Single Cycle Structure-Based Humanization of an Anti-Nerve Growth Factor Therapeutic Antibody

Most forms of chronic pain are inadequately treated by present therapeutic options. Compelling evidence has accumulated, demonstrating that Nerve Growth Factor (NGF) is a key modulator of inflammatory and nociceptive responses, and is a promising target for the treatment of human pathologies linked to chronic and inflammatory pain. There is therefore a growing interest in the development of therapeutic molecules antagonising the NGF pathway and its nociceptor sensitization actions, among which function-blocking anti-NGF antibodies are particularly relevant candidates. In this respect, the rat anti-NGF αD11 monoclonal antibody (mAb) is a potent antagonist, able to effectively antagonize rodent and human NGF in a variety of in vitro and in vivo systems. Here we show that mAb αD11 displays a significant analgesic effect in two different models of persistent pain in mice, with a remarkable long-lasting activity. In order to advance αD11 mAb towards its clinical application in man, anti-NGF αD11 mAb was humanized by applying a novel single cycle strategy based on the a priori experimental determination of the crystal and molecular structure of the parental Fragment antigen-binding (Fab). The humanized antibody (hum-αD11) was tested in vitro and in vivo, showing that the binding mode and the NGF neutralizing biological activities of the parental antibody are fully preserved, with even a significant affinity improvement. The results firmly establish hum-αD11 as a lead candidate for clinical applications in a therapeutic area with a severe unmet medical need. More generally, the single-cycle structure-based humanization method represents a considerable improvement over the standard humanization methods, which are intrinsically empirical and require several refinement cycles.


Introduction
The neurotrophin Nerve Growth Factor (NGF) [1] exerts a wide range of physiological functions not only in the development and maintenance of specific neuronal populations of the vertebrate nervous system [2], [3], but also in some non neuronal cells, including cells of the immune system such as mast cells, basophils and monocytes [4]. It is worthy of note that, besides its broad range of physiological effects, NGF is involved in several disease states such as in certain chronic inflammatory or neuropathic pain states [5], [6] and in several human malignancies [7].
There has been an increasing recognition that NGF regulates the function of adult peripheral sensory neurons including smalldiameter nociceptive afferents, thereby exerting a pain modulation activity through nociceptor sensitization [8]. Interestingly, NGFinduced activation of the Tropomyosin-related receptor kinase A (TrkA) receptor on mast cells as well as on macrophages and monocytes recruited at an injured or inflamed site determines the release of mediators that further contribute to the sensitization of sensory nociceptors [6]. Therefore, NGF modulates pain responses and changes pain thresholds by two principal mechanisms: a direct TrkA-mediated activation of pain signaling through receptors and channels on nerves such as Transient Receptor Potential cation channel subfamily V member 1 (TRPV1) and TetrodoToXin (TTX) insensitive voltage-gated sodium channel Na v , and indirectly through the TrkA mediated degranulation of mast cells and basophils.
Thus, the NGF-TrkA system appears to be a master control system for pain, in spreading inflammation and increasing the electrical neuronal response in nerve endings, functionally placed upstream in the hierarchy of the pain regulation process.
Besides a large body of evidence in animal models, the clinical relevance of the functional role of the NGF-TrkA system in pain has received considerable and compelling validation in humans.
First of all, increased NGF levels are found in inflamed tissues and fluids from patients with pathological conditions such as arthritis [9], pancreatitis and prostatitis [10]. In humans, exogenous NGF infusions either locally or systemically, induce pain [11]. Finally, humans harboring mutations in the NGFB [12], [13] and TrkA genes [14] suffer congenitally from a complete loss of pain sensations, leading to severe self-mutilation.
For all these reasons, there has been a great interest in the development of antagonists of NGF as analgesic drugs for chronic and inflammatory pain conditions [15] such as osteoarthritis [16]. In this respect, antibodies against NGF constitute the strategy of choice to antagonize the actions of NGF, ever since the pioneering ''immunosympathectomy'' experiments by Levi-Montalcini [1], [17]. Indeed, the potent analgesic effects of anti-NGF antibodies have been well documented in a variety of animal pain models [6].
