The human autonomous V_H and V_L sdAb scaffolds used in this study (Table 1; Figure S1 in Supplementary Material) were isolated as previously described by To et al. (38) and Kim et al. (39). Disulfide-stabilized versions of each V_H/V_L sdAb (bearing an intradomain disulfide linkage formed between Cys residues at IMGT positions 54 and 78) were produced by overlap extension PCR as described in Kim et al. (40).

All three CDRs of each V_H/V_L scaffold were simultaneously exchanged for those listed in Table 2 (20 V_H/V_L scaffolds × 12 CDR sets = 240 CDR-shuffled variants). DNA constructs encoding the CDR-shuffled variants were synthesized commercially (GeneArt/Life Technologies, Regensberg, Germany) and subcloned into the pSJF2H bacterial expression vector (12).

Three phage-displayed sdAb libraries were constructed by in vitro randomization of the sdAb scaffolds VH428, VHB82_SS and VL383_SS. Briefly, nondegenerate oligonucleotides spanning each sdAb were chemically synthesized using the phosphoramidite method (GeneArt/Life Technologies) and purified by HPLC. CDRs were randomized via incorporation of defined mixtures of trinucleotide phosphoramidite building blocks (41) during oligonucleotide synthesis. Oligonucleotides were assembled without amplification by Klenow fragment extension, gel purified using a GeneJET™ gel extraction kit (Thermo-Fisher, Waltham, MA, USA) according to the manufacturer’s instructions, and resuspended in a total volume of 1 mL TE buffer (10 mM Tris-HCl, 1 mM EDTA, pH 8.0). Non-amplified library DNA was quantified by real-time PCR using Fast SYBR™ Green master mix (Thermo-Fisher), and a StepOnePlus™ real-time PCR system (Thermo-Fisher), then approximately 232–1,069 ng (4.3 × 10¹¹ to 1.5 × 10¹² molecules) of DNA was PCR-amplified with Phusion^® high-fidelity DNA polymerase (Thermo-Fisher) using M13F and M13R primers (sequences in Table S1 in Supplementary Material), gel purified using a GeneJET™ gel extraction kit, then digested and ligated between the NcoI and NotI sites of the phagemid vector pMED1 (42). Escherichia coli TG1 cells were transformed with the ligation products by electroporation, yielding final library sizes between 1.0 × 10¹⁰ and 2.3 × 10¹⁰ independent transformants, ≥95% of which carried V_H/V_L sdAb inserts as shown by colony PCR.

Recombinant human CEACAM6 (residues 145–232) and mouse GITR (residues 20–153) ectodomains were produced by transient transfection of HEK293-6E cells as previously described (43). HSA was from Sigma-Aldrich (St. Louis, MO, USA; Cat. No. A9511). Recombinant human CTLA4 ectodomain (residues 1–162) was from Sino Biological (Beijing, China; Cat. No. 11159-H08H). E. coli 0157:H7 intimin (residues 658–934) fused C-terminally to maltose-binding protein (MBP-intimin) was from GenScript (Piscataway, NJ, USA). Horseradish peroxidase-conjugated antibodies used in ELISA (mouse anti-M13, Cat. No. 27942101; mouse anti-c-Myc, Cat. No. 11814150001; rabbit anti-6 × His, Cat. No. A190-114P) were from GE Healthcare (Piscataway, NJ, USA), Bethyl Laboratories (Montgomery, TX, USA) and Roche Diagnostics (Basel, Switzerland), respectively. M13K07 helper phage was from New England BioLabs (Ipswich, MA, USA) and M13K07ΔpIII hyper phage was from Progen Biotechnik (Heidelberg, Germany).

