Selection and characterization of bispecific ADAPT molecules for enhanced biodistribution in cancer therapy

(1)

EXAMENSARBETE INOM BIOTEKNIK, AVANCERAD NIVÅ, 30 HP

STOCKHOLM, SVERIGE 2020

Selection and characterization of bispecific ADAPT molecules for enhanced biodistribution in cancer therapy

And a validation of library design JESPER BORIN

KTH

SKOLAN FÖR KEMI, BIOTEKNOLOGI OCH HÄLSA

(2)

(3)

i

ABSTRACT

Established biopharmaceuticals such as antibodies and derivatives thereof are relatively large.

In cancer therapy, this creates a steep drug concentration gradient within tumors, leaving cells far from blood vessels effectively untreated. Continuous pseudo treatments should foster the development of drug resistance and might lead to eventual disease relapse. Drug

concentration gradients can be operationalized as tissue penetration efficiencies, which are functions of molecular size. However, small particles are also subject to potent renal clearance, collapsing the therapeutic window beyond clinical applications. In this master’s thesis, spatial bispecificity was engineered into a single albumin binding domain (ABD).

Resulting ABD derived affinity proteins (ADAPTs) are saved from urinary excretion by the grace of HSA, but in the more static microenvironment of tumors, following HSA

dissociation, they are capable of tissue penetration efficiencies bestowed only upon smaller particles. To this end, phage display was used to raise ADAPTs against the cancer associated proteins human epidermal growth factor receptor 2 (HER2) and carcinoembryonic antigen- related cell adhesion molecule 6 (CEACAM6), but also the inflammation marker C-reactive protein (CRP). Via Sanger sequencing, 9 variants were picked for protein production and characterization, among which two spatially bispecific binders were found. ADAPTs were also evaluated for aggregation tendencies, structural conformity to library design, and thermal stability. One ADAPT, binding HER2, passed all tests of initial characterizations. Deep sequencing was used to analyze selection output, from which many more binders should be screened in future experiments.

KEY WORDS

ABD; ADAPT; HER2; CEACAM6; HSA; Phage display; Cancer therapy; Spatial bispecificity; Half-life extension; Biodistribution

(4)

ii

SAMMANFATTNING

Etablerade bioläkemedel liksom antikroppar och deras derivat är relativt stora protein. Som cancerterapeutiska skapar de således branta koncentrationsgradienter utgående från

tumörpenetrerande blodkärl. Detta riskerar att lämna vissa cancerceller utanför det

terapeutiska fönstret. Det svaga selektionstryck som således verkar i tumörperiferin fostrar cancerceller till att utveckla resistens mot detsamma. Koncentrationsgradienten beror på proteinets vävnadspenetrarande förmåga, vilken är en funktion av proteinets storlek. Mindre proteiner borde därmed lättare ackumuleras i hela tumören och förebygga resistensutveckling.

Problemet med små proteiner är deras mycket korta halveringstid i serum, en följd av relativt obehindrad filtrering ut i urinen via njurarna. I det här examensarbetet utvecklades

rumsbispecifika bindare av cancerassocierade protein med hjälp av fagdisplayselektioner från ett proteinbibliotek baserat på en enda albuminbindande domän (ABD). Resulterande ABD deriverade affinitetsprotein (ADAPT) undkommer ovan nämnda filtrering tack vare sin naturligt starka interaktion med humant serumalbumin (HSA). I den mer långsamt flödande tumörmikromiljön tillåts ADAPTerna efter albumindissociation sedan utöva en bland bioläkemedel överlägsen vävnadspenetration. Tre parallella selektionsspår utfördes mot de cancerassocierade målproteinerna human epidermal growth factor receptor 2 (HER2) och carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6) samt den utsöndrade inflammationsmarkören C-reaktivt protein (CRP). Via Sangersekvensering kunde flera

kandidater identifieras. Bland 6 karakteriserade ADAPTer uppvisade samtliga hög HSA- affinitet, tre konstaterades interagera specifikt med sitt målprotein, och två verkade binda även rumsbispecifikt. ADAPTer utvärderades även för sin benägenhet att bilda aggregat, strukturell överensstämmelse med experimentell design, och värmestabilitet. Endast en bindare, mot HER2, klarade sig genom alla prövningar som proteinkarakteriseringen innebar utan underkänt. Även en högparallel sekvensering utav selektionsresultat utfördes, men utanför de tidsramar som tillät ytterligare karakterisering.

(5)

iii

INTRODUCTION

Background

Ever since the invention of recombinant gene technology in the 70’s, interests within the industry of drug development have shifted from small synthetics to proteins, and antibodies in particular. Antibody based cancer drugs have been extensively utilized in both diagnostic and therapeutic applications [1, 2]. Especially in therapy, antibodies are generally a good option to other affinity proteins due to their inherent synergistic effects with the immune system and an excellent half-life. Antibodies can recruit the immune system via mechanisms initiated by the recognition of the fragment crystallizable (Fc) domain by Fc-gamma receptors (FcγR) on effector cells [3]. Besides means for targeted cytotoxicity or cytostasis, a good therapeutic should have a long distribution half-life. This reduces the minimum effective dose (MED) since the likelihood of disease specific accumulation increases. Antibodies naturally have a long half-life due to their size (150 kDa) limiting the rate of glomerular filtration. Molecules with a hydrodynamic radius smaller than 5 nm, which corresponds to approximately 50 kDa for globular proteins, experience a great increase in renal clearance rate compared to larger molecules [4]. Antibodies have their half-life further increased by a pH-dependent affinity to the neonatal Fc receptor (FcRn), which enables escape from lysosomal degradation. Upon endocytosis and subsequent endosomal acidification, antibodies can bind FcRn in the vesicle membrane. The FcRn then sorts itself, and the antibodies by extension, into a budding recycling endosome, retrieving the antibodies to the physiological pH outside the cell, where the antibody is released back into circulation [3].

