key: cord-0020856-5eojx11o
authors: Shen, Zhuolun; Xiang, Yufei; Vergara, Sandra; Chen, Apeng; Xiao, Zhengyun; Santiago, Ulises; Jin, Changzhong; Sang, Zhe; Luo, Jiadi; Chen, Kong; Schneidman-Duhovny, Dina; Camacho, Carlos; Calero, Guillermo; Hu, Baoli; Shi, Yi
title: A resource of high-quality and versatile nanobodies for drug delivery
date: 2021-08-21
journal: iScience
DOI: 10.1016/j.isci.2021.103014
sha: 7ab9972034668711d0a217e6c460bbfbe55e76c8
doc_id: 20856
cord_uid: 5eojx11o

Therapeutic and diagnostic efficacies of small biomolecules and chemical compounds are hampered by suboptimal pharmacokinetics. Here, we developed a repertoire of robust and high-affinity antihuman serum albumin nanobodies (Nb(HSA)) that can be readily fused to small biologics for half-life extension. We characterized the thermostability, binding kinetics, and cross-species reactivity of Nb(HSA)s, mapped their epitopes, and structurally resolved a tetrameric HSA-Nb complex. We parallelly determined the half-lives of a cohort of selected Nb(HSA)s in an HSA mouse model by quantitative proteomics. Compared to short-lived control nanobodies, the half-lives of Nb(HSA)s were drastically prolonged by 771-fold. Nb(HSA)s have distinct and diverse pharmacokinetics, positively correlating with their albumin binding affinities at the endosomal pH. We then generated stable and highly bioactive Nb(HSA)-cytokine fusion constructs “Duraleukin” and demonstrated Duraleukin's high preclinical efficacy for cancer treatment in a melanoma model. This high-quality and versatile Nb toolkit will help tailor drug half-life to specific medical needs.

Chemical compounds and small biomolecules (<50 kDa) are known to be rapidly eliminated in the plasma (from min to h) by glomerular filtration. Such poor pharmacokinetics has greatly limited the efficacy of numerous clinically important peptides and small proteins including hormones, cytokines, coagulation factors, and antibody fragments such as nanobodies (Nbs). Various strategies have been developed to improve the half-lives of the target biomolecules. These include chemical modifications by PEGylation, fusion with an antibody Fc domain, and piggy-back delivery by targeting serum albumin (Czajkowsky et al., 2012; Harris and Chess, 2003; Kratz, 2008; Larsen et al., 2016) . Human serum albumin (HSA) is a highly abundant blood protein with a long plasma half-life of approximately three weeks (Larsen et al., 2016) . HSA has a molecular weight of $67 kDa and is refractory to renal filtration. Moreover, it can bind to the neonatal Fc receptor (FcRn) that is ubiquitously expressed in various cell types. This interaction requires the endosomal pH (<pH 6.5) and is critical for the FcRn-mediated transcytosis of albumin that prevents its endolysosomal degradation to substantially improve albumin's half-life (Merlot et al., 2014; Sand et al., 2015) . Although HSA is the most abundant serum protein, its intracellular concentration is likely very low (Merlot et al., 2014; Peters, 1996) .

The high serum concentration and the exceptionally long in vivo half-life make HSA an excellent trojan horse for drug delivery. Different albumin-based approaches have been developed including direct fusion to HSA and development of albumin nanoparticles as well as albumin-binding domains (ABDs) derived from grampositive bacteria to improve the pharmacokinetics of target therapeutics (Larsen et al., 2016; Merlot et al., 2014; Sleep et al., 2013) . Despite these advances, however, successful preclinical and clinical applications of these technologies remain limited (Kontermann, 2016) . Development of robust, highly specific and costeffective new HSA binders, ideally with inherent low immunogenicity, high bioavailability, and versatility in modulating pharmacokinetic properties based on specific medical needs remains highly desirable.

Nbs are robust and small antigen-binding fragments ($15 kDa) that are derived from camelid's heavychain-only antibodies (HcAbs) (Hamers-Casterman et al., 1993) . They are characterized by marked physiochemical properties including high solubility and stability (Xiang et al., 2020a (Xiang et al., , 2021 . Nbs can be rapidly produced in microbes such as Escherichia coli and yeast cells at low costs. Because of the small size and robustness, Nbs have fast tissue distributions and high bioavailability for in vivo therapeutic and diagnostic applications (Jov cevska and Muyldermans, 2020; Kijanka et al., 2015) . Monomeric Nbs can be easily bioengineered into multivalent forms to improve binding affinities and bioactivities (Fridy et al., 2014; Harmsen and De Haard, 2007; Xiang et al., 2020a Xiang et al., , 2021 . Importantly, camelid Nbs have high sequence and structural similarity with human IgG heavy chains (Kijanka et al., 2015; Vincke et al., 2009) . To date, over 35 clinical trials have been conducted on more than 1,000 patients and healthy volunteers, revealing high biosafety of Nbs in humans. In addition, the first Nb drug (caplacizumab) has recently been approved by the FDA (Scully et al., 2019) . Stable Nbs are flexible for administration including direct inhalation therapy by Nb aerosols, making their use against respiratory viruses very appealing (Cunningham et al., 2021; Larios Mora et al., 2018; Liang et al., 2020; Nambulli et al., 2021; Patton and Byron, 2007; Van Heeke et al., 2017) .

Recently, we have developed a robust proteomic technology for large-scale identification and affinity classification of drug-quality Nbs (28). Using this pipeline, we identified thousands of high-affinity anti-HSA Nbs (Nb HSA ) (Xiang et al., 2021) . In this study, we produced and systematically characterized a repertoire of structurally diverse and sub-nM affinity Nb HSA s for drug delivery. We developed a multiplexed pharmacokinetic assay by quantitative proteomics and determined the half-lives of dozens of selected Nb HSA s in an Alb À/-FcRn-humanized mouse model (B6.Cg-Tg(FCGRT)32Dcr Alb em12Mvw Fcgrt tm1Dcr /MvwJ) (Roopenian et al., 2015) . Our analysis reveals that Nb HSA s have distinct and diverse pharmacokinetic profiles with highly extended half-lives ranging from 1.6 to 7.6 days in the HSA model. The half-lives of Nb HSA positively correlate with their albumin binding affinities at the mildly acidic pH. To demonstrate the utility of our Nb HSA s, we generated ''Duraleukin" by fusing several cross-species Nb HSA s to human interleukin-2 (hIL-2) -a small anticancer cytokine whose clinical efficacy has been substantially compromised by its small size and short serum half-life. Duraleukin can be rapidly produced from E. coli cells and have excellent stability and bioactivity. Using a melanoma mouse model, we demonstrated the high preclinical efficacy of Duraleukin for cancer treatment. This versatile and well-characterized Nb resource is shared with the research community to help advance biomedical discoveries into therapeutic development.