Phage particles displaying monovalent V_H/V_L sdAbs were prepared by rescue of the three synthetic human V_H/V_L sdAb phagemid libraries with M13K07 helper phage as previously described (42). Briefly, 2 × YT broth (4 L) containing 1% (w/v) glucose and 100 µg/mL ampicillin was inoculated with 1 mL phagemid-bearing E. coli TG1 cells (7.3 × 10¹⁰ to 1.6 × 10¹¹) and grown at 37°C with 250 rpm shaking to an OD₆₀₀ of 0.5. Phagemid-bearing cells were superinfected with M13K07 helper phage at a multiplicity of infection of 20:1 at 37°C for 30 min with no shaking, then pelleted and resuspended in 6 L of 2 × YT broth containing 100 µg/mL ampicillin and 50 µg/mL kanamycin. The next day, phage particles were purified from bacterial supernatants using two rounds of polyethylene glycol precipitation; neither heat treatment nor filtration steps were used to remove residual bacterial cells. For panning, 5 µg of each protein in 35 µL phosphate-buffered saline (PBS), pH 7.4, was adsorbed overnight at 4°C in wells of NUNC MaxiSorp™ 96-well microtiter plates (Thermo-Fisher). The next day, wells were blocked with 200 µL PBS containing either 2% (w/v) bovine serum albumin (BSA) or 2% (w/v) skim milk for 1 h at 37°C, then rinsed 3 × with PBS. Approximately 5 × 10¹¹ infective library phage particles (5 × 10¹² virions) were added to wells in 100 µL of PBS containing 1% BSA or skim milk and 0.1% (v/v) Tween-20 for 2 h at room temperature. The blocking protein (BSA and skim milk) was switched in alternate rounds of panning, except for panning on HSA in which only skim milk was used. Wells were washed 5 × with PBS containing 0.1% Tween-20, 2 × with PBS and then bound phage were eluted for 10 min with 50 µL of 100 mM triethylamine, neutralized with 50 µL 1 M Tris-HCl, pH 7.5, and used to reinfect exponentially growing E. coli TG1 cells. The cultures were superinfected with M13K07 helper phage. The next day, amplified phage were purified by polyethylene glycol precipitation from 10 mL overnight cultures and used in subsequent panning rounds. After four or five rounds of selection, antigen-specific V_H- or V_L-displaying phage clones were identified by their binding in polyclonal and monoclonal phage ELISA as previously described (42).

Phage particles displaying monovalent V_H/V_L sdAbs were prepared by rescue of phagemid libraries with M13K07 helper phage as described above. Phage particles displaying multivalent V_H/V_L sdAbs on all copies of pIII were prepared as described above, except that: (i) smaller volumes (100 mL) of 2 × YT broth were inoculated with 1 mL phagemid-bearing E. coli TG1 cells, (ii) phagemid libraries were rescued with M13K07ΔpIII hyper phage at a multiplicity of infection of 10:1, and (iii) the final overnight culture volume was 150 mL of 2 × YT broth.

For panning, 5 µg of either protein A (for the VHB82_SS library; Thermo-Fisher) or protein L (for the VL383_SS library; Thermo-Fisher) in 50 µL PBS, pH 7.4, was adsorbed overnight at 4°C in wells of NUNC MaxiSorp™ 96-well microtiter plates. The next day, wells were blocked with 300 µL PBS containing either 2% BSA or skim milk for 1 h at 37°C, then rinsed 1 × with PBS. Approximately 10¹¹ infective library phage particles (10¹² virions) were added to wells in 100 µL of PBS containing 1% skim milk and 0.1% Tween-20 for 2 h at room temperature. Wells were washed 5 × with PBS containing 0.1% Tween-20, 2 × with PBS and then bound phage were eluted for 10 min with 50 µL of 100 mM triethylamine, neutralized with 50 µL 1 M Tris-HCl, pH 7.5, and used to reinfect exponentially growing E. coli TG1 cells. The cultures were superinfected with either M13K07 helper phage or M13K07ΔpIII hyper phage. The next day, amplified phage were purified by polyethylene glycol precipitation from 10 mL overnight cultures and used in subsequent panning rounds. After three rounds of selection, pools of sdAb-phage (either phage bound by and eluted from protein A/L, or phage amplified from overnight cultures) were interrogated using next-generation DNA sequencing (NGS) as described below.

