Five amino acids (AAs) were chosen in each complementarity-determining region (CDR) (there are three CDRs, so 15 AAs in total were chosen) to mutate randomly into other AAs, in this way we generated a mutation library consisting of about 10⁷ samples. These samples varied little in pI and molecular mass and had the same length, thus it was helpful to focus on the variables of AA composition. Then, 65 mutated HSA dAbs excluding terminator mutants (AUA, CCU, CCC, AGA and AGG) or sequential repeat mutants were chosen randomly as experimental subjects and 10 were chosen as verification subjects. These mutated sequences are listed in Table 1.

The dAb fragments were cloned into vector pBY (an efficient expression vector constructed by a coworker in our lab) and introduced into E. coli strain BL21(DE3). The transformed cells were plated onto Luria-Bertani (LB) agar plates (Solarbio® Life Sciences, Beijing, PR China) and incubated at 37 °C overnight. After that, single colonies were selected and inoculated into 25 mL of LB medium (containing 15 μg/mL of tetracycline (Shanghai Shenggong Co. Ltd., Shanghai, PR China) in 250-mL flasks and incubated at 37 °C for 7 h with shaking at 230 rpm. Stock solutions were prepared by mixing 500 μL of culture with 500 μL of 20% glycerol (Shanghai Hushi Laboratorial Equipment Co. Ltd., Shanghai, PR China) solution in 1.5-mL tubes, and the cells were stored at −80 °C.

Cultivation can be divided into three phases: seed culture, growth and induction phase. Forty-eight square multititer plates (48-MTP; Thermo Fisher Scientific, Shanghai, PR China) were used to culture the 66 strains (65 mutated strains and a control strain) to achieve parallel fermentation. In the seed culture phase, 2 mL of LB medium containing 15 μg/mL of tetracycline were added into each well of the 48-MTP. After inoculation with 20 μL of stock cell solution, 48-MTPs were incubated in a shaker at 230 rpm and 30 °C for 16 h. In the growth phase, the seed solutions were transferred to fresh 48-MTPs containing 2 mL of Terrific Broth/Super Broth (TB/SB; Solarbio® Life Sciences) medium with 15 μg/mL of tetracycline and cultured under the same conditions as described above. The inoculum volume was calculated by the following equation, thus fixing the initial A_{595 nm} at 0.05:V(inoculum)=(0.05∙V(fermentation)/A(seed culture))/mL /1/where V is the volume, 0.05 is the initial absorbance (A) at 595 nm and A is the absorbance of seed culture solution.

Seven hours after the second inoculation, isopropyl-β-d-thiogalactoside (IPTG; Solarbio® Life Sciences) was added to each well to a final concentration of 0.1 mM and the culture temperature was lowered to 23 °C simultaneously. The induction phase lasted for 16 h. After centrifugation of the culture broth at 6000×g (centrifuge model Sorvall ST 16R; Thermo Fisher Scientific, Shanghai, PR China), the supernatants were collected, the cell pellets were resuspended in phosphate-buffered saline (PBS; Shanghai Hushi Laboratorial Equipment Co. Ltd) and lysed using Precellys 24 (Bertin Technologies, Paris, France), and then supernatants were collected.

The whole process of cultivation was repeated six times; batches with small deviation of dAb production by control strain were chosen for further analysis, and in this way, parallel operations were guaranteed.

Two amounts of soluble expression of dAbs were measured by direct ELISA, i.e. soluble dAbs in broth supernatant and in pellet lysate supernatant. Flat-bottomed 96-well plates (Thermo Fisher Scientific) were first coated with 50 μL of supernatant. After blocking with 5% non-fat milk in PBS with Tween 20 (PBST; Shanghai Hushi Laboratorial Equipment Co. Ltd), the dAbs were detected using HRP-labelled protein A (Boster Biological Technology Co. Ltd., Beijing, PR China) with the substrate tetramethylbenzidine (Zhengzhou Biocell Biotechnology Co. Ltd., Zhengzhou, PR China). The reactions were stopped by the addition of 100 μL of 2 M sulfuric acid, and the absorbance was measured at 450 nm/620 nm using an EZ Read 800 (Biochrom, Cambridge, UK). The amount of dAb was calculated from a standard curve made using reference sample. Total protein mass fraction was detected using a modified Bradford protein assay kit (Sangon Biotech Co. Ltd., Shanghai, PR China). To avoid the difference caused by different degrees of cell lysis, standardized amounts of dAbs in μg per g of total protein were calculated as follows and used in the data analysis (Table 2):