We hypothesize that the in vivo half-lives of Nb HSA s are related to their biophysical (such as binding kinetics, solubility, stability, etc.) and structural properties (Ovacik and Lin, 2018) . To maximize these properties, we included a collection of $90 Nb HSA s with highly diverse complementarity-determining region (CDR) 3 sequences, which form hypervariable loop structures (fingerprints) for specific antigen binding. As shown in Figure 1B , the CDR3s of these highly selected Nbs vary substantially in amino acid composition, sequence length, isoelectric point (pI), and hydrophobicity. To produce recombinant Nbs, their amino acid sequences were reverse translated into DNA fragments ($350 bp), which were subsequently synthesized and cloned into a his6-tag vector (pET-21) for recombinant protein production in E. coli. As expected, the majority of Nbs (71/90) were highly soluble and can be rapidly purified by the His-cobalt affinity resin ( Figure 1A ; Data S1).

To assess binding and cross-species reactivity of our Nb HSA s, we performed the enzyme-linked immunosorbent assay (ELISA) for serum albumins from five different species including human, nonhuman primate (NHP, cynomolgus monkey), rodent (mouse), bovine (cow), and camelid (llama). Optical density (O.D.) values were (G) Validation of Nb cross-species reactivity by immunoprecipitation. Three Nbs (Nb 143 , Nb 85, and Nb 132 ) were immunoprecipitated by affinity resins coupled with albumin of different species, including human, monkey, mouse, bovine, and llama. (H) The plot of the thermostability of 20 Nb HSA selected for measurement by differential scanning fluorimetry.

iScience 24, 103014, September 24, 2021 3 iScience Article calculated from the ELISA curves to generate a heat map that summarizes the binding results ( Figure 1C ). Our data confirm that all selected Nb HSA s can bind strongly to HSA. Around three quarters (53/71) of Nb HSA s bind monkey albumin and 11% of them (8/71) can also interact with mouse albumin. These results likely reflect the high sequence similarity (92%) between HSA and cynomolgus monkey albumin and relatively poor residue conservation (78%) between HSA and mouse albumin. Interestingly, we identified six Nb HSA s that bind all ( Figure 1C ; STAR methods) but llama and bovine albumins ( Figure 1G ). These Nbs may target relatively conserved epitopes specifically shared among HSA, NHP, and mouse albumins. C ross-species Nbs will be highly useful in the translation of preclinical studies in the animal models into clinal trials.

To measure the binding kinetics of Nb HSA s, we selected 20 Nb HSA s for surface plasmon resonance (SPR) analysis ( Figures 1C, 1D , and S1). Three quarters (15/20) of Nb HSA s bind extremely tight to HSA with subnanomolar affinities. The remaining Nb HSA s span single digit to hundreds of nanomolar affinities for HSA binding ( Figure S1 ). Consistent with the previous report (Heinrich et al., 2010) , we found that ELISA O.D. correlates well with the SPR affinity (Pearson r = 0.78, p < 0.001) ( Figure 1E ) and can be used to approximate binding affinity.

Next, we assess the thermostability of a cohort of representative Nb HSA s by differential scanning fluorimetry (Niesen et al., 2007) . While Nbs were produced from E. coli whole-cell lysis, they remain stable with melting temperatures (T m ) ranging between $39 C and 70 C ( Figures 1H and S2 ; Data S2). The thermostability could be further improved by using periplasmic preparations (Pleiner et al., 2015) . In summary, our highly selected Nb HSA s are characterized by marked physicochemical properties such as high solubility and stability, which are critical for drug development, storage, and delivery (Carter, 2006; Holliger and Hudson, 2005) .

Epitope mapping can provide insights into how Nbs uniquely bind the target with high affinity and selectivity and may guide the rationale design of therapeutics such as chemical compounds and small peptides/ proteins for HSA binding (D Brooks, 2014; Flyak et al., 2018; Nivarthi et al., 2017) . To systematically map Nb epitopes, we used a hybrid approach that integrates chemical cross-linking/mass spectrometry (CXMS) and structural modeling for high-throughput structural analysis. CXMS identifies short cross-links at amino acid resolution, which can be used as spatial restraints (molecular rulers) to calculate structural models to help map binding interfaces (Chait et al., 2016; Rout and Sali, 2019; Russel et al., 2012; Shi et al., 2014 Shi et al., , 2015 Xiang et al., 2020b; Yu and Huang, 2018) .

Thirty-eight HSA-Nb complexes were reconstituted in vitro and cross-linked by disuccinimidyl suberate, which can reach up to 30 Å between two cross-linked amines (e.g., lysine residues). After cross-linking, the complexes were separated by SDS-PAGE, in-gel digested, and the resulting cross-linked peptides were analyzed by liquid chromatography coupled to mass spectrometry (LC/MS) (STAR methods). A total of 125 intersubunit cross-links were confidently identified by MS (Data S3) and were used to compute the models (STAR methods). 29 HSA-Nb complex models converged with an average root mean square deviation (RMSD) of 7.4 Å ( Figure 2C ). The majority (87.5%) of cross-links were satisfied by the models ( Figure S3B ).

Previously, X-ray crystallographic analysis revealed three domains of HSA (I, II, and III). Each domain is composed of two subdomain structures. In addition, domains II and III are characterized by a small pocket composed of hydrophobic and positively charged residues that allow binding of specific metabolites and chemical compounds (Sudlow's sites) (Ghuman et al., 2005; Sudlow et al., 1975; Sugio et al., 1999) . Our analysis identified six distinct Nb epitope clusters (clusters A to F) ( Figure 2A ; Data S3). More than three-quarters of Nbs bind two major epitope clusters (A: 31% and E: 44.8%) ( Figure 2B ). The rest (B, C, D and F) bind four minor clusters. Moreover, cluster A Nbs bind epitopes between domains I and II of HSA. Cluster B also targets these domains but localizes on the opposite side of A. Cluster C binds domain I, while the other three epitope clusters (D, E, and F) mainly target domain III. Noticeably, two dominant clusters of A and E overlap substantially with Sudlow's site I and II, respectively. Moreover, our models indicate that highaffinity Nbs mainly bind concave epitopes ( Figure 2D ). Critically, most of our Nb HSA s do not target the major crevice of HSA which is essential for FcRn binding ( Figure S3A ). The only exception is Nb 126 , which based on our structural model, can interfere with the HSA-FcRn interaction and therefore may have inferior pharmacokinetics. Next, we used size-exclusion chromatography (SEC) to verify our structural proteomic results. Three Nbs (Nb 13 , Nb 29, and Nb 80 ) which bind distinct epitopes (A, C, and E, respectively) were reconstituted in complex with HSA. The resulting complexes (HSA-Nb 80 , HSA-Nb 80 -Nb 29 , and HSA-Nb 80 -Nb 29 -Nb 13 ) of different molecular weights were subsequently analyzed by SEC. As shown in Figure 3A , our analysis reveals that these Nbs do not compete with each other for HSA binding and do not share overlapping epitopes, which is consistent with epitope mapping by structural proteomics.