Monomeric V_H/V_L sdAbs bearing C-terminal 6 × His and c-Myc tags were expressed from overnight cultures of E. coli TG1 cells grown in 250 mL to 1 L of 2 × YT broth under IPTG (isopropyl β-D-1-thiogalactopyranoside) induction, then extracted from periplasmic space by osmotic sucrose shock and purified by immobilized metal affinity chromatography as previously described (42). For small-scale expression screening, 5 mL overnight cultures were grown as above, lysed using FastBreak™ reagent (Promega, Madison, WI, USA) and sdAbs purified using PureProteome™ nickel magnetic beads (EMD Millipore, Billerica, MA, USA). Expression yields from 5 mL cultures were determined using the Bradford protein assay (Bio-Rad, Hercules, CA, USA) as per manufacturer’s instructions with a V_HH sdAb of known concentration as the protein standard. Titration ELISAs using soluble V_H/V_L sdAbs were performed as previously described (42, 44, 45), using either anti-6 × His or anti-c-Myc secondary antibodies to detect binding.

Size exclusion chromatography analyses of monomeric V_H/V_L sdAbs were conducted using a Superdex™ 75 GL column (GE Healthcare) connected to an ÄKTA FPLC protein purification system (GE Healthcare) as previously described (42). UPLC-SEC-MALS analyses of V_H/V_L sdAbs were conducted essentially as previously described (39) using an Acquity BEH-125 column (Waters, Milford, MA) connected to an Acquity UPLC H-Class Bio system (Waters) with miniDAWN™ MALS detector and Optilab^® UT-rEX™ refractometer (Wyatt Technology, Santa Barbara, CA, USA). V_H/V_L sdAbs (10–20 µg) were injected at 30°C in a mobile phase consisting of calcium- and magnesium-free DPBS (GE Healthcare) at a flow rate of 0.4 mL/min. Weighted average molecular mass (M_MALS) was calculated using a protein concentration determined using A₂₈₀ from the PDA detector with extinction coefficients calculated from amino acid sequences. Data were processed using ASTRA 6.1 software (Wyatt).

All monomeric human V_H/V_L sdAbs were SEC-purified and buffer-exchanged into HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% (v/v) surfactant P20, pH 7.4) prior to SPR analyses. All SPR analyses were performed on a Biacore™ 3000 instrument (GE Healthcare) at a temperature of 25°C. Briefly, proteins (CEACAM6, 370 resonance units (RUs); HSA, 1614 RUs; GITR, 796 RUs; MBP-intimin, 1536 RUs) were immobilized via amine coupling on research-grade CM5 sensor chips (GE Healthcare) in 10 mM acetate buffer, pH 4.0, according to the manufacturer’s instructions. V_H/V_L sdAbs were injected at concentrations ranging from 1 nM to 10 µM, at flow rates of 20–50 µL/min and with contact time between 120 s and 300 s, then allowed to dissociate for 7–10 min. All surfaces were regenerated using 10 mM glycine, pH 1.5. Ethanolamine-blocked flow cells served as reference surfaces. The data were fitted to a 1:1 interaction model and binding affinities (K_Ds) and/or kinetic parameters were determined either by steady-state analysis or by multicycle kinetic analysis using BIAevaluation 4.1 software (GE Healthcare).

Surface plasmon resonance analyses of two V_L sdAbs (VL_SS-2 and VL_SS-5) against MBP-intimin were conducted as described above, except that 234 RUs of MBP-intimin were immobilized on a research-grade C1 sensor chip (GE Healthcare) via amine coupling in 10 mM acetate buffer, pH 4.5, according to the manufacturer’s instructions. These two V_L sdAbs were injected in HBS-EP buffer containing either 150 or 500 mM NaCl with an extended contact time (600 s).