However, in cancer medicine, potency is related to drug concentrations around the tumor cells. The disorganized vascularization of tumors results in sporadically large distances in between vessels. This promotes the need for good tissue penetration of a drug to effectively treat all target cells. Poor lymphatic drainage limits the rate of fluid convection in the extravascular space. Therefore, the penetration of particles is mainly dependent on their diffusion coefficients. Because tissue penetration efficiency then becomes inversely correlated with molecular size [5, 6], the drug concentration gradient within tumors will generally be steeper for antibodies than for smaller drugs. While the top layers of tumor cells might succumb to antibody therapy, cells experiencing an apparently lower dose and effect have an opportunity to proliferate and develop therapy resistance. Hence, one barrier to greater cancer drugs is size. Granted, tumor penetration is influenced also by target affinity and rate of internalization [7]. These properties can however be manipulated via protein engineering.

It should also be mentioned that the permeability of particles through the mucosa of the nasal cavity is directly size dependent [8]. The absorption of macromolecules through the lungs is also possible, but here the relationship between bioavailability and size is not as clear [9].

This enables the use of alternative routes for drug administration of small proteins compared to traditional biopharmaceuticals, which are still limited to intravenous or subcutaneous injections.

(7)

2

One problem with small protein drugs is poor distribution half-lives due to accelerated renal clearance [4]. This can divert treatment effects to renal cells, which normally decreases TD50

values and maximal therapeutic dosages [10]. It is however possible to design a drug

benefiting from long half-life and good tissue penetration simultaneously, e.g. a small protein with bispecificity for albumin and a disease target [11, 12]. The distribution half-life of human serum albumin (HSA) is 19 days. This is partly due to low glomerular filtration, and partly due to exploitation of the FcRn rescue pathway [13]. Moreover, albumin accumulates in tumors, hence positively contributing to both half-life extension and tumor retention. This can be attributed to a multitude of reasons. For example, tumor vasculature is generally porous and disorganized. Combined with poor lymphatic drainage, this facilitates an

extravascular accumulation of macromolecules, including albumin, a phenomenon called the enhanced permeability and retention effect (EPR). Moreover, the Serum protein, acidic and rich in cysteine (SPARC) binds serum albumin and is overexpressed in different cancer species [14]. SPARC affinity might further increase tumor retention of albumin due to an increase in protein complex size. These mechanisms of albumin mediated tumor retention are complemented by the disease target affinity, especially in a spatially bispecific setting.

Simultaneous binding negates an intramolecular competition of specificities. It also allows for a synergistic cooperation were the albumin mediated accumulation enables a larger pool of proteins to compete for disease target binding than would be the case if bispecificity was only temporal. Importantly, the drug has the ability to dissociate from albumin and begin to

exercise its superior tissue penetration to a greater extent than without the distribution help of bispecificity. It is the combination of these opposing physical properties, long half-life and good tissue penetration, in one molecule that promises improved drug efficacy compared to market competition.

Project

In this project, a phage library of small albumin binding domain (ABD) derived affinity proteins (ADAPTs) will be used for selections of binders to human epidermal growth factor receptor 2 (HER2), carcinoembryonic antigen-related cell adhesion molecule 6 (CEACAM6) and C-reactive protein (CRP) via phage display. The ADAPTs will preferably have retained albumin affinity while simultaneously binding the target to maximally benefit from the flexible pharmacokinetic properties of small bispecific proteins. The selected candidate binders will be screened against the desired product profile by evaluations of binding kinetics, structure, and stability.

Library

As previously stated, engineering spatial bispecificity into a small affinity scaffold is

preferable to a large one. The modified ABD used in this library is merely 54 amino acids in length. However, when the size decreases, the risk of sterical hindrance inhibiting

simultaneous binding increases. Earlier attempts to create such binders have only achieved temporal bispecificity [12, 15, 16]. It was reasoned that moving the target binding site to the other side of the ADAPT molecule should reduce the likelihood of sterical interference between albumin and target (Figure 1). 11 randomized positions in helix 1 and 2 were thus

(8)

3

selected for trinucleotide synthesis excluding all cysteine, proline, and glycine amino acids due to the possibility of unwanted helix breaks and protein dimerization.

A good scaffold used for engineering of affinity proteins should be highly stable to tolerate the introduction of many mutations. Functionally distinct bacterial membrane proteins, including orthogonal ABDs, have been shown to display high degrees of sequence homology to each other. A couple such ABDs were structurally characterized to uncover a

thermodynamically stable three helical fold, binding to HSA by interactions involving helix 2 and a connecting loop [17]. The third albumin binding domain of streptococcal protein G from strain G148 (G148-GA3) is one albumin binding domain that has since been subjected to many different protein engineering approaches [18, 19, 20]. Gulich et. al. developed an ABD with increased alkaline and thermic stability (ABD*) [19], and Jonsson et. al. matured G148-GA3 to femtomolar affinity (ABD035) via phage display [20]. A previous ADAPT library was constructed with 11 mutated surface exposed positions in helix 1 and 3 of ABD*

[21]. Despite the albumin binding being restricted to helix 2, produced binders experienced decreased HSA affinity [12]. The current ADAPT library instead focuses the randomization of amino acids on helix 1 and 2. Because of the increased risk of damaging the albumin affinity, ABD035 was chosen as the library scaffold, but with an N-terminal leader sequence shown to increase expressivity in Escherichia coli.

0 50100 150 0 50 100 150

Data 1

1 2 500 100150

Data 1

1 2 500 100150

A

C

B

Figure 1. 3D views of ABD and HSA protein structures, generated from PDB files 1TF0 and 1GJT via Chimera.

