII rHSA fusions in the work of Larsen et al. (16).
In our laboratory, we have used nucleic acid assemblies to multifunctionalize HSA at Cys-34 (12). These could be further utilized for the triple-thiol rHSA variants introduced in this work, to further maximize multifunctionalization. Additionally, use of the HBII rHSA triple-thiol engineered variant should allow a longer blood circulatory time. Conjugation of drugs to the three available cysteines is a future focus of our laboratory.
The potential for enhanced DAR, multifunctionality, and high hFcRn engagement promotes this new class of engineered albumins as a highly attractive drug delivery platform.

HSA variant engineering: Protein design
A range of candidate positions for replacement with cysteine were identified in surface-exposed regions of HSA. Crystal structures of HSA complexed with hFcRn are available (17,18) and were inspected using Yasara software (24). Positions that could directly impact the hFcRn binding interface when replaced with cysteine were avoided. Consideration was also given to avoid positions where introduction of a free thiol group could adversely affect the HSA structure or function or where conjugation of an additional moiety may sterically hinder hFcRn binding.

Plasmid construction
Escherichia coli 5 ␣ (New England Biolabs) was used for propagation and preparation of the thiol rHSA DNA. Expression cassettes contained Saccharomyces cerevisiae PRB1 promoter, leader sequence, HSA gene, and S. cerevisiae ADH1 terminator sequence (9). An HSA overexpression plasmid pDB5102 (Fig. S2), which contains the HSA gene codon optimized for expression in S. cerevisiae and a unique NheI site for cloning, was mutated to C34A to remove the single free thiol at position 34. Briefly, the template, pDB5102, was dam-methylated following the manufacturer's instructions (New England Biolabs); 5 ng was used in a round plasmid PCR using primers C34AForward (5Ј-TTCGCTCAATACTTGCAACAAGC-TCCATTCGAAGATCACGTCAAG-3Ј) and C34AReverse (5Ј-TTGTTGCAAGTATTGAGCGAAAGCGATCAAGAC-CAA-3Ј); and the PCR product was treated with DpnI and transformed into E. coli. The resulting plasmid, pDB5155, was used as the backbone for the construction of the singlethiol variants; the primers used to generate the variants are listed in Table S1. Expression cassettes were generated by digesting the single-thiol plasmids with NsiI and PvuI, and the required DNA fragment was purified from an agarose gel using the Qiagen gel extraction kit. Gap repair into S. cerevisiae was performed as described previously (9).

Microtiter plate (MTP) analysis
The S. cerevisiae transformants were grown in 500 l of buffered minimal media sucrose (BMMS) ϩ CSM-Leu in a 48-well MTP. The cultures were grown in a humidity chamber at 30°C and 200 rpm for 96 h. The plate was then centrifuged at 1500 rpm for 5 min, and the supernatant was removed for DTNB analysis.

Free thiol profiles
The number of free thiols on a protein can be determined using mass spectrometric analysis of the protein sample treated with Ellman's (DTNB) reagent. TNB has a molecular mass of 199 Da; thus, an increase in mass of 197 Da, when TNB loses H2 during disulfide bond formation with the free thiol group on the test protein, indicates the presence of one free thiol group on the test protein. Thus, 4 l of buffer (4 mg/ml DTNB, 500 mM sodium phosphate, pH 7.0) was added to 200 l of the test protein culture sample in a 96-well MTP format. The preparation was incubated for 25 min at ambient temperature (20 Ϯ 5°C) to allow TNB labeling. Protein intact mass was determined by UHPLC-MS. UHPLC separation was carried out on 10 l of sample using a Waters Acquity system on a BEH 50 ϫ 2.1-mm ACQUITY BEH 1.7-m 300A C4 column and a 5-min analytical gradient of buffer A (0.1% formic acid) and buffer B (100% acetonitrile, 0.1% formic acid). Eluted proteins were directly introduced to a Bruker MicrOTOF II mass spectrometer via an electrospray ionization source. Instrument control and sample acquisition was performed using BioPharma Compass TM .All data were manually processed over the leading edge of the protein peak between 2.9 -3.0 minutes within BioPharma Compass TM DataAnalysis software. This included spectral smoothing using a Gauss smoothing algorithm set at 0.0765 Da and a baseline correction setting of 0.8 flatness. Deconvoluted intact mass spectra were obtained using the Max. Entropy algorithm.