V_H/V_L sdAb samples (0.2 mL) in 1.5 mL microcentrifuge tubes were heated to 85°C in a heating block for 10 min, then allowed to cool to room temperature for 30 min. Absorbance at 360 nm was measured pre- and post-heat treatment in a quartz NanoQuant Plate™ (Tecan, Männedorf, Switzerland) using an Infinite^® M200 PRO microplate reader (Tecan). V_H/V_L sdAb concentrations were adjusted to 0.25 mg/mL in PBS prior to analysis.

V_H/V_L sdAb samples (45 µL) in 96-well thin-wall optical plates (Bio-Rad) were mixed with 5 µL SYPRO^® Orange (diluted 1:100 from 5,000 × stock; Life Technologies, Carlsbad, CA, USA) and sealed with optical quality sealing tape (Bio-Rad). Using an iQ™ 5 real-time PCR system (Bio-Rad), a temperature ramp of 1°C/min was applied and thermal unfolding was monitored by measuring fluorescence at 0.5°C intervals as previously described (46, 47). The wavelengths for excitation and emission were 490 and 575 nm, respectively. Melting temperatures (T_ms) were calculated as the temperature at which the maximum rate of change in fluorescent signal (d(RFU)/dt) was achieved. V_H/V_L sdAb concentrations were adjusted to 1 mg/mL in PBS prior to analysis.

T_ms were also determined by circular dichroism as previously described (39, 40). Ellipticity of V_H/V_L sdAbs (100 µg/mL) was measured at wavelengths between 205 and 210 nm in 100 mM sodium phosphate buffer, pH 7.4. Ellipticity measurements were normalized to a percentage scale and then T_ms were determined by plotting percent folded versus temperature and fitting the data to a Boltzmann distribution.

V_H/V_L sdAb libraries were interrogated using an Illumina MiSeq instrument as described previously (44, 45, 48). Amplicons for NGS were prepared using FR1- and FR4-specific barcoded primers (sequences in Table S1 in Supplementary Material) using as template either phagemid DNA (~10 ng) or phage virions (~10⁶ particles).

Descriptive statistics (mean, median) were used to describe datasets as described in figure legends. No inferential statistical tests were used.

The T_ms and aggregation tendencies of 22 potential V_H sdAb scaffolds and 18 potential V_L sdAb scaffolds were determined by circular dichroism and SEC (Tables S2 and S3 in Supplementary Material). Five V_H and five V_L sdAb scaffolds as well as their disulfide-stabilized equivalents (Table 1) were selected as the most promising candidates for library construction based on: (i) primarily monomeric folding, (ii) high thermostability with reversible thermal unfolding, and (iii) reasonable expression yields (>1 mg/L) from overnight E. coli TG1 cultures. As reported previously (39, 40), incorporation of a disulfide linkage spanning IMGT positions 54–78 increased the T_m of each scaffold by ~5–20°C but also had unpredictable effects on expression yields.

As a preliminary investigation into which of the 20 V_H/V_L sdAb scaffolds might best tolerate library randomization, we grafted 12 sets of exogenous CDRs (Table 2) into each scaffold and assessed the resulting molecules’ expression level (Bradford assay), aggregation tendency (SEC-MALS), T_m (thermal shift assay), and ability to refold after thermal denaturation (turbidity assay). For V_L sdAb scaffolds, the 12 sets of exogenous CDRs were derived from non-antigen-specific human V_L sdAbs isolated from previously constructed libraries with expression yields ≥10 mg/L and for V_H sdAb scaffolds, the 12 sets of exogenous CDRs were selected from either human V_H sdAbs of unknown antigen specificity isolated from previously constructed libraries or camelid V_HH sdAbs with expression yields ≥3 mg/L.