(A) The ALB8-GA ABD as complexed with HSA. (B) The G148-GA3 ABD depicted here can be assumed to retain the ALB8-GA mode of albumin binding due to their high sequence (59 %) and structure homology.

Residues numbered 1 to 11 represent the surface exposed residues of the helix 1 and helix 2 interface randomized in the current ADAPT library. In (C) is the amino acid sequence of the current library scaffold ABD035. Residues marked with black dots correspond to the numbered positions in (B).

(9)

4

Target validation

HER2 is part of a larger family of 4 epidermal growth factor receptors (EGFRs) capable of dimerization following ligand binding. HER2 seems to lack a natural ligand but contains an intracellular tyrosine kinase domain that auto phosphorylates upon ectodomain dimerization with other members of the EGFR family. This induces transduction signaling pathways connected to proliferation and invasiveness. HER3, as an example, exists in its inactivated form until a ligand binds, inducing dimerization receptivity. HER2 on the other hand is always activated, why it is the most common partner in dimerization events. It is

overexpressed in 15-30 % of breast and ovarian cancers but expressed at low or undetectable titers in normal somatic cells [22, 23], which makes it ideal as a therapeutic target. HER2 is also implicated as a potential target in a range of other cancer types [23, 24]. Among approved therapeutics for HER2 positive breast cancer treatment are both antibodies and small molecule drugs [25]. Treatments with the most successful therapy sometimes fail, and many patients quickly develop resistance. Many pathways of resistance have been described [26, 27]. To prevent resistance, HER2 positive therapies with different mechanisms of action can be combined [25]. Since no small protein drug has yet been approved, the contribution of deep tumor treatment and biopharmaceutical grade specificity of future ADAPTs to existing therapies might further increase overall survival.

CEACAM6 is a membrane anchored, heavily glycosylated cell adhesion protein expressed in epithelial and myeloid tissues. All members of the CEA protein family have one or two terminal IgV-like N domains and a number of constant domains, termed A and B. Their structural similarities facilitates a broad heterotypic binding repertoire, also encompassing proteins outside of the CEA family. Homotypic or heterotypic interactions with CEACAM6 activate the AKT/PI3K signaling pathway, important for regulation of intercellular

organization and differentiation. CEACAM6 is upregulated in a range of epithelial and myeloid cancers. It imparts cancer cells with increased invasiveness via degradation of the surrounding extracellular matrix (ECM), abnormally regulated cellular polarity and

differentiation, and anoikis resistance [28, 29]. Anti-CEACAM6 mAb binding mediated inhibition of dimerization was found to reverse the invasive phenotype in vitro [30].

Moreover, CEACAM6 functions as an independent predictor of overall-, and disease-free survival in colorectal cancer (CRC) patients [31]. In addition to the aforementioned mechanisms of action, CEACAM6 over-expression in pancreatic ductal adenocarcinoma (PDA) and breast cancer confer gemcitabine and tamoxifen resistance [32, 33]. CEACAM6 over-expression was identified across breast cancers of all four major categories and in 67 % of HER2 positive cases. There is a linkage proposed where HER2 could induce CEACAM6 transcription via the TGF-β pathway [29]. Furthermore, CEACAM6 and HER2 are now thought to interact directly with each other, were CEACAM6 might inhibit HER2

internalization and Trastuzumab efficacy [34]. This suggests HER2 and CEACAM6 could be used as targets for combination therapies in the future. CEACAM6 targeted therapies include siRNA gene silencing, mAbs, scFvs and ADCs [29]. Because of size advantages, there might be a gap in the market of also CEACAM6 targeted therapies befitting the ADAPT molecule.

(10)

5

CRP was chosen not because of its potential as a target for cancer therapeutics, but because of its previous success as a target protein in phage display in house. It was reasoned that

including CRP in selections would serve as an indicator on a successful selection and in addition deliver binders that could be developed as therapeutics. CRP is an acute-phase inflammatory protein secreted by hepatocytes as a homopentamer. It has diverse roles including tissue damage control and pathogen recognition. It can bind membrane

polysaccharides on one side and recruit the complement system by interactions with C1q on the other. Microenvironments of inflammation can induce structural changes in pentameric CRP (pCRP) causing its dissociation into monomers (mCRP) able to interact with FcγR to recruit circulating leukocytes at a higher rate. Transcription is triggered by the release of common cytokines such as IL-6 and TNF-α [35]. Because of its central role in the innate immune response, and its stable circulatory presence, CRP has become popular for the detection of infections. More recently, serum CRP levels have been investigated for clinical significance in disease related to chronic inflammation, e. g. cardiovascular disease (CVD), where mCRP was found to have not only a robust risk predictive power but also a

participatory role [36]. Elevated CRP concentrations provokes tissue-factor expression in blood monocytes. This might serve a causal role in vascular complications [37]. Other conditions of chronic inflammation where CRP was discovered to be valuable as a risk predictor include type 2 diabetes mellitus (T2DM), age related macular degeneration (AMD) and hemorrhagic stroke [36]. There are therapies for CRP inhibition under development, including the small molecule drug 1,6-bisPC, which inhibits CRP pro-inflammatory mechanisms by occupation of the phosphocholine (PC) binding site. No severe side effects were observed in animal trials [37]. However, the high evolutionary conservation of the CRP gene flags for consequences of abrogating the acute-phase response too much. Ideally, therapeutics should be administered in dosages optimized for mitigation of the long term serum CRP elevation present in chronic diseases while still allowing for the massive acute- phase production of the protein.

An ADAPT raised against CRP would be able to compete with small molecule drugs for treatment because of a putative specificity increase and a favorable administration routine due to its long half-life. The ADAPT might also bring advantages in detection over current

techniques using either methods of light scattering or antibody detection [38]. Affinity based detection is obviously more specific than morphology based detection in heterogenous blood sera, and if the ADAPTs could replace any antibodies in detection assays the comparatively easy production would improve the cost to benefit ratio.