Shake flask expression and purification
BMMS medium (10 ml) was inoculated with S. cerevisiae and grown for 2 days at 30°C with orbital shaking at 200 rpm. An aliquot of each starter culture (5 ml) was used to inoculate two flasks of 200 ml of BMMS medium and grown for a further 5 days at 30°C with orbital shaking at 200 rpm. Cells were harvested by filtration through a 0.2-m vacuum filter membrane (Nalgene Sterile Top Filter), and the supernatant was retained for purification. A single-step chromatography procedure was used to prepare purified material from the thiol-rHSA variants. The purification step used a column (bed volume ϳ2 ml) Engineered albumin with enhanced drug-carrier potential packed with AlbuPure matrix (ProMetic Biosciences Ltd., Cambridge, UK). This was equilibrated with 50 mM sodium acetate, pH 5.3, and loaded with neat shake flask culture supernatants, at pH ϳ5.5-6.5. The column was washed with a ϳ10 column volumes each of 50 mM sodium acetate, pH 5.3, and 50 mM ammonium acetate, pH 8.0, respectively. Bound protein was eluted using ϳ10 column volumes of 50 mM ammonium acetate, 10 mM octanoate, pH 7.0. The flow rate throughout was 240 cm/h using an AKTA Explorer system (GE Healthcare).

Aggregation characteristic profiles
rHSA variants were analyzed for the tendency to remain as a monomer in solution. Individual rHSA variants had a single free thiol group (single-thiol) and were compared with WT HSA. The concentration and percentage monomer of the eluate samples were determined by GP-HPLC. Injections of 25 l were made onto a 7.8-mm internal diameter ϫ 300-mm length TSK G3000SWXL column (Tosoh Bioscience), with a 6.0-mm internal diameter ϫ 40-mm length TSK SW guard column (Tosoh Bioscience) at 1 ml/min, with a run time of 15 min. Alternatively, using an Agilent 1260 isocratic UHPLC instrument, injections of 4 l were made onto a 4.6-mm inner diameter ϫ 150-mm length BEH 200A, 1.7-m column (Waters), at 0.5 ml/min, with a run time of 5 min. Samples were analyzed by chromatography in 25 mM sodium phosphate, 100 mM sodium sulfate, 0.05% (w/v) sodium azide, pH 7.0. Samples were quantified by UV detection at 280 nm, by peak area, relative to a 10 mg/ml rHSA standard. The samples were reanalyzed to determine the change in percentage monomer after storage at 2-8°C.

Conjugation efficiency profiles
rHSA variants were conjugated with biotin (Thermo Scientific, EZ-Link maleimide-PEG2-biotin) using a 3.2-fold molar excess of maleimide-PEG2-biotin to protein. Eluates from the AlbuPure chromatography were diluted with PBS buffer, pH 7.4, to give 10-ml solutions at 0.3 mg/ml (45.15 nmol). Stock solutions of 2 mg/ml biotin were prepared by the addition of 5 ϫ 200-l aliquots of PBS buffer, pH 7.4, to each of two 2-mg preweighed EZ-Link microtubes, and the vials were rinsed to maximize recovery of the lyophilized product. The two 1-ml volumes were pooled into a 7-ml container with a lid. From the biotin stock solution, 38 l (144.5 nmol) was added to the 10-ml rHSA samples to give a ϳ3.2-fold molar excess of biotin over each rHSA variant. Samples were gently mixed and incubated at ambient temperature overnight. Following incubation, the samples were subjected to MS to determine the intact protein mass after conjugation according to the method described previously, but using a 15-min analytical gradient, and processing data for the protein peak between ϳ7 and 10 min.