To obtain detailed structural information of Nb binding, we collected a homogenous fraction of the tetrameric complex composed of HSA and three Nbs for negative stain electron microscopy (EM) analysis. $22,000 EM particles were subsequently selected to reconstruct the 3D density map ( Figure S4 , STAR methods). The EM map fits well the cross-linking model ( Figure 3B ). Moreover, compared to Nb 29 of relatively lower affinity (K D = 1.2 nM), we found that ultrahigh-affinity Nbs of Nb 80 (K D = 166 pM) and Nb 13 (K D = 201 pM) bind more concave epitopes ( Figures 3C-3E ). To further validate this integrative structure, we performed site-directed mutagenesis followed by immunoprecipitation to evaluate the binding of Nb 80 to two critical HSA residues (K383 and E400). Our model predicts that these residues form salt bridges with the CDR residues of Nb 80 (K383-D60 and E400-R107, Figure 3F ). Therefore, simple charge reversals should disrupt the covalent bonds to reduce the ultrahigh-affinity of Nb 80 for HSA binding. Consistent with this model, we found that a single-point mutation E400R can substantially weaken Nb 80 binding. The binding was largely abolished by the double-residue mutations (K383D;E400R) that reversed two critical charges ( Figure 3G ).

ELISA has been extensively used to quantify the pharmacokinetics of biologics. This method requires the use of mAbs for specific, accurate, and reproducible quantifications of low-abundance signals in the complex serum background. However, high-quality mAbs are often difficult to obtain and may have large batch-to-batch variations (Baker, 2015; Sakamoto et al., 2018; Weller, 2016) . The challenges are exacerbated for high-throughout pharmacokinetic analysis, which requires the use of a large number of laboratory animals (minimally, three animals per molecule such as Nb). The experiments can be highly laborious.

To alleviate these potential issues and to minimize experimental biases, we developed a specific, sensitive, and paralleled pharmacokinetic assay based on parallel reaction monitoring or PRM mass spectrometry (MS) (Gallien et al., 2012; Peterson et al., 2012) . For proof-of-concept, we chose four benchmark Nbs for assay development. Purified Nbs were spiked in the llama serum. The protein mixtures were then efficiently proteolyzed in solution by endoproteases such as trypsin to generate distinct Nb peptides that represent Nb signals. Desalted peptides were subsequently analyzed by LC coupled to label-free MS. The unique peptides and their specific fragment ions were selected for quantification by a high-resolution Orbitrap mass analyzer. Our analysis shows that the assay is sensitive and accurate. Less than 3 nM of Nbs in the llama serum can be detected with excellent quantification linearity (R 2 = 0.999) ( Figure S5 ; Data S4).

To facilitate the pharmacokinetic analysis of Nb HSA (most do not bind mouse albumin), we used a humanized mouse model (B6.Cg-Tg(FCGRT)32Dcr Alb em12Mvw Fcgrt tm1Dcr /MvwJ) (Roopenian et al., 2015) to approximate the half-life and physiology of HSA in humans. In this model, mouse Alb is removed, and Fcgrt is replaced with a human ortholog. HSA can be later administered in this model to study the pharmacokinetics of HSA-based drugs. These transgenic animals are healthy and fertile, consistent with the previous observations (Low and Wiles, 2016; Roopenian et al., 2015) .

Following the HSA injection, 22 Nbs were mixed with equal molarity in PBS and were administered to the animals (n = 3) via single-bolus intravenous (i.v.) injections (STAR methods). This mixture includes 20 wellcharacterized and highly diverse Nb HSA s and two control Nbs that do not bind HSA. These Nb HSA s can bind albumin at the mildly acidic pH and span a wide range of binding affinities. Analysis of these Nbs may help better understand the contribution of binding at the endosomal pH to their in vivo pharmacokinetic iScience Article properties (van Faassen et al., 2020) . The molar ratio between Nbs and HSA was 1:5 to limit potential competitions among Nbs that target the same epitopes. Blood was collected at 17 different time points ranging from predose to 21 days postinjection (STAR methods). Serum samples were efficiently proteolyzed, and the diagnostic Nb HSA peptides and the fragment ions were accurately quantified as described above (Figure 4A ; Data S4). The quantifications were highly reproducible and the median coefficient of variation (CV) of three biological replicates was 15.3% ( Figure S6 ).

To measure pharmacokinetics, the average relative abundance of Nbs was calculated and fitted into a twophase decay model to derive the half-lives (STAR methods). Our results reveal that control Nbs were rapidly cleared from the serum, with the half-lives of approximately 26-60 min, which are consistent with the previous ELISA measurements (Hoefman et al., 2015) . In comparison, Nb HSA s were substantially more stable in the serum. Nb HSA s are generally characterized by fast distribution rates followed by slow elimination rates. However, each Nb HSA has a unique pharmacokinetic profile. The distribution half-lives span between 1.0 h (Nb 80 ) and 30.5 h (Nb 85 ). The elimination half-lives range from 1.6 days (Nb 100 ) to 7.6 days (Nb 75 ), which correspond to approximately 771-fold improvement compared to the control Nb ( Figures 4B and 4C ; Data S2). The half-lives of Nb HSA s agree well with the relatively short half-life of HSA ($8 days) that we determined in this model ( Figure S7C ). Nevertheless, we expect that their pharmacokinetics will be substantially improved when applied to humans.

Serum proteomes contain rich molecular information regarding clinical manifestations and drug treatment (Durrum et al., 1956) . To explore the potential serum proteomic alterations post-HSA administration and to understand the relatively fast half-life of HSA in this mouse model (8 days versus $3 weeks in humans), we quantified the longitudinal blood proteomes by label-free quantitative proteomics (Anderson, 2010; Geyer et al., 2017) .

A total of 441 blood proteins were quantified by MS for 10 days post-HSA administration. The quantified proteins were then clustered based on their relative abundance changes to generate the heatmap (Figure S8) . Our analysis revealed a group of acute-phase inflammatory proteins, such as serum amyloid A and C-reactive protein, were drastically increased by hundreds of folds in 12-24 h postinjection, before returning to the baseline levels in 24-48 h (Sproston and Ashworth, 2018; Uhlar and Whitehead, 1999) . This observation may be related to the acute response to i.v. injection and frequent blood sampling within the first 24 h or potential inflammations caused by administration of a large quantity of exogenous protein (HSA). In addition, we detected an increase of serum IgGs 5-7 days postadministration, which indicates a potential antibody response against the exogenous HSA and may explain the faster-than-expected clearance of HSA in this model (Schmidt et al., 2013) . Further studies will be needed to evaluate these changes during Nb treatment.