MATERIALS AND METHODS

Evaluation of biotinylation of target proteins

Selection target proteins HER2 (Sino Biological), CRP (produced in house) and CEACAM6 (Sino Biological) were evaluated for biotinylation degrees and stoichiometry of binding to streptavidin (SA) coated Dynabeads^TM M-280 (Invitrogen) via sodium dodecyl sulphate polyacrylamide gel electrophoresis (SDS-PAGE). Biotinylated protein was washed in 25 µl phosphate buffered saline with 0.05 v/v% Tween 20 (PBST) supplemented with 0.25 mg SA-

(11)

6

beads by 15 min incubation on a rotary mixer in four cycles, after which followed an additional cycle with 200 µl PBST and 2 mg beads. After each wash, beads were separated from supernatants which were carried onto the next wash cycle. Beads were washed with PBST, isolated, and resuspended in reducing 2 w/v% SDS, 62.5 mM TRIS, 50 mM TCEP, 10 v/v% glycerol, 2w/v‰ bromophenol blue, SDS-PAGE loading dye. The final supernatant was dried from heating at 95 °C before treated with the same loading dye. Following thermal denaturation at 70 °C, samples were loaded onto a mini PROTEAN® TGX^TM gel (Biorad) and run according to the manufacturer’s recommendations. All beads were twice previously washed with PBST.

Phage display selections

An M13 phage library of 2.45x10¹² colony forming units (CFUs) per selection track was precipitated into phosphate buffered saline (PBS) with 0.1 v/v% Tween 20 and 0.1 w/v%

Gelatin (selection buffer) using 2 w/v% polyethylene glycol 8000 with 1.5 w/v% NaCl (PEG/NaCl). The library was negatively selected against 0.5 mg SA-beads via a 1 h incubation followed by isolation of unbound phages in the supernatant. All beads had been washed with 1 ml PBS and 0.1 v/v% Tween 20 (wash buffer) twice before blocking in 1 ml wash buffer supplemented with 0.5 w/v% Gelatin (block buffer) for >30 min. Biotinylated target protein pooled with phage library, and HSA in rounds 3 and 4 (Sigma Aldrich), to 1 ml in selection buffer was incubated for 2 h at 500 rpm after which phage-target complexes were mixed with beads for 30 sec at 1500 rpm, collected during 10 min and washed for 30 sec at 1500 rpm, all done according to Table 1 using a TANbead magnetic rotary mixer (Taiwan advanced nanotech). Phages were eluted with 50 mM Glycine at pH 2 (elution buffer) for 10 min and restored to neutral pH by a 50 v/v% mix with pH 8 0.1 M

Tris(hydroxymethyl)aminomethane-HCL in PBS (neutralization buffer) after bead removal on a magnetic rack. In selection round 1, target protein was coupled with beads and purified on a magnetic rack prior to panning. All incubations were carried out in room temperature (RT) in a rotary mixer. All 1 ml of the first round-, and 500 µl of subsequent panning eluates were used to infect XL-1Blue cells grown to 0.5<OD600<0.8 and an excess of 100 times phage CFUs in at least 5 ml tryptic soy broth (TSB) with 1 w/v% glucose and 10 µg/ml tetracycline, starting at an OD600 of 0.05-0.1 using an overnight (O/N) culture in TSB and 10 µg/ml tetracycline for inoculation. After 30 min incubation, the cell culture was diluted 1:2 in TSB and antibiotics to a final concentration of 100 µg/ml ampicillin and 10 µg/ml tetracycline after which cells were propagated for another 1-1.5 h. M13K07 helper phages were added to the cultivations at 10 times excess compared to library phages at infection, and cultures were incubated for 1.5 h at 90-120 rpm. Bacteria were pelleted via a 10 min centrifugation at 4000xg, resuspended in 50 ml TSB with yeast extract (TSB+Y), 100 µg/ml ampicillin, 25 µg/ml kanamycin and 0.1 mM isopropyl β-D-1-thiogalactopyranoside (IPTG) before transferred to a 500 ml E-flask for O/N incubation at 30 °C. All incubations of bacterial cultures were done in 37 °C at 90-150 rpm unless stated otherwise. Before dilution and centrifugation, culture samples were extracted and streaked onto agar plates with ampicillin, kanamycin, or no antibiotics to be incubated at 37 °C ON. Colonies were counted to estimate infection frequencies. 1 ml O/N culture was mixed with 240 µl 85 v/v% glycerol and stored in

(12)

7

-80 °C in preparation for upcoming deep sequencing. O/N cultures were centrifuged at 3500xg for 20 min whereby the supernatant phages were precipitated with PEG/NaCl on ice for 1-2 hours. Then followed 4500xg centrifugation and resuspension in 10 ml MilliQ (MQ) water before Ø0.45 µm filtration and another precipitation step. After yet another 4500xg centrifugation, the phage pellets were dissolved in selection buffer prior subsequent panning rounds. Precipitated phages not used in subsequent selections were diluted 1:2 in 85 v/v%

glycerol and stored in -80 °C. Precipitated and eluted phages were serially diluted 1:10 in duplicates along the rows of a 96 well plate, used to infect early log phase XL-1 Blue cells for 5 min and streaked onto ampicillin-, or 45 min and streaked onto kanamycin containing agar plates. After 37 °C O/N incubations, the colonies were counted to determine phage titers.

Table 1. A selection tree detailing target protein concentrations and number of washes per selection track and round. The selective pressure was increased throughout the selections by decreasing target protein

concentrations and increasing the number of washes. In round 3, all in tracks were bifurcated. Half of round 3 selections were performed in presence of 1.5 µM free HSA, adding an extra layer of selective pressure, negatively targeting library members incapable of spatial bispecificity.