Combining thiol variants with rHSA variants with different hFcRn affinity
The triple-thiol rHSA variant was constructed by a two-step cloning strategy. First, the single-thiol variants K93C (pDB5623) and E294C (pDB5624) were combined; this was performed by DNA fragment swaps. Both plasmids were digested with NheI and SphI, and the insert containing the E294C position was cloned into the backbone containing the K93C position, resulting in plasmid pDB5625 (C34A/K93C/E294C). For the second step of the cloning, both plasmids pDB5102 and pDB5625 were digested with PstI and SacII, and the Cys-34 insert from pDB5102 was cloned into the pDB5625 backbone, resulting in the plasmid pDB5628 (Cys-34, K93C, and E294C). Expression cassettes used for in vivo homologous recombination were generated by digesting the plasmid with NsiI and PvuI, and the required DNA fragment was purified from an agarose gel using the Qiagen gel extraction kit. S. cerevisiae transformations and albumin purifications were performed as described above. The triple-thiol rHSA variant was further modified to create expression cassettes for proteins that also exhibit modified hFcRn-binding affinity: NB, HBI, and HBII, described previously (16). S. cerevisiae transformations and albumin purification using AlbuPure matrix and diethylaminoethyl-Sepharose Fast Flow matrix (GE Healthcare) were performed as described previously (9,25).

Variant stability over long-term storage
rHSA variants at 5 mg/ml in 25 mM sodium phosphate, 215 mM sodium chloride, pH 6.5, were analyzed for monomer content over 24 months, using GP-HPLC as described earlier.

Conjugation with Alexa Fluor 680
Triple-thiol rHSA variants were further modified with AF680 for developing a detection system for in vivo studies. Variants were conjugated with AF680 dye with a maleimidereactive group (Molecular Probes), using a 3.1-fold excess of AF680-maleimide to thiol. Combination variants in 25 mM sodium phosphate, 215 mM sodium chloride, pH 6.5, were diluted with PBS, pH 7.4, to give 20-ml solutions at 1 mg/ml (300 nmol). AF680 C2-maleimide (1 mg) was solubilized in 200 l of DMSO, and 186 l (930 nmol) was added to the 20-ml rHSA samples to give an ϳ3.1 molar excess of fluorescence dye over rHSA variant. Samples were gently mixed and incubated at 2-8°C overnight, wrapped in foil to protect against light. The unreacted fluorescence-maleimide compound was removed by two cycles of concentration and dilution with PBS in Viva-Spin-20 centrifugal concentrators. The conjugation was stabilized by hydrolysis (ring-opening) of the maleimide ring (26) by the addition of 20 mM glycine buffer, pH 9.0. The hydrolysis was performed for 6 h at 37°C and terminated by neutralizing to pH 7.4. Because the overall yield after hydrolysis is ϳ50%, the material was subjected to another cycle of conjugation and hydrolysis. Following the second round of hydrolysis, unreacted and hydrolyzed fluorescence-maleimide compound was removed by ultra/diafiltration in a VivaSpin-20 centrifugal concentrator with repeated cycles of concentration and dilution with PBS, pH 7.4. The degree of labeling was determined based on the ratio of absorbance from the fluorophore at 680 nm and albumin at 280 nm using the GP-HPLC method described previously, with a dual-wavelength collection at 280 and 680 nm.