To understand the diverse pharmacokinetics of Nb HSA s better, we calculated the correlations of half-lives (including both elimination and distribution rates) with different biophysical and physicochemical properties such as HSA binding (at both neutral and mildly acidic pH), melting temperature (T M ), pI, and hydrophobicity. The resulting heatmaps were shown in Figure 4D . We found that 1) elimination and distribution rates of Nbs are positively correlated (p = 0.018), 2) compared to low/mediocre-affinity Nbs, high-affinity Nbs (sub-nM) have slower clearance rates, 3) the elimination half-lives of Nbs are positively correlated with binding affinities (ELISA O.D.) at the endosomal pH (p = 0.003), and 4) elimination seems to be positively correlated with the binding off-rates at neutral pH (p = 0.006), which may contribute to high-affinity binding at low pH. However, we did not observe any significant correlation between the elimination and binding affinity at neutral pH when both on-and off-rates are considered. Conceivably, both slow off-rates and binding at the mildly acidic pH are required to enable stable complex formation of the low-abundance, intracellular Nb HSA -HSA-FcRn complex. Nb HSA that do not bind strongly to the complex have reduced in vivo half-life. In addition, Nb structures and epitopes may also contribute to the half-lives. For example, an ultrahigh-affinity 

Interleukin-2 (IL-2) is a cytokine that can modulate immune cells such as T cells and natural killer cells to inhibit tumor growth (Boyman and Sprent, 2012). Human IL-2 (hIL-2) is the first FDA-approved cancer iScience Article immunotherapy drug to treat renal cell carcinoma and metastatic melanoma (Rosenberg, 2014) . High-dose hIL-2 (proleukin) is effective against these devastating cancers, and $10% of patients who receive the treatment can expect a complete response. However, clinical efficacy of this cytokine-based immunotherapy is substantially limited by the small size ($15 kDa) and the poor pharmacokinetics of hIL-2 (elimination half-life of 85 min in humans) (Boyman and Sprent, 2012; Rosenberg, 2014) . High-dose and repetitive administration (every 8 h), which are necessary for the sustained antitumor activity and clinical efficacy, can lead to severe side effects such as vascular leak syndrome and liver toxicity (Rosenberg, 2014; Schwartz et al., 2002) .

Here, we developed Nb-fusion constructs to improve the antitumor efficacy of hIL-2 while potentially reducing the clinical side effects associated with high dose. Specifically, we generated constructs by genetically fusing several cross-species albumin Nbs (Nb 77 , Nb 80, or Nb 158 ) N-terminal to hIL-2 ( Figure 5A ). A short flexible GS linker was inserted between IL-2 and the Nb. Based on structural models, this design will ensure that 1) the N-terminal fusion Nb does not sterically interfere with the hIL-2 binding to its receptor IL-2R and 2) hIL-2 does not block Nb binding to the FcRn receptor ( Figure S9A ). The resulting constructs ''Duraleukins'' were purified for in vitro and in vivo evaluations.

As shown in Figure 5B , we developed a protocol to isolate Duraleukin (DL 77 , DL 80, and DL 158 ) from E. coli inclusion bodies (STAR methods). Recombinantly purified DLs demonstrated excellent thermostability (T m z 60 C, Figure 5C ) and remained highly stable when incubating at 37 C for days with fresh mouse sera that contain protease activities ( Figure 5E ). Critically, DLs had comparably high bioactivities to hIL-2 in vitro in stimulating the proliferation of CD8 + T cells, as shown by the T cell proliferation assay (STAR methods; Figure 5D ). Moreover, DLs retained ultrahigh HSA binding affinities ( Figure 5E ). Our results demonstrated the robust physicochemical properties and high bioactivities of Duraleukin. DL 80 was used for preclinical evaluations in an animal model. While the selection was primarily based on its fast distribution rate, high affinity, and partial resistance to acidic pH ( Figure 5G ), other DLs also have excellent stability and bioactivities that may translate into high preclinical efficacy as well.

To evaluate the preclinical efficacy of Duraleukin for cancer treatment, we generated a B16-F10 melanoma mouse model in the C57BL/6J background (Overwijk and Restifo, 2001) . First, we confirmed the prolonged half-life of DL 80 ($46-fold improvement compared to hIL-2) in this model ( Figure S9B ). Next, tumor-bearing mice were randomly grouped into three arms and were intraperitoneally administered with DL 80 , hIL-2, or PBS control for 24 days. The hIL-2 treated mice were dosed daily to counteract the limited bioactivity imposed by the short half-life of hIL-2. In comparison, the DL 80 -treated arm was dosed every six days to evaluate the impact of half-life extension. In addition, a monoclonal antibody, TA99 was cotreated with both therapeutic agents. TA99 recognizes a melanoma surface marker TRP1 and has been shown to synergize with hIL-2 for improved anticancer activity in the mouse model (Moynihan et al., 2016; Silva et al., 2019; Sun et al., 2019; Zhu et al., 2015) (STAR methods). The tumor sizes and the survival of the treated animals were continuously monitored for up to 62 days before termination of the project (Figures 6A and 6B ). Compared to the PBS control animals, both DL 80 -and hIL-2-treated animals showed substantial reductions in tumor loads and significantly improved survival rates. In addition, we found that animals responding to DL 80 were almost entirely free of tumors, whereas rebounds of tumorigenesis in the hIL-2 treated mice were evident. Here, despite the low administration frequency and dose (as little as 13% of the hIL-2 dose), the high in vivo stability and sustainable anticancer activity of DL 80 may ultimately lead to the improved recovery of the treated animals.

To explore the mechanisms that underlie Duraleukin's high preclinical efficacy, we isolated the tumors three days post-treatment, sorted, and analyzed tumor-infiltrating immune cells by flow cytometry ( Figures  6C and S10 ). We observed significant and more persistent increases of tumor-infiltrating immune cells, including both CD8 + T cells and natural killer cells in the DL 80 -treated tumor tissues compared to controls. Collectively, these results suggest that Duraleukin can consistently promote tumor-infiltrating immune cells to obtain highly sustainable antitumor activities.

In this study, we have characterized a large collection of high-quality HSA Nbs for drug delivery. Using multidisciplinary approaches that span biophysics, structural proteomics, and modeling, we systematically mapped HSA epitopes and resolved the structure of a tetrameric HSA-Nb complex. Our analysis reveals that most of ll OPEN ACCESS iScience 24, 103014, September 24, 2021 iScience Article Nbs bind to HSA at sub-nanomolar affinities, which are unusual for single-domain antibody fragments (Xiang et al., 2021) . Several Nbs also bind strongly with monkey and mouse albumins which could be useful for efficacy studies in these preclinical models. Moreover, our systematic structural investigations have identified multiple Nb epitope classes on HSA. This information has provided a structural basis to better understand Nb binding and may help future design of novel therapeutics to target HSA as an important drug target. (C) Thermostability of Duraleukins by differential scanning fluorimetry. (D) In vitro CTLL-2 cell proliferation assay of Duraleukin and hIL-2. The X axis is the concentration of hIL-2 (ng/mL). The Y axis is the absorbance at 450 nM subtracted by absorbance at 690 nm (background signal) for the measurement of T cell proliferation.