Round 1

Her2 CRP CEACAM6

150 nM 150 nM 150 nM

2x wash 2x wash 2x wash

Round 2

Her2 CRP CEACAM6

100 nM 100 nM 100 nM

4x wash 4x wash 4x wash

Round 3

Her2 Her2+HSA CRP CRP+HSA CEACAM6 CEACAM6+HSA

50 nM 50 nM 50 nM 50 nM 50 nM 50 nM

6x wash 6x wash 6x wash 6x wash 6x wash 6x wash

Round 4

Her2 Her2+HSA CRP CRP+HSA CEACAM6 CEACAM6+HSA

25 nM 25 nM 25 nM 25 nM 25 nM 25 nM

12x wash 12x wash 12x wash 12x wash 12x wash 12x wash

Quality control of enriched library

Round 4 phage eluates were used to infect early log phase XL-1 Blue cells at a bacterial excess of 1x10⁵ for 30 min in 37 °C and subsequently spread on ampicillin containing agar plates. Single colony polymerase chain reaction (PCR) amplified library inserts were screened for correct sizes on a 1 % agarose gel supplemented with 0.01 v/v% GelRed for easy

nucleotide detection.

Enzyme linked immunosorbent assay (ELISA) experiments were run to assess binding enrichments on a population level. A clear and flat bottomed half area 96-well high binding polystyrene plate (Corning) was incubated O/N in 4 °C with human polyclonal IgG (phIgG, Star lab) diluted in 50 mM carbonate buffer of pH 9.6 (coating buffer) to a concentration of 0.075 µg/ml. The plate was washed twice with PBS and blocked for 2 h by PBS supplemented with 0.5 w/v% Casein (PBSC). PBST was used to wash the plate twice. Phage glycerol stocks from rounds 2 through 4 were diluted 0, 5, 10, 20 and 40 times from roughly 1x10¹¹ CFU/ml,

(13)

8

added to the ELISA plate in duplicates and incubated for 1 h. After three 10 min PBST washes anti-M13-HRP antibody (GE healthcare) suspended in PBSC was added and

incubated for 1 h. Three PBS washes later, TMB/H2O2 was added and incubated until color emerged. 2 M H2SO4 was added to 50 v/v% whereby reaction absorbances were measured at 450 nm wavelength using a Clariostar plate reader (BMG Labtech). The assay results were used to optimize the phage titers for absorbances within the linear range, to aid in

normalizations for concentration. The assay was repeated with phIgG, HER2, CEACAM6 and CRP coated wells using the optimized phage count. Target protein was coated in 4 °C O/N at 10 µg/ml in 30 µl coating buffer. All reactions were performed in duplicates. Assay reagents were added in volumes of 50 µl, and washes were done with 180 µl per well. All incubations were performed in RT unless otherwise specified.

Sanger sequencing and subcloning

48 single colony PCR amplified ADAPT inserts per track, provided they passed the quality controls, were PCR purified and Sanger sequenced (Eurofins). Recurring sequences were PCR amplified, PCR purified and restricted with AscI (NEB) and NcoI-HF (NEB) enzymes.

Cleaved fragments were then ligated into an E. coli expression vector carrying ampicillin resistance and a T7 promoter, inducible from IPTG inhibition of LacI. The vector had previously been restricted with the same enzymes and purified via a 2 % GTG agarose gel using a QIAquick® gel extraction kit (Qiagen). KCl, CaCl2, and MgCl2 competent Top10 E.

coli cells were transformed with the ligation mixtures by heat shocking for 45 sec at 42 °C and plated in the presence of ampicillin in 37 °C ON. Monoclonal ADAPT inserts were PCR amplified and analyzed on a 1 % agarose gel for correct amplicon sizes. 3 or 4 PCR products per construct passing the analysis were Sanger sequenced by Eurofins. Plasmids were

amplified and purified from relevant colonies as identified via Sanger sequencing, transformed into BL21* E. coli, and plated on ampicillin containing agar for O/N colony expansion.

Illumina MiSeq sequencing

Small scale TSB cultures supplemented with ampicillin were inoculated with cell glycerol stocks of interest isolated during phage display selections, grown in 37 °C at 150 rpm O/N and used as sources for QIAprep spin miniprep kit (Qiagen) mediated phagemid purification.

Library inserts were amplified via PCR, for 15 cycles with 50 ng template, to introduce sequencing indices unique to the corresponding track and round of selections. 5 pmol each of forward and reverse primers were used. The amplicons were extracted from a 2 % GTG agarose gel, using the same kit as during subcloning, and diluted to 10 nM in StAq. All samples were pooled and sent for sequencing using one flow cell in an Illumina MiSeq v2 instrument (National genomics infrastructure). Sequence logotypes were created with MATLAB functions multialign and seqlogo.

(14)

9

Protein production and purification

50 ml TSB+Y and 100 µg/ml ampicillin was inoculated with BL21* E. coli O/N culture of the same medium and incubated to an OD600 of ~0.8. IPTG was added to 1 mM where after cultures were incubated O/N in 25 °C. Cells were harvested by 8 min via a 4 °C 4000 rpm centrifugation in a JLA 16.250 rotor (Beckman Coulter), resuspended in 10 ml 25 mM TRIS, 1 mM EDTA, 200 mM NaCl, and 1 v/v% Tween 20 (TST) and sonicated with a Vibra Cell instrument (Sonics) for 1.5 min at 40 % energy output. Lysate was centrifuged for 10 min at 16000 rpm in a JA 17 rotor (Beckman Coulter). Supernatant was extracted, Ø0.45 µm syringe filtered and loaded onto HSA-sepharose packed PD-10 columns equilibrated with TST buffer.