In this study we demonstrate the potent and remarkably long lasting analgesic activity of the mAb aD11 on different rodent models of tonic/chronic pain. In order to pursue its therapeutic development, mAb aD11 was humanized by a novel strategy, exploiting the a priori 3D crystal structure determination of the parental rat Fab aD11 (PDB_ID: 1ZAN) [26], [27]. This resulted to be a crucial approach, that allowed to humanize aD11 antibody variable regions, by Complementary Determining Regions (CDRs) grafting, in a single cycle, obtaining a humanized version (hum-aD11) whose binding characteristics and NGF antagonizing activity, both in vitro and in vivo, are fully preserved with respect to the parental rat counterpart.

Results
Mab aD11 binds human NGF equally well as mouse NGF A basic requirement for mAb aD11 to be employed in human clinical applications is that its binding affinity for hNGF should be comparable to that for rodent NGF. The epitope recognized by mAb aD11 was identified in Loops I and II of mNGF. While the sequences of rat and human NGF in this region are identical, for the mouse and human NGF they differ at position 40 ( Figure 1A). A structural alignment of Loop I and Loop II of the mouse and human NGF crystallographic structures (PDB_ID: 1BTG, PDB_ID: 1WWW) [28], [29] respectively, shows a good superposition ( Figure 1B).
Thus, we can reliably predict that mAb aD11 binds to hNGF equally well as to mNGF. Indeed, an ELISA assay, with solidphase coated mNGF and hNGF and serial dilutions of mAb aD11, confirms that mAb aD11 recognizes hNGF and mNGF with a comparable affinity ( Figure 1C). At the functional level, the potency of mAb aD11 to neutralize the activity of NGF from different species was ascertained by the TF-1 cell proliferation assay [33] exhibiting a similar concentration-dependent inhibition of cell proliferation for human, rat and mouse NGF, respectively (data not shown).
In vivo analgesic properties of anti-NGF mAb aD11 on formalin-induced pain and on neuropathic pain The antagonistic properties of mAb aD11 are well established, as this anti-NGF antibody is extremely effective at neutralizing the biological actions of NGF in a wide variety of in vivo systems [20], [21], [22], [23], [24], [25], thanks to its extremely high binding affinity [19] and epitope specificity [20], [27]. In order to confirm the therapeutic potential of the aD11 antibody, its analgesic properties were assessed in vivo on two different models of tonic/ chronic pain in mice.
In the formalin-induced inflammatory pain model, formalin injection resulted in the typical biphasic response with the highest peak after 5 min and a second phase of licking that started 15 min after the treatment. The mAb aD11 was administered, either as IgG or Fab fragment format, 45 min before formalin injection and showed a significant analgesic effect (Figure 2A) clearly specific for the second phase (late inflammatory phase, i.e. 15-40 min) of the pain response. The analgesic effect was superior for the mAb aD11 in the Fab format, by halving the response of persistent pain, as compared either to saline (p,0.01) or to control mAb treatment (p.0.05) (Figure 2A). The strong analgesic potency of Fab aD11 in relation to that of the whole IgG counterpart, may be due to its higher diffusion rate and hence greater tissue penetration and bioavailability.
The analgesic potency of mAb aD11 was further evaluated in a mouse model of neuropathic pain, the Chronic Constriction Injury (CCI) of the sciatic nerve [34], following two treatment protocols, a short and a long lasting protocol (see Materials & Methods). In both protocols ( Figure 2B and Figure 2C), mAb aD11 (Intra-Peritoneal injected (I.P.)) exhibited a very significant analgesic effect, as compared to mouse IgG mock. In a first set of experiments (short protocol) ( Figure 2B), four I.P. injections of mAb aD11 (from day 3 to day 6 after ligation of the nerve) were able to significantly reduce mechanical allodynia, starting from day 4 after surgery. On this basis, a second set of experiments with a longer observation period (long lasting protocol, observation up to 31 days following sciatic nerve ligature), was performed. The observation of animals undergoing long lasting protocol revealed a quite unexpected temporal profile for the strong analgesic activity induced by mAb aD11 ( Figure 2C). Two phases can be recognized in the analgesic action: the first one identifies a pharmacological effect of mAb aD11 (an effect which declined in parallel with the drop of the antibody concentration in circulating blood, reaching a minimum analgesic effect around day 17, i.e. one week after the end of the treatment). After the gradual decline of the anti-allodynic effect, in the second phase (from day 21 to day 31) mAb aD11 again reduced neuropathic pain, displaying a longterm analgesic effect of the anti-NGF antibody ( Figure 2C). This long lasting analgesic effect is likely to involve persistent changes in gene expression in sensory neurons, demonstrating that aD11 antibody is not just as a potent analgesic, but also a long-term disease-modifying drug.