The software package R (24) was used to analyze the contributions of factors such as AA composition, dAb charge and polarity on dAb soluble expression level. Factors with p<0.05 were considered significant. Categories of AAs based on Vector NTI® (25) are listed in Table 3. Two levels of analysis were run, including dividing expression levels into high and low by Clustal Omega (26), and identifying the factors that had an effect on the expression level by t-test. A linear regression model was constructed, then factors that had a significant influence were removed in turn to identify the most significant ones based on Akaike information criterion (AIC) values (27). We used SWISS-MODEL (20) to get 3D structure, and PyMOL (21) to decorate CDRs in three different colours.

It is widely accepted that AA sequence is significantly correlated with protein production, which was also shown in this study through analysis of the consistency of cluster results based on AA sequences and the corresponding soluble expression levels of dAbs (Table 4). AA compositions of the whole dAb (in percentage) were set as variables to explore their effect on the dAb soluble expression level by a stepwise regression analysis, and the results are summarized in Table 5.

Stepwise regression was taken to analyse AA effect on dAb soluble expression level in broth supernatant, in pellet lysate supernatant and total soluble dAb. Results showed that the combination of AAs S, R, N, D, Q, Y, F and G had a significant influence on dAb soluble yield in broth supernatant, with the p-value of 0.002. Specifically, S, N, D and Q had positive effects, with p-values of 0.0006, 0.02, 0.03 and 0.05, respectively, which means that the soluble yield of dAb in broth supernatant increased with increasing content of these AAs. However, R had a negative effect (p=0.001), thus dAb would be more difficult to express in soluble form in broth supernatant with a higher content of R. Moreover, the combined composition of G, R, C, N, S, Y, K and A had a significant effect on dAb soluble yield in the pellet lysate (p=0.002). Again, R showed a significantly negative effect on the soluble expression (p=0.02), while G, C, N and S showed significantly positive effects, with p-values of 0.01, 0.02, 0.03 and 0.03, respectively. When analyzing AA effect on total amount of soluble dAb, the combined composition of R, S, G, N, Y, C, Q and F showed a significant influence (p=0.0007). The most significant AAs were R (negative), S (positive) and G (positive), for which the p-values were 0.0008, 0.006 and 0.03, respectively (Table 2). Additionally, stepwise regression analysis of the features of the dAbs, including charge, polarity, hydrophobicity, acidity and alkalinity, showed that polarity was the most important feature that had a positive influence on dAb soluble yield (p=0.02).

A linear model was built using total soluble dAb yield data:y=−0.5810∙R−0.3711∙F+0.4278∙G+0.4233∙S+0.2783∙C+0.2737∙Q+0.2519∙Y+0.2267∙N /3/where y indicates the soluble expression score in %, R is arginine, F is phenylalanine, G is glicine, S is serine, C is cysteine, Q is glutamine, Y is tyrosine and N is asparagine.

The higher the score, the higher the soluble expression level of dAb. Clustering results divided the sequences of the 65 experimental subjects and the control dAb into high- and low-expression groups; the score distribution is shown in Fig. 2. Twenty out of 25 dAbs belonging to the low-expression group had a score <2.5, while 31 out of 41 high-expression dAbs had a score >2.5. We conclude that dAbs with a score <2.5 are likely to be expressed at a low level in soluble form and the soluble yield would possibly be <(2.4±0.9) μg/g. On the other hand, dAbs with a score >2.5 are likely to be expressed at a high level in soluble form, with the potential soluble yield higher than (4.0±0.5) μg/g.

Using the same cultivation and detection methods as in the experiments above, expression data were obtained for 10 verification subjects and a control. Comparing the predicted expression levels from the model with the actual soluble yield of these dAbs, the accuracy of the prediction model was 80% (Table 6).