Washing was performed with TST and 5 mM NH4Ac at pH 6. Proteins were subsequently eluted with 0.5 M HAc and analyzed via spectrophotometry. Relevant fractions were lyophilized in a scanvac coolsafe (Labogene) O/N, dissolved in PBS supplemented with 50 mM L-Arginine and pH adjusted to 9-10 by NaOH to aid in solvation, and measured for protein concentration in a nanodrop 1000 spectrophotometer. SDS-PAGE was employed on protein pools to assess purity and yield.

Structure and stability measurements

Molecular mass was evaluated on a 4800 MALDI-TOF/TOF analyzer (Applied biosystems).

Sample aliquots of 1 µl were mixed with equal volumes 5 mg/ml α-Cyano-4-

hydroxycinnamic acid (CHCA) dissolved in 50 % acetonitrile (ACN) supplemented with 0.1

% trifluoroacetic acid (TFA). Volatile solvents were evaporated on a matrix assisted laser desorption/ionization-time of flight (MALDI-TOF) metal plate before 1 µl more CHCA was added. The analysis was carried out upon complete solvent evaporation.

The secondary structure and its thermostability was measured using a Chirascan circular dichroism (CD) spectrophotometer (Applied biosystems). Protein was diluted, if applicable, to 0.2 mg/ml in 400 µl 1xPBS and scanned with light from 260 to 195 nm wavelength at 20 °C while monitoring ellipticity. Samples were subsequently heated to 90 °C at a rate of 5 °C/min while continuously monitoring ellipticity at light of 221 nm wavelength. Sample was cooled back to 20 °C before the scanning light spectrometry was repeated.

Binding kinetics screening

Binding kinetics of ADAPTs interacting with HSA and biotinylated HER2, CRP and CEACAM6 were determined using a Biacore 8K instrument (GE Healthcare). Five

concentrations of ADAPTs serially diluted 1:2 in PBST downwards from 1 µM were pumped at 30 µl/min over intended target proteins and HSA immobilized via amine coupling

according to manufacturer’s recommendations to separate flow cells of the surface of an S series CM5 chip (GE Healthcare). Immobilization levels settled at 716.4, 286.7, 1233.2 and 1576.7 RU for HER2, CRP, CEACAM6 and HSA, respectively, after surface deactivation.

Association and dissociation times were set in a multicycle method to 120 and 300 s.

Regeneration was carried out using 10 mM HCl. A sensorgram reference of analyte flowed over an inactive flow cell, and a sensorgram blank of running buffer (PBST) pumped over an

(15)

10

active flow cell, i.e. immobilized with ligand, were subtracted from the binding data of each interaction to reduce background noise contributions. Sensorgrams were then fitted a binding curve simulating 1:1 kinetics to estimate rate constants and affinity. Surface saturation was calculated from sensorgrams of 500 nM analyte as actual maximal response unit divided by theoretical maximal response unit. ADAPTs were also evaluated for cross reactiveness with parallel selection targets at 500 nM.

Biosensor capture assay

ADAPTs diluted to 500 nM in PBST were flown over the same sensor chip used in the previously described binding kinetics screening at 30 µl/min for 180 s. 500 nM analyte was subsequently injected for 180 s after which followed a 300 s dissociation phase. Flow cells used where coated with HSA and biotinylated HER2 and CEACAM6. For HER2 and CEACAM6 surfaces, HSA acted the analyte, and vice versa. A reference inactive reference surface was used for subtraction of chip binding background, generated anew with each injection. Blank subtractions included ADAPT capture solutions and running buffer as analyte substitute, flowed across active surfaces. Sensorgrams were then fitted a curve simulating 1:1 kinetics to estimate rate constants and affinity.

Evaluation of aggregation

A Superdex 75 increase 5/150 GL column (GE Healthcare) was used with the ÄKTA pure purification system (GE Healthcare) according to manufacturer’s recommendations in evaluation of the different ADAPT’s aggregation tendencies. Void volume was determined from injections of blue dextran 2000. A calibration kit with four proteins (Conalbumin, Carbonic Anhydrase, Ribonuclease A and Aprotinin) of known molecular weight were

purified together and used to construct a standard curve used in correlations between partition coefficients and the base ten logarithm of molecular weights. Purified proteins were diluted to 0.4 mg/ml in PBS when applicable, syringe filtered Ø0.45 µm and applied to the column via a 50 µl injection loop at 0.45 ml/min. Running buffer was PBS.

RESULTS

Phage display selections and sequencing

Four rounds of equilibrium based phage display selections were performed against the biotinylated target proteins HER2, CRP and CEACAM6 in parallel. In the last two rounds of selections, free HSA was included to introduce a negative selective pressure for ADAPTs incapable of simultaneous HSA and target protein binding. A lift in phage survival rate was seen in round 3 for all tracks but HER2 with HSA (Table S1). Selection stringency was iteratively increased with the number of washes and reduction of target concentrations according to Table 1. A polyclonal phage ELISA was performed on samples from each selection round. Only HER2 and CRP tracks including free HSA failed to enrich for target binding (Figure 2). Interestingly, the HER2 binding capacity greatly increased in round 3 but decreased in round 4. The two most common HER2 mutants were highly enriched to round 3, but only the second most common mutant was depleted to round 4.

(16)

11

Besides ELISA, initial quality control of selection outputs included PCR screening for the correct size of library insert. Tracks conducted with HSA supplementation carried empty phagemids to a much greater extent than tracks void of HSA (Figure 3). 48 clones from each track exhibiting binding enrichments in the phage ELISA were Sanger sequenced to register 16 clusters of differentially repeated library variants (data not shown). The 9 most common variants were continued down the pipeline for subcloning and expression.