The V_L sdAb scaffolds yielding the best-expressing CDR-shuffled variants were VL383_SS, VL325_SS and VL335_SS (Figure 1A). CDR-shuffled variants of VL383_SS and VL325_SS were primarily monomeric by SEC-MALS, while CDR-shuffled variants of VL335_SS showed some tendency to aggregate (Figure 1C). All three of these V_L sdAb scaffolds, especially VL325_SS, yielded CDR-shuffled variants with high T_ms (Figure 1E); however, many CDR-shuffled variants of VL325_SS showed significant turbidity upon heat denaturation (Figure 1G), reflecting their inability to refold as soluble monomers. The V_H sdAb scaffolds yielding the best-expressing CDR-shuffled variants were VH420, VH428, VHB82, VHB82_SS, and VHM81_SS (Figure 1B). CDR-shuffled variants of all five of these V_H sdAb scaffolds were similarly monomeric (Figure 1D) but as expected, the two disulfide-stabilized sdAb scaffolds (VHB82_SS and VHM81_SS) yielded CDR-shuffled variants with higher T_ms (Figure 1F). Similar proportions of the CDR-shuffled variants of all five V_H sdAb scaffolds showed minor turbidity upon heat denaturation (Figure 1H). On the basis of these data, we selected one V_L sdAb scaffold (VL383_SS) and two V_H sdAb scaffolds (VH428 and VHB82_SS) for library construction, reflecting a balance of their tolerance to CDR modification as well as their differing CDR3 lengths and germline gene rearrangements.

Three synthetic sdAb libraries (VL383_SS, VH428, and VHB82_SS) were constructed by limited in vitro randomization of the CDRs of these V_H/V_L sdAb scaffolds and cloning of the resulting DNA into the pMED1 phagemid vector (Figures 2A,B). CDR3 length was varied in each library (Figures 2A,D; VL383_SS library: 8, 9, or 10 residues; VH428 and VHB82_SS libraries: 10, 12, 14, 16, 18, 20, or 22 residues) while CDR1 and CDR2 lengths were constant. The major technical advances of these over our previously described phage-displayed synthetic human V_H/V_L sdAb libraries (12, 21, 31) were (i) trinucleotide mutagenesis allowed for incorporation of defined mixtures of amino acids at each randomization position, based on alignments of human and llama antibodies with additional bias toward solubility-promoting residues (Asp, Glu) and against undesirable residues (Asn, Cys, and Met) and stop codons (Figures 2B,C; Figure S2 in Supplementary Material) and (ii) near-complete randomization of all three CDRs of each V_H/V_L sdAb scaffold resulted in much higher sequence diversity for the vast majority of library molecules (Figure 2E). All three V_H/V_L sdAb libraries were transformed into E. coli TG1 cells, yielding final library sizes (1.0–2.3 × 10¹⁰ independent transformants) that, as expected, vastly under-sampled theoretical library diversity (VL383_SS: 2.9 × 10¹³ unique sequences; VHB82_SS: 2.3 × 10³³ unique sequences; VH428: 2.3 × 10³⁵ unique sequences). At this stage, prior to helper phage rescue, no major clonality biases or deviations from library design were observed (data not shown).

All three V_H/V_L sdAb libraries were rescued with M13K07 helper phage and panned for four or five rounds against five model antigens (CEACAM6, CTLA4, GITR, HSA, and MBP-intimin). All library pannings were performed in duplicate by two independent operators at the same time and using the same materials, except for panning against MBP-intimin, in which panning was done in triplicate. High attrition rates of V_H/V_L sdAbs showing false positive binding either as sdAb-phage or as soluble sdAb proteins (VL383_SS: 53% attrition; VH428: 69% attrition; VHB82ss: 71% attrition; Table 3) resulted in a final yield of 20 unique antigen-specific V_H/V_L sdAbs against four targets. Of the 15 library-target screens we conducted, eight (53%) yielded at least one antigen-specific V_H/V_L sdAb (four screens yielded one sdAb, two screens yielded two sdAbs, one screen yielded three sdAbs, and two screens yielded five sdAbs). Most of the recovered sdAbs were monomeric (Figure 3A; Table 4; Figure S3 in Supplementary Material), thermostable (Figure 3B; Table 4), and had affinities for antigen ranging from 5 nM to ~12 μM (Figures 3C,D; Table 4). One notable exception was the α-GITR V_H sdAb VHB82_SS-6, whose elution volume by SEC suggested it existed in solution as a strict dimer (Figure 3A). Stable homodimerization has previously been reported for several human V_H sdAbs and was critical for antigen binding (49–52); this phenomenon is distinct from the general tendency of some V_H domains to forms soluble aggregates. Antigen-specific V_H/V_L sdAbs isolated from all three libraries had aggregation tendencies and thermostabilities reflective of the parental V_H/V_L sdAb scaffold from which they were derived, but generally poorer expression yields (Figure 4A). Reproducibility in the selection of specific sdAb sequences from the libraries was modest, with only 45% (9/20) of antigen-specific V_H/V_L sdAbs recovered by both operators (Figure 4B) and no apparent connection between antigen-binding affinity and consistency of isolation.