(E) The resistance of three DLs to serum proteases. Purified DLs were incubated with the mouse serum for the indicated periods. After incubation, DLs were detected by western blot using an anti-IL-2 monoclonal antibody (BG5). High-quality Nb HSA s may offer several major advantages over the Fc-based methods for drug delivery.

Firstly, compared to Fc ($50 kDa), Nbs are substantially smaller ($15 kDa). Therefore, Nb-fusion constructs have higher molar concentrations per administration (and effective therapeutic dose) than the Fc-fusion counterparts. Secondly, Nbs have a robust fold to facilitate bioengineering and retain the bioactivities of the modified targets (Larsen et al., 2016; Sleep et al., 2013) . Affinity-matured Nbs are characterized by high stability that allows for direct inhalation therapy by Nb aerosols for the treatment of pulmonary infections highly efficiently (Larios Mora et al., 2018; Liang et al., 2020; Nambulli et al., 2021; Patton and Byron, 2007) . Their marked physiochemical properties will also be critical for drug storage, administration, and delivery (Holliger and Hudson, 2005) . Finally, Nbs do not recruit functional immune cells as the Fc domain of antibody does. While this may not be ideal for applications in which the Fc-mediated effector's functions are required, it will be useful when elevated immune responses (e.g., against the IL-2R + cells) should be minimized (Hassanzadeh-Ghassabeh et al., 2013; Kang and Jung, 2019; Steeland et al., 2016) . In addition, for the neutralization of viruses, Fc has been associated with antibody-dependent enhancement (ADE) which may increase the risk of exacerbating disease severity (Katzelnick et al., 2017; Polack et al., 2003) . In these cases, it might be necessary to mutate Fc to abolish its associated immune responses.

To demonstrate the utility of this resource, we have generated IL-2-fusion constructs (Duraleukin) using high-quality, cross-species Nb HSA s that help translate the preclinical results into clinical trials. Critically, despite using a substantially lower administration dose, Duraleukin outperformed hIL-2 for melanoma treatment. The improved efficacy is likely the results of enhanced half-life of Nb HSA -modified hIL-2 and the accumulation of tumor-infiltrating immune cells. Presumably, it is also related to the high local concentration of albumin in the tumor interstitium and high bioactivity of Duraleukin in the tumor tissue (Finicle et al., 2018; Greish, 2007; Merlot et al., 2014) .

This proof-of-concept study is readily extended to other small antitumor cytokines (Steel et al., 2012), ultimately leading to enhanced antitumor activity, low-dosing regimen, and improved safety profiles for human applications. It is conceivable that a combination of Duraleukin and an immune checkpoint blockade therapy like PD-1/L1 or CTLA-4 may further synergistically improve cancer treatment (Khair et al., 2019; Ring et al., 2013) . It is also feasible to quickly generate a trifunctional construct by fusing Duraleukin with another Nb that specifically recognizes a tumor surface marker such as EGFR or HER2 for a more target-specific therapy (Hirsch et al., 2009) . Future studies will be needed to investigate Nb HSA and Duraleukin in the preclinical and clinical settings.

In this study, we have determined the pharmacokinetics of highly selected albumin Nbs in an Alb À/-FcRnhumanized mouse model. In this model, we found that the half-lives of Nb HSA s are drastically improved compared to control Nbs. However, these transgenic animals may or may not fully recapitulate the physiology of albumin and its transcytosis biology in humans. Further investigations in NHPs and preferentially humans would help more accurate determination of the half-lives of our Nb HSA s for clinical development.

Moreover, in the current study, we did not investigate the optimal dosing of Duraleukin for tumor growth inhibition. It would also be interesting to assess the efficacy of Duraleukin in renal cell carcinoma models.

Detailed methods are provided in the online version of this paper and include the following: iScience Article Coppieters, K., Dreier, T., Silence, K., de Haard, H., Lauwereys, M., Casteels, P., Beirnaert, E., Jonckheere, H., Van de Wiele, C., Staelens, L., et al. (2006) . Formatted anti-tumor necrosis factor alpha VHH proteins derived from camelids show superior potency and targeting to inflamed joints in a murine model of collagen-induced arthritis. Arthritis Rheum. 54, 1856 -1866 . Cox, J., and Mann, M. (2008 

Amino acid sequences of nanobodies generated in this paper are available in Data S1. Request for HSA mutant plasmids should be fulfilled by the lead contact Dr. Yi Shi.

Data and code availability d Mass spectrometry data of this study have been deposited in MassIVE and are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

d This paper does not report original code.

d Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

All mouse strains used in this study were obtained from The Jackson Laboratory. B6.Cg-Tg(FCGRT)32Dcr Alb em12Mvw Fcgrt tm1Dcr /MvwJ (JAX #025201) mice were fed on LabDietâ 5K20. 2 male and 1 female mice used in this study were aged 25 weeks before the experiments. Gender is not considered as a confounding factors in this study since grouping is not involved. C57BL/6J (JAX #000664) mice were fed on LabDietâ 5K52. All C57BL/6J mice used in this study were male and aged 8 weeks before the experiments. All animal experimentation was approved by the Institutional Animal Care and Use Committee (IACUC) at University of Pittsburgh.

All cell lines used in this study were obtained from ATCC and cultured according to the provided protocols. The complete culture medium of CTLL-2 cell line was formulated as 10% FBS, an additional 2 mM L-glutamine, and 1 mM sodium pyruvate in ATCC-formulated RPMI-1640 containing 10% T-STIM with Con A (Corning). The complete medium of B16-F10 cell line is formulated as 10% FBS in ATCC-formulated DMEM.