0 1 2 3 4 5

0 2 4 6

Selection round OD450 (nm)

HER2 HER2+HSA CEACAM6 CEACAM6+HSA CRP

CRP+HSA

Figure 2. A graph on polyclonal phage ELISAs using phage stocks from each selection round. Optical density at 450 nm wavelength (OD450) is normalized against phage titers. All but two tracks show a binding increase with advanced selection. The CEACAM6+HSA track is the sole HSA-track exhibiting a binding enrichment, reaching the second highest signals of the assay. HER2+HSA track OD450 signals broke the background noise for neither IgG nor HER2 wells. The last round of HER2 selections performed worse than its predecessor, seeming to enrich for functionalities other than target binding.

Figure 3. A DNA gel electrophoresis image showing amplicons from colony PCRs using vector specific primers.

Amplicons containing ADAPT inserts are expectedly 500 bp. Lanes 3 to 38 are representative results of the CEACAM6+HSA selections. Lanes 39 to 54 are typical amplicon sizes from non HSA selection tracks. The multiplicity of bands preset in wells with amplicons from empty phagemids might be due to partial

complementarity of primers inside the dummy fragment. However, since the sizes were still greater than that of ADAPT inserts, investigations into the matter were deemed unnecessary.

(17)

12

Illumina MiSeq deep sequencing generated thousands of reads per track and round.

Investigations into the amino acid diversity of randomized positions revealed a difference in degrees of residue conservation in between tracks (data not shown). Helix 1 contained the most highly conserved amino acids across all tracks. The CEACAM6 track displayed strong enrichments focused to select helix 1 residues. The HER2 track also exhibited impressive enrichment, but with a close to uniform diversity profile across randomized positions. Other tracks contained information up to 2 bits per position and maintained a much more complex composition of enriched amino acids across the protein. Furthermore, deep sequencing only partially mirrored the results of the Sanger sequencing, in that the most abundant mutant identified from deep sequencing only matched a characterized Sanger sequencing identified mutant in the case of the HER2 selection. Due to time constraints, additional constructs identified through deep sequencing could not be included in the protein characterization. The relatedness of variants within clusters was higher in HSA free selection tracks (Figure 4).

1 2 050 100150

1 2 500 100150

CRP

1 2 500 100150

HER2

CEACAM6 CEACAM6+HSA

CRP

A B

C D

Figure 4. Dendrograms depict UPGMA constructed phylogenetic trees using the blosum62 scoring matrix on deep sequencing data from the fourth selection rounds. Only repeated sequencing reads with mutations exclusively matching the intended positions were considered. (A – D) Selection tracks without HSA show short distances in between variants of the same cluster. The CEACAM6+HSA tree instead performs poorly when it comes to intra cluster relatedness. Sanger sequences were found again in deep sequencing clusters, but only represented by the most abundant variant in the case of the HER2 track.

(18)

13

Expression and purification

Four out of nine proteins were very poorly expressed, yielding concentrations undetectable via SDS-PAGE (Figure 5) and a total protein amount of less than 30 µg after purification.

ADAPTCEACAM6-3 was purified to ~2 mg per liter culture. The extant four ADAPTs reached yields of ~20 mg per liter culture. Only proteins with gel presence were continued through the entirety of characterization experiments.

Figure 5. An SDS-PAGE image displaying the wildly different expression levels of candidate binders. Mutants ADAPTHER2-1, ADAPTCRP-2, ADAPTCEACAM6-4, and ADAPTCEACAM6-6 all have strong gel presence. ADAPT6CEACAM6- 3 shows modest expression. No other ADAPTs are visible in respective elution pools. The presence of ADAPT in ADAPTCEACAM6-5 flow through (FT) is not distinguishable from background host cell proteins. One unidentified protein band above 10 kDa can be seen in tracks of ADAPTCEACAM6-3 and ADAPTCEACAM6-6 purifications.

Screening of binding kinetics

Purified ADAPTs were screened for selection target binding via multicycle SPR experiments where four variants bound (Figure 6 A – D). No protein of the HSA supplemented selections registered any target interaction (Figure 6 E). Due to issues with poor quality of fit between sensorgrams and proposed 1:1 kinetics the affinities could not be determined for

ADPAPTHER2-1 or ADAPTCRP-2. For ADAPTCRP-1 and ADAPTCEACAM6-3 affinities of 59 and 350 nM, respectively, were estimated. ADAPTCRP-1 being the exception, all interactions had very fast dissociation kinetics, in the range of 10^-1-10^-2/s. Comparing affinities operationalized as surface saturation, despite experiments not reaching equilibrium within the association phase, clearly distinguishes ADAPTHER2-1 from ADAPTCRP-2 as the better binder.

Furthermore, ADAPTs were crosschecked for reactivity with parallel selection targets, where lower levels of binding with HER2 was measured as compared to the ADAPTHER2-1 HER2 interaction (9 vs. 22 % surface saturation). In contrast, many ADAPTs raised against HER2 or CEACAM6 displayed relatively strong responses when flowed over CRP despite a similar

1 2 500 100150

Data 1

ADAPT^H^E

R2-1

ADAPT^C^R

P-1

ADAPT^C^R

P-2

ADAPT^C^E

ACAM6-1

ADAPT^C^E

ACAM6-2

ADAPT^C^E

ACAM6-3

ADAPT^C^E

ACAM6-4

ADAPT^C^E

ACAM6-5

ADAPT^C^E

ACAM6-6

FT ADAPT^C^E

ACAM6-5

wash ADAPT^C^E

ACAM6-5

(19)

14

immobilization level to the HER2 surface (18 % surface saturation). All ADAPTs maintained a strong nanomolar or sub nanomolar affinity to HSA (Figure 7).