To better understand potential constraints on the CDR sequences of monomeric and stable human V_H/V_L sdAbs, we examined the effects of stability selection (three rounds of protein A or L selection followed by amplification of the eluted phage overnight in E. coli TG1 cells; Figure 5) on randomized sequence diversity of two V_H/V_L sdAb libraries (VL383_SS and VHB82_SS). We performed this experiment in duplicate using either phage particles displaying monovalent sdAb (rescued with M13K07 helper phage; simultaneous expression of pIII from helper phage and pIII-sdAb from phagemid) or phage particles displaying multivalent sdAb (rescued with M13K07ΔpIII hyper phage; only pIII-sdAb expressed from phagemid). Out-of-frame and/or stop-codon-encoding V_H/V_L sdAb clones were rare in both libraries (“Library Cells”), but rose substantially in frequency upon phage rescue (“Library Phage”) and upon amplification in E. coli (“Amplification in E. coli”), suggesting a growth advantage for non-sdAb expressing cells over sdAb-expressing ones (Figures 5A,B); this phenomenon was mitigated somewhat using hyper phage rescue. After three rounds of protein A/L selection and overnight amplification in E. coli with helper phage rescue, there were clear clonal biases observed in the resulting populations of sdAb-phage (Figures 5C,D). Using an arbitrary frequency cutoff (0.00006% for VL383_SS and 0.00004% for VHB82_SS; sdAbs at frequencies greater than these cutoffs were not present in any other dataset) to identify unique sdAb sequences enriched by stability selection, we found that different sdAb clones were selected in the two replicate pannings, although both sets of sdAbs were heavily biased toward the parental V_H/V_L sdAb scaffold’s CDR3 length (9 residues for the VL383_SS library and 10 residues for the VHB82_SS library, which is the shortest CDR3 length possible in the library design and the nearest to the VHB82_SS scaffold’s CDR3 length of 6 residues; Figure S4 in Supplementary Material). However, stability selection reproducibly yielded V_H/V_L sdAbs with biased CDR amino acid composition (Figures 5E,F; Figures S5 and S6 in Supplementary Material). In order of magnitude, stability biases in the VL383_SS library favored Asp, His, Pro, Ser and Thr and disfavored aromatic (Phe, Trp, Tyr) and hydrophobic (Ile, Leu, Val) residues. Similarly, stability biases in the VHB82_SS library favored Asp, His, Pro, and Gln and disfavored aromatic (Phe, Trp, Tyr) and hydrophobic (Ile, Leu, Val) residues. These biases were especially acute at the C-terminus of CDR1 and N-terminus of CDR2, both of which flank FR2. Identical molecular signatures were observed to a lesser degree in the pools of all sdAb-phage (irrespective of frequency or mono- vs. multivalent display format) subjected to three rounds of stability selection (Figures S7 and S8 in Supplementary Material) as well as in the pools of sdAb-phage after phage rescue and eluted from protein A/L (data not shown), suggesting a common ontogeny.