Nb genes were codon-optimized for expression in Escherichia coli and the nucleotides were in vitro synthesized (Synbio). The Nb genes were cloned into a pET-21b (+) vector at BamHI and XhoI restriction sites, or a pMAL-c5X vector at BamHI and EcoRI restriction sites. iScience Article NaCl, 0.2% TX-100 with protease inhibitor). After lysis, soluble protein extract was collected at 15,000xg for 10 min, and recombinant his-tagged Nbs were purified by His6-Cobalt resin (Thermo) and eluted by imidazole; while MBP-fusion Nbs were purified by amylose resin (NEB) and eluted by maltose. The eluted Nbs were subsequently dialyzed in the dialysis buffer (e.g., 1x DPBS, pH 7.4). A HiLoad 16/600 Superdex 200 pg column (GE Healthcare) on an Ä KTA FPLC protein purification system (GE Healthcare) was used to further purify the Nbs and ensure their monomeric states. The purified Nbs were analyzed by SDS-PAGE. For the animal experiments, excessive endotoxin in the Nbs was removed (below 0.1 EU/dose) by the ToxinEraser Endotoxin Removal column (Genscript) and measured by ToxinSensor Chromogenic LAL Endotoxin Assay Kit (Genscript). The purified Nbs were further sterilized by passing a 0.2 mm filter (Millex) and were stored at À80 C before use.

Indirect ELISA was carried out to measure Nb-albumin bindings. Albumin was coated onto a 96-well ELISA plate (R&D system) at 1-10 ng/well in coating buffer (15 mM sodium carbonate, 35 mM sodium bicarbonate, pH 9.6) overnight at 4 C and was blocked with a blocking buffer (DPBS, 0.05% Tween 20, 5% milk) at room temperature for 2 hrs. Nbs were diluted in the blocking buffer and incubated with albumin for 2 hrs. HRP-conjugated secondary antibodies against His-tag (Genscript) were diluted 1:5,000-10,000 and incubated with the well for 1 h at room temperature. After PBST (DPBS, 0.05% Tween 20) washes, the samples were further incubated under dark with freshly prepared w3,3 0 ,5,5 0 -Tetramethylbenzidine (TMB) substrate for 10 min to develop the signals. After the STOP solution (R&D system), the plates were read at multiple wavelengths (the optical density at 550 nm wavelength subtracted from the density at 450 nm) on a plate reader (Multiskan GO, Thermo Fisher). The relative affinity of each Nb was determined by the average of O.D. readouts. For human and monkey albumin, the relative affinity of each Nb was tested at Nb concentrations of 1 mM, 100 nM, 10 nM, 1 nM, and 100 pM. An average ELISA O.D. of >0.1 was used as the cutoff value to assign albumin binding Nbs. For mouse albumin, relative affinity was tested at one mM and a cutoff O.D. of >1 was used.

In vitro pull-down assay (immunoprecipitation)

Nb or serum albumin was coupled to CNBr-activated sepharose resin (GE Healthcare). For the in vitro pulldown assay, different concentrations of Nbs were incubated with the human serum albumin coupled resin. Samples were incubated for 15-30 min at 4 C with gentle agitation. The resin was collected and washed three times with a washing buffer and was boiled in the LDS sample loading buffer (Thermo) before SDS PAGE analysis.

Surface plasmon resonance (SPR, Biacore 3000 system, GE Healthcare) was used to measure Nb affinities. Briefly, human serum albumin was immobilized to the flow channels of an activated CM5 sensor-chip. Protein analytes were diluted to 10 mg/mL in 10 mM sodium acetate, pH 4.5, and injected into the SPR system at 5 mL/min for 420 s. The surface was then blocked by 1 M ethanolamine-HCl (pH 8.5). For each Nb analyte, a series of dilution (spanning $1,000-fold concentration range) was injected in duplicate, with HBS-EP + running buffer (GE-Healthcare) at a flow rate of 20-30 mL/min for 120-180 s, followed by a dissociation time of 10-20 min based on dissociation rate. Between each injection, the sensor chip surface was regenerated twice with a low pH buffer containing ten mM glycine-HCl (pH 1.5-2.5) at a flow rate of 40-50 mL/min for 30 s -1 min. Binding sensorgrams for each Nb were processed and analyzed using BIAevaluation by fitting with the 1:1 Langmuir model.

A phylogenetic tree was generated by Clustal Omega (Sievers and Higgins, 2014) with the input of unique Nb HSA CDR3 sequences and the adjacent framework sequences (i.e., YYCAA to the N-terminus and WGQG to the C-terminus of CDR3s) to help alignments. The data were plotted by iTOL (Interactive Tree of Life) (Letunic and Bork, 2007) . Isoelectric points and hydrophobicities of the CDR3s were calculated using the BioPython library. The sequence logo was plotted using WebLogo (Crooks et al., 2004) .

The thermal stability of Nb HSA was measured by differential scanning fluorimetry (DSF iScience Article Serum samples were reduced in 8M urea digestion buffer (with 50 mM Ammonium bicarbonate, 5 mM TCEP, and DTT) at 57 C for 1 h, and alkylated in the dark with 30 mM Iodoacetamide for 30 min at room temperature. The samples were in-solution digested with 1:100 (w/w) trypsin and Lys-C at 37 C overnight, with incubation with another bolus of trypsin for 4 h. After proteolysis, the peptide mixtures were desalted and analyzed with an easyLC 1200 device coupled with a Q Exactiveä HF-X Hybrid Quadrupole Orbitrapä mass spectrometer (Thermo Fisher). Briefly, Nb peptides were loaded onto a picoChip column (C18, 1.9 mm particle size, 120 Å pore size, 75 mm 3 25 cm; New Objective) and eluted using a 45-min liquid chromatography gradient (5% B-8% B, 0-3 min; 8% B-42% B, 3-35 min; 42% B-100% B, 35-40 min; 100% B, 40-45 min; mobile phase A consisted of 0.1% formic acid (FA), and mobile phase B consisted of 0.1% FA in 80% acetonitrile (ACN). The flow rate was $300 nL/min. The QE HF-X instrument was operated in the data-dependent mode, where the top 6 most abundant ions (mass range 300-2,000, charge state 2 -8) were fragmented by high-energy collisional dissociation (HCD). The target resolution was 120,000 for MS and 60,000 for MS2 analyses. The quadrupole isolation window was 1.8 Th, and the maximum injection time for MS/MS was 120 ms.

Serum proteins including Nbs were first identified by MaxQuant version 1.6.1.0 (Cox and Mann, 2008). Up to 4 unique and specific tryptic peptides with the preference of CDR3-containing peptides with high intensities were selected from each Nb.

For the MS1-based quantification, the area under the curve (AUC) of the selected peptides was calculated from the raw data by using the Xcalibur Qual Browser. For each Nb, AUCs of the selected peptides were averaged to represent the abundance of the Nbs. Selected peptides were validated by MS/MS.

For the MS2-based quantification, an inclusion list of the Nb peptides to be monitored was generated. An LC RT window of À2$+3 min (for a 90-min LC gradient) was used to ensure that all the targeted peptides were included. The instrument was operated in a data-independent model where the targeted peptides were specifically selected for fragmentation (HCD normalized energy 27). Up to 5 abundant fragment ions from each peptide were chosen to increase the specificity of our quantification. AUCs of the fragment ions were calculated using Xcalibur Qual Browser. The relative abundance of Nbs was presented as the median AUC intensities of their respective MS2 fragment ions. It was further normalized based on the total MS1 ion current (TIC) of each LC run. An in-house script was developed to enable the automatic ll OPEN ACCESS iScience 24, 103014, September 24, 2021 23 iScience Article quantification described above. The samples were analyzed in triplicates. The abundance of serum Nbs in pharmacokinetics analysis was determined by averaging technical and biological replicates; the resulting data were used to fit into a two-phase decay model by Prism GraphPad 7.