0 100 200 300 400 500

0 5 10 15 20

ADAPT_HER2-1

Time (s)

Response units

0 100 200 300 400 500

0 5 10 15 20 25

ADAPT_CRP-1

Time (s)

Response units

0 100 200 300 400 500

0 2 4 6 8

ADAPT_CRP-2

Time (s)

Response units

0 100 200 300 400 500

0 20 40 60 80 100

ADAPT_CEACAM6-3

Time (s)

Response units

0 100 200 300 400 500

0.0 0.2 0.4 0.6 0.8

ADAPT_CEACAM6-4

Time (s)

Response units

0 100 200 300 400 500

0 1 2 3 4

ADAPT_CEACAM6-4

Time (s)

Response units

A

D

B

E

C

F

Figure 6. Sensorgrams are generated from multicycle SPR experiments with analyte (ADAPT) diluting downwards from 1 µM (or 500 nM in (B)). Dotted curves display the best 1:1 kinetic fittings. (A – E) The immobilized protein is the intended selection target for each ADAPT. (A – D) Four ADAPTs displayed binding to their respective selection target proteins. Rate constants were difficult to calculate due to a poor 1:1 binding fit in all cases but ADAPTCEACAM6-3, with ka=8.19x10⁵ M^-1, kd=2.86x10^-1 s and a resulting affinity of 350 nM.

ADAPTCRP-1 was fitted a ka=2.98x10⁴ M^-1 and a kd=1.75x10^-3 s, amounting to an affinity of 59 nM. Employing surface saturation evaluations, for the relative affinity of binders, without extrapolation produces results of approximately 22, 50, 7 and 36 % for ADAPTHER2-1, ADAPTCRP-1, ADAPTCRP-2, and ADAPTCEACAM6-3, respectively.

The saturation of ADAPTs cross reacting with other selection target proteins were as low as 5 % with a mean of 9 % for HER2. Cross reactivity against CRP was higher, with a minimum response of 5 % and mean response of 18 % (Figure S1). Both supposed CEACAM6 binders originating from the HSA selections did not show

significant binding signals. In (E), ADAPTCEACAM6-4 shows representative binding signals under 1 RU. (F) ADAPTCEACAM6-4 sensorgrams from interactions with immobilized HER2 (one sensorgram) and CRP (one sensorgram) show higher binding signals compared to ADAPTCEACAM6-4 when interacting with its intended target.

(20)

15

0 100 200 300 400 500

0 50 100 150

ADAPT_HER2-1

Time (s)

Response units

0 100 200 300 400 500

0 10 20 30 40

ADAPT_CRP-1

Time (s)

Response units

0 100 200 300 400 500

0 50 100 150

ADAPT_CRP-2

Time (s)

Response units

0 100 200 300 400 500

0 50 100 150

ADAPTCEACAM6-3

Time (s)

Response units

0 100 200 300 400 500

0 50 100 150

ADAPTCEACAM6-4

Time (s)

Response units

0 100 200 300 400 500

0 50 100 150

ADAPTCEACAM6-6

Time (s)

Response units

A

D

B

E

C

F

Figure 7. All characterized ADAPTs had retained strong affinities towards the immobilized HSA. KD was calculated to 45 pM, 1.2 nM, 0.16 nM and 0.25 nM for ADAPTHER2-1, ADAPTCRP-1, ADAPTCRP-2, and ADAPTCEACAM6-3, respectively. Dotted curves display the best 1:1 kinetic fittings.

Screening for spatial bispecificity

Target binding protein of sufficient concentration (ADAPTHER2-1 and ADAPTCEACAM6-3) were analyzed for spatial bispecificity in an SPR capture assay with immobilized HSA, ADAPTs in capture solution and selection targets acting the analytes. The fitting of rate constants to the binding curve was suboptimal, disallowing credible rate constant estimates. Both binders produced significant response levels upon analyte injection compared to blank injections.

Negative controls for binding between selection targets and HSA showed no binding (Fig 8).

(21)

16

0 200 400 600

0 2 4 6 8

ADAPT_HER2-1

Time (s)

Response units

0 200 400 600

0 5 10 15 20 25

ADAPT_CEACAM6-3

Time (s)

Response units

0 200 400 600

0.0 0.2 0.4 0.6 0.8 1.0

HER2 & CEACAM6

Time (s)

Response units

0100 200

300 400

500 100 2030 40

captureprinciple

A B

C D

Figure 8. A capture assay in which ADAPTs were flown over immobilized HSA to establish a baseline for analyte (target protein) binding was conducted. Both ADAPTHER2-1 and ADAPTCEACAM6-3 displayed simultaneous binding to their respective target proteins and HSA. In (A) and (B) the reference and blank subtracted

sensorgrams after analyte injections show selection target binding to captured ADAPTs. (C) No HER2 or CEACAM6 interactions with HSA were detected when HSA was flown over HER2 or CRP. The assay principle is illustrated in (D), where HSA is the immobilized ligand, ADAPT is the capture reagent and selection target (HER2 or CEACAM6) is the analyte. This was the setup for (A – B). In (C), the selection target was the immobilized ligand and HSA the analyte.

Structure and stability

All highly expressing ADAPTs displayed CD spectra typical for protein with high alpha helical content (Figure 9). The ellipticity background to noise ratio of the more modestly expressed ADAPTCEACAM6-3 was insufficient for highly resolved secondary structure or melting temperature determination. However, a rough alpha helical spectral profile can be seen in Figure 9 C. ADAPTHER2-1 was alone in its ability to completely refold post thermal denaturation (Figure 9 A), during which a melting temperature of 62 °C could be measured (Figure 9 F). All protein analyzed via MALDI TOF MS registered with experimental molecular weights close to theoretical values (Figure 11).

Selection and characterization of bispecific ADAPT molecules for enhanced biodistribution in cancer therapy