hFcRn affinity by Biolayer Interferometry
A BLItz system (Fortebio, Pall Life Sciences, Menlo Park, CA) was used to detect the affinity between selected triple-thiol rHSA variants (K93C and E294C mutations combined with WT, NB mutations, or HBII mutations) or WT control rHSA Engineered albumin with enhanced drug-carrier potential (Sigma-Aldrich, A6608) and hFcRn. Streptravidin-coated biosensors were hydrated in Dulbecco's PBS (Sigma-Aldrich, D8537) supplemented with 0.01% Tween 20 (PBST) (pH 7.4) for 10 min and incubated in 5 l of 70 nM hFcRn-biotin (Immunitrack, Copenhagen, Denmark) for 15 min followed by a wash in PBST. For kinetic measurements, 2-fold dilution series of the rHSA and triple-thiol rHSA variants were made (0.1875-3 M) and performed in PBST association buffer supplemented with 25 mM sodium acetate, 25 mM NaH 2 PO 4 , and 150 mM NaCl, pH 5.5. A 120-s association phase followed by a 300-s dissociation phase were subtracted from a baseline phase of 60 s in association buffer. Biosensors were regenerated by a 30-s wash in PBST to allow multiple use. Binding curves were fitted to a 1:1 binding model after intercept correction using global fit in the BLItz PRO version 1.2 software.
The binding of single-thiol rHSA variants (WT, HBI, and HBII) to hFcRn and mFcRn was compared using an Octet RED96 system. Biotinylated soluble hFcRn and mFcRn (Immunitrack) were immobilized on streptavidin-coated biosensors (PALL/Fortebio) in PBST. Nonspecific binding to the sensors was minimized by preblocking in albumin (diluted to 0.5 mg/ml in PBS, pH 7.4) for 5 min followed by a 5-min PBS rinse, pH 7.4, and then MilliQ water rinse. The sensors were kept in storage by prior soaking in 15% sucrose and air-dried. For kinetic characterization, a seven-step 2-fold dilution series was prepared in association buffer, pH 5.5. Binding kinetics were performed at 30°C with a 120-s association phase and 300-s dissociation phase. The sensors were regenerated with PBST, pH 7.4, and equilibrated between samples in association buffer, pH 5.5. All data were referenced with FcRn-streptavidin sensors in association buffer without HSA. The Octet data analysis software version 8.0 (PALL/Fortebio) was used for data analysis using curve fitting to a 1-1 model for estimation of kinetic parameters.

Animals
Twenty-four 8 -12-week-old female BALB/cAnNRj mice were purchased from Janvier Labs (Le Genest-Saint-Isle, France). All mouse experiments were performed at the biomedical animal facilities of Aarhus University. All animal studies were approved and performed according to the guidelines from the Danish Animal Experiments Inspectorate under the auspices of the Danish Ministry of Environment and Food.

Blood circulation time of Alexa Fluor 680 rHSA variants
Mice were intravenously injected in the tail vein with the 2 mg/ml triple-thiol rHSA (K93C and E294C mutations combined with WT, NB, or HBII mutations) variants (10 mg/kg) bearing three AF680 molecules each (n ϭ 7). A PBS-injected control group was also included. Blood samples (15 l) were taken after 1 min from vena sublingualis and from the tail vein at 30 min, 4 h, 24 h, 48 h, 72 h, and 96 h and placed into heparincoated 20-l capillaries. Blood samples were centrifuged and then scanned using Xenogen IVIS (PerkinElmer Life Sciences). Spectral unmixing was performed to subtract autofluorescence (from the control plasma) from the AF680 signal on the rHSA variants using Living Image version 4.4 (PerkinElmer Life Sciences). Signal intensity from plasma was obtained from regions of interest (ROIs) using equivalent area selection and surface radiance within the ROIs measured as the number of photons per second that leave a square centimeter of tissue and radiate into a solid angle of one steradian (photons/s/cm 2 /sr). Surface radiance from plasma from each time point was normalised to the surface radiance from the 1 minute plasma sample for each mouse.
Author contributions-C. J. A. F., J. C., K. B., K. N., and K. A. H. conceived the research and planned the studies; K. N., H. R., M. W., B. A., K. K. S., and F. D.-H. performed experiments and analyzed and interpreted data. K. A. H., K. K. S., K. N., and H. R. drafted the manuscript. All authors read and approved the final manuscript.