Serum proteomic quantification 1 mg HSA was administered to a B6.Cg-Tg(FCGRT)32Dcr Alb em12Mvw Fcgrt tm1Dcr /MvwJ mouse. Blood samples were collected from the tail vein at the following time: predose, 10 min, 24, 48, 72, 96 h, 7, 10 days. Serum sample preparation and in-solution digestion of serum proteins were processed, as previously mentioned. Peptide samples were analyzed with a nano easyLC 1000 device coupled with a Q Exactiveä HF-X Hybrid Quadrupole Orbitrapä mass spectrometer (Thermo Fisher). Briefly, Nb peptides were loaded onto an analytic column (1.6 mm particle size, 100 Å pore size, 75 mm 3 25 cm; IonOpticks) and eluted using a 90-min liquid chromatography gradient. A QE HF-X instrument was used for analysis. The quadrupole isolation window was 1.6 Th, and the maximum injection time for MS/MS was set at 100 ms. Serum protein levels were quantified using MaxQuant version 1.6.1.0. The resulting raw data were searched against a mouse Uniprot FASTA protein database (December 2018) with fixed modification of carbamidomethylation and variable modifications of methionine oxidation and N-terminal acetylation. Trypsin was set as a digestion enzyme, and a maximal missed cleavage of 2 was allowed. Protein's false discovery rate was set to be 0.01. Minimal peptide length was 7, and maximal peptide mass was 4600 Da. The minimal ratio count of label-free quantification was set to be 2. Serum protein abundance at each time point is derived by averaging ion intensities of 3 technical replicates and normalized based on baseline abundance. Serum proteins were then clustered using the K-means clustering algorithm (K = 10), and heatmap was generated by software Morpheus.

hIL-2 (C125A) sequence was fused to C-termini of Nb sequences at the DNA level with a flexible linker (GGGGS) 2 (Gly-Gly-Gly-Gly-Ser-Gly-Gly-Gly-Gly-Ser). The resulting DNA sequences were synthesized and cloned into pET-21b(+) (Synbio). Plasmids were transformed into BL21(DE3) E. coli cells to produce Duraleukins. After IPTG induction, Duraleukins were purified from the cell pallets. Cell pellets containing the inclusion bodies of Duraleukins were washed with a solubilization buffer A (0.1 M Tris, 2% sodium deoxycholate (SDC), 5 mM EDTA, pH 8) and sonicated on ice. Cell lysates were spun down at 12,000 3 g for 20 min 4 C. Supernatants were discarded, and the pellets were saved for analysis. Cell pellets were washed with Milli-Q water and spun down at 12,000 3 g for 20 min at 4 C. Solubilization buffer B (0.1 M Tris, 6 M guanidine hydrochloride [GuHCl], pH 8) was added to the pallets and incubated for 1 h incubation at room temperature with gentle agitation. After centrifugation 12,000 3 g for 20 min at 4 C, the supernatants that contain soluble forms of Duraleukins were collected and subsequently diluted with a Duraleukin refolding buffer (0.1 M Tris, pH 8, the final concentration of 10 mM reduced and 1 mM oxidized glutathione) to a GuHCl concentration of 2 M and a protein concentration of $0.3 mg/mL. The solutions were then incubated at room temperature for 16 h with gently rotating to allow protein refolding. The insoluble fraction was removed by centrifugation at 16,000 3 g for 20 min at 4 C and refolded Duraleukins were obtained in the supernatants. IMAC and FPLC purification, detoxification, and sterilization were performed as described in ''Nb purification'' section of STAR methods.

In vitro T cell proliferation assay CTLL-2 cells (ATCC) were cultured according to ATCC's protocol. For the CTLL-2 proliferation assay, cells were grown to a density of 1 3 10 5 cells/mL. Duraleukins or hIL-2 (PeproTech) was added into the wells of a flat-bottom 96-well tissue culture microplate. The final concentration of hIL-2 and Duraleukins was adjusted to 0.01-25 ng/mL and 0.025-50 ng/mL by 2-fold serial dilution. Every concentration group was tested in duplicate. 10 4 cells were seeded to each well of the plate, and the total volume in each well was 100 mL. The cells were then incubated at 37 C, 5% CO 2 for 56 h. 10 mL of WST-1 reagent (Roche) was added to each well and the cells were incubated for another four h. The plate was shaken at 1,000 rpm for 1 min before analysis. A spectrometer read absorbance at 450 nm. 690 nm was used as a reference wavelength. A dose-response curve was fit via nonlinear regression using GraphPad Prism 7.

Generation and treatment of melanoma mouse model B16-F10 (ATCC) was cultured according to ATCC's protocol. All C57BL/6J mice (JAX) were aged eight weeks before tumor inoculation. B16-F10 melanoma cells were grown into the logarithmic growth phase ll OPEN ACCESS

The clinical plasma proteome: a survey of clinical assays for proteins in plasma and serum

Reproducibility crisis: blame it on the antibodies

Targeted proteomic quantification on quadrupole-orbitrap mass spectrometer

Revisiting biomarker discovery by plasma proteomics

Structural basis of the drug-binding specificity of human serum albumin

Enhanced permeability and retention of macromolecular drugs in solid tumors: a royal gate for targeted anticancer nanomedicines

Naturally occurring antibodies devoid of light chains

Properties, production, and applications of camelid single-domain antibody fragments

Effect of pegylation on pharmaceuticals

Nanobodies and their potential applications

Comparison of the results obtained by ELISA and surface plasmon resonance for the determination of antibody affinity

Predictive value of EGFR and HER2 overexpression in advanced non-small-cell lung cancer

Pre-clinical intravenous serum pharmacokinetics of albumin binding and non-half-life extended Nanobodiesâ

Engineered antibody fragments and the rise of single domains

The therapeutic potential of nanobodies

Boosting therapeutic potency of antibodies by taming Fc domain functions

Antibody-dependent enhancement of severe dengue disease in humans

Combining immune checkpoint inhibitors: established and emerging targets and strategies to improve outcomes in melanoma

Nanobody-based cancer therapy of solid tumors

Integrative structure and functional anatomy of a nuclear pore complex

Half-life extended biotherapeutics

Albumin as a drug carrier: design of prodrugs, drug conjugates and nanoparticles

Delivery of ALX-0171 by inhalation greatly reduces respiratory syncytial virus disease in newborn lambs