Structure-based humanization design of anti-NGF mAb aD11
In order to advance its development towards clinical evaluation in patients, mAb aD11 was humanized by a novel structure-based strategy, taking advantage of the available 3D crystal structure of the rat Fab aD11 (PDB_ID: 1ZAN) [26], [27]. The structural information gained from the rat Fab aD11 crystallographic structure was exploited to optimize the selection of the acceptor human antibody framework regions (FWRs), onto which the CDRs of the donor murine anti-NGF mAb aD11 were grafted. The acceptor human FWRs for mAb aD11 humanization were selected by a novel approach [35], based on a comparison of both the primary (% sequence homology/identity) and the tertiary structures (degree of structural similarity based on the root mean square deviations (r.m.s.d) calculated by taking into account the C a skeleton atoms) of the parental antibody with all the available  [30], [31] of Loop I (in magenta) and Loop II (in green) from mNGF (PDB_ID: 1BTG, cyan) [28] and hNGF (PDB_ID: 1WWW, green) [29]. This Figure was prepared with PyMOL [32]. (c) ELISA assay with mNGF and hNGF coating (5 mg/ml) and serial dilutions of parental mAb aD11. The experiments were done in duplicate. doi:10.1371/journal.pone.0032212.g001 experimental crystal structures of human and humanized antibodies. A cluster of candidate human FWRs acceptors was identified, scored on the basis of the highest level of homology/ identity of the sequences present in the Protein Data Bank [36], for which high resolution X-ray structural data (i.e. not lower than 2.5 Å ) were available. The search was performed at first by considering the overall antibody variable regions (i.e. the F v fragment) and then by narrowing down to the FWRs. The selected crystallographic structures were superimposed to that of the parental F v aD11, using the ''superimpose'' software [37], and the r.m.s.d for each individual structural comparison were calculated, considering only the C a atoms at the correspondent positions, on the respective backbones, closer than 2.0 Å . Therefore, the selection of the optimal human FWRs acceptors for humanization was configured as a three-variable problem, resulting in a 3D plot that combined the information from the primary structure comparison (% sequence homology/identity both at the level of the F v or only of the FWRs) and from the tertiary structure alignment (the degree of structure similarity expressed in terms of r.m.s.d. and by the % of C a atoms employed in the r.m.s.d. calculations). Figure S1, the 3D distributions of the clusters were mutually consistent for all of the combinations of the variables taken into account. Moreover, by comparing the distances ( Figure 3) between the point of each of the selected acceptors to the one having the coordinates corresponding to 100% sequence homology/identity both at the level of the variable regions of the heavy (V H ) and of the light (V k ) chains F v or only of the FWRs, 0.00 Å r.m.s.d. and 100% of C a atoms employed in the r.m.s.d. calculations, i.e. the ideal human or humanized antibody, it was straightforward to unequivocally identify the best FRWs acceptor candidate among the human or humanized antibody of choice, both at the levels of the primary and tertiary structures. Thus, on the basis of the described method, the humanized antibody (PBD_ID: 1JPS) [38] was chosen as acceptor FWRs in the ensuing process of CDR grafting in the humanization of the aD11 antibody. The similarity of their FWRs is displayed at the level of their F v region, both by sequence alignment (Figure 4) and the 3D structural superimposition ( Figure 5A).