Nanobodies: site-specific labeling for super-resolution imaging, rapid epitope-mapping and native protein complex isolation

A role for nonprotective complement-fixing antibodies with low avidity for measles virus in atypical measles

FcRn: the architect behind the immune and nonimmune functions of IgG and albumin

Adrenaline-activated structure of beta2-adrenoceptor stabilized by an engineered nanobody

Albumin-deficient mouse models for studying metabolism of human albumin and pharmacokinetics of albumin-based drugs

IL-2: the first effective immunotherapy for human cancer

Principles for integrative structural biology studies

Putting the pieces together: integrative modeling platform software for structure determination of macromolecular assemblies

Enzyme-linked immunosorbent assay for the quantitative/ qualitative analysis of plant secondary metabolites

Unraveling the interaction between FcRn and albumin: opportunities for design of albumin-based therapeutics

RELION: implementation of a Bayesian approach to cryo-EM structure determination

Crystal structure of an HSA/FcRn complex reveals recycling by competitive mimicry of HSA ligands at a pHdependent hydrophobic interface

PatchDock and SymmDock: servers for rigid and symmetric docking

Modeling of multimolecular complexes

Managing toxicities of high-dose interleukin-2

Caplacizumab treatment for acquired thrombotic thrombocytopenic purpura

Structural characterization by cross-linking reveals the detailed architecture of a coatomerrelated heptameric module from the nuclear pore complex

A strategy for dissecting the architectures of native macromolecular assemblies

Clustal Omega, accurate alignment of very large numbers of sequences

De novo design of potent and selective mimics of IL-2 and IL-15

Serum amyloid A, the major vertebrate acute-phase reactant

Serum albumin-binding V Hs with variable pH sensitivities enable tailored half-life extension of biologics

Nanobodiesâ as inhaled biotherapeutics for lung diseases

Nanobodies targeting the hepatocyte growth factor: potential new drugs for molecular cancer therapy

Comparative protein structure modeling using MODELLER

Quality issues of research antibodies

Versatile and multivalent nanobodies efficiently neutralize SARS-CoV-2

Chemical cross-linking and mass spectrometric analysis of the endogenous yeast exosome complexes

Integrative proteomics identifies thousands of distinct, multi-epitope, and high-affinity nanobodies

Cross-linking mass spectrometry: an emerging technology for interactomics and structural biology

Synergistic innate and adaptive immune response to combination immunotherapy with anti-tumor antigen antibodies and extended serum half-life IL-2

The samples were analyzed by a 7900HT Fast Real-Time PCR System (Applied Biosystems) in triplicate. The temperature was programmed to increase from 25 C to 95 C with a ramp rate of 1 C/min to generate the melting curves. The melting point was calculated by first derivatives method

Chemical cross-linking and mass spectrometry (CXMS)

ThermoFisher Scientific) for 23 min at 25 C with gentle agitation. The crosslinking reaction was then quenched with 50 mM ammonium bicarbonate (ABC) for 10 min at room temperature. After protein reduction and alkylation, the cross-linked samples were separated by a 4-12% SDS-PAGE gel (NuPAGE, Thermo Fisher). The regions corresponding to the cross-linked species ($130 kDa) were cut and in-gel digested with trypsin and Lys-C as previously described

100% B

1% formic acid (FA), and mobile phase B consisted of 0.1% FA in 80% acetonitrile. The QE HF-X instrument was operated in the data-dependent mode, where the top 8 most abundant ions (mass range 380-2,000, charge state 3-7) were fragmented by high-energy collisional dissociation (normalized collision energy 27). The target resolution was 120,000 for MS and 15,000 for MS/MS analyses. The quadrupole isolation window was 1.8 Th, and the maximum injection time for MS/MS was set at 120 ms. After the MS analysis, the data was searched by pLink 2 (Chen et al., 2019) for the identification of crosslinked peptides. The mass accuracy was specified as 10 and 20 p.p.m. for MS and MS/MS, respectively. Other search parameters included cysteine carbamidomethylation as a fixed modification and methionine oxidation as a variable modification. A maximum of three trypsin missed-cleavage sites was allowed. The initial search results were obtained using the default 5% false discovery rate, estimated using a targetdecoy search strategy. The cross-link spectra were then manually checked to remove potential false-positive identifications essentially as previously described

Each Nb model was then docked to the HSA structure (PDB: 4g03) by an antibody-antigen docking protocol of PatchDock software that focuses the search to the CDRs (Schneidman-Duhovny et al., 2005) and optimizes CXMS-based distance restraints satisfaction

The antigen interface residues (distance <6Å from Nb atoms) among the 10 best scoring models according to the SOAP score, were used to determine the epitopes. Convergence was measured as the average RMSD among the 10 top-scoring models. Once the epitopes were defined, we clustered the Nbs based on the epitope similarity using hierarchical clustering and displayed the models by Chimera X

Protein was diluted to a final concentration of 30 mg/mL (HSA in PBS; HSA-Nb complexes in Tris 20 mM, 200mM NaCl, 3% glycerol) and deposited on carbon-coated CF400-CU grids (EMS) which were freshly glow discharged. After 30 s of incubation, the excessive protein was removed, and the grid was stained with two drops of uranyl acetate 2% w/v. Electron micrographs were recorded in an FEI TECNAI T12 operating at 120 kV with a 2k x 2k Gatan UltraScan 1000. Raw images were converted by IMOD 4.8 ll OPEN ACCESS (Mastronarde and Held, 2017) and particles were selected using EMAN2

Site-directed mutagenesis

Plasmids bearing wild type HSA and the mutants were transfected to HeLa cells using Lipofectamine 3000 transfection kit (Invitrogen) and Opti-MEM (Gibco) according to the manufacturer's protocol. The cells were cultured overnight before a change of medium to DMEM without FBS supplements to remove BSA. After a 48 h culture at 37 C, 5% CO 2 , the media expressing HSA were collected and stored at À20 C. The media were

After injections, the whole blood samples were collected from the tail veins of the mice. The blood sampling time for HSA was pre-dose (0 min), 10 min, 24 h, and 48 h. The sampling time for Nb mixture was pre-dose (0 min), 5, 15

For flow cytometry analysis, a single i.v. admindissection and Collagenase IV Cocktail digestion (formulated as 7.4 kU Collagenase Type 4, 53 kU Deoxyribonuclease I, 20 mg Soybean Trypsin Inhibitor in 10 mL cocktail; Worthington Biochemical). The suspensions were passed through 40 mm cell strainers to ensure singularity. Cells were first treated with BioWhittaker ACK lysing buffer (Lonza), followed by TruStain FcX (BioLegend) and stained with antibodies against CD45 (clone 30-F11), CD3 (clone 17A2), CD8a (clone 53-6.7), NK1.1 (clone PK136)