As shown in
In order to design the final humanized form of the aD11 antibody (hum-aD11), first the CDRs residues of the parental mAb aD11 (underlined in Figure 4) were combined with the  (15-40 min) in the course of the formalin test. Treatment consisted in saline (negative control) or antibody injection (single doses: 12.5 mg of mock mouse mAb or two different molecular formats of aD11, i.e. mAb or Fab) performed (in the same paw as for formalin) 45 min before formalin injection and testing. Statistical analysis was performed on each phase (ANOVA and Fisher's Test for comparison of each couple of groups). Each experimental group included at least 8 animals. (b) Analgesic effects of aD11 antibody in the short lasting protocol of neuropathic pain model: mAb aD11 significantly increased the value of ipsilateral/contralateral index (ratio between the threshold forces measured for the two hind paws, the one ipsilateral to surgery and the contralateral one. Mean value 6 s.e.), starting from day 4 to day 14, one week after the last antibody injection. Control mice were injected with either mock mouse IgG, (1.4 mg/Kg) or saline solution (sal). ANOVA test for repeated measures resulted in statistical significance for treatment (p,0.0001), time (p,0.0001) and the interaction between the two factors (treatment6time) (p,0.0001). (c) Analgesic efficacy of mAb aD11 (one dose: 2 mg/kg) in the long lasting protocol of neuropathic pain model. MAb aD11 increased the ipsilateral/contralateral index, starting either from day 5. The analgesic effect, which disappeared around days 17-19, increases again to reach a plateau between day 27 and day 31, identifying a late phase in the action of mAb aD11 (long-term effect). ANOVA test for repeated measures resulted in statistical significance for treatment (p,0.005), time (p,0.005) and the interaction between two factors (treatment6time) (p,0.005). doi:10.1371/journal.pone.0032212.g002 FWRs of PDB_ ID: 1JPS [38] (CDR grafting). A few retromutations were introduced, as follows: positions L46 and L87 were mutated (to the rat mAb aD11 residue at this position) to preserve the interface between V H and V k , whilst the H71 position was retro-mutated (ARR) to maintain the characteristics of the Vernier zones [39], [40]. Subsequently, following the comparison with the main consensus sequences of human immunoglobulins, a   [26], [27] with the selected template for humanization (PDB_ID: 1JPS) [38] and hum-aD11, obtained by CDRs grafting (highlighted in yellow) on PDB_ID: 1JPS [38] FWRs (highlighted in magenta) with the retro-mutations (highlighted in green) and the mutation (highlighted in cyan) . The six CDRs are underlined and the residues belonging to the Vernier zones are colored in red. The residues numbering is according to Kabat [39]. doi:10.1371/journal.pone.0032212.g004  [26], [27] in magenta (a) with the selected FWRs acceptor 1JPS [38] (in cyan) and (b) with the model of the resulting humanized antibody after CDRs grafting (whose FWRs are depicted in cyan, while its CDRs are in magenta); (c) Model of the hum-aD11 after CDR grafting and mutagenesis in the chosen FWRs. The residues of human and animal origin are highlighted in cyan and violet, respectively. doi:10.1371/journal.pone.0032212.g005 H67 (LRF) mutation was introduced, considering that the residue in this position was unusual both in the donor and in the acceptor FWRs, being therefore substituted by a well conserved residue in human antibodies. The resulting alignment of the Fv of hum-aD11 to the donor (parental aD11) and acceptor (PDB_ID: 1JPS) [38] sequences is presented in Figure 4.
The structural model of the designed hum-aD11 was then refined by energy minimization using the program CNS [41]. In Figure 5B, an overlay of the F v region of the parental aD11 and the model of the F v region of the hum-aD11 is shown. The resulting model of the Fab fragment of hum-aD11 and the modeled F v region being assembled in a composite immunoglobulin are shown in Figure 5C.

Hum-aD11 shows an enhanced in vitro NGF binding affinity
In order to obtain the hum-aD11 as an antibody protein in the IgG1 format, for its functional evaluation and characterization, DNA sequences encoding hum-aD11 V H and V k were synthesized using overlapping oligonucleotides (Table S1), genetically fused to the human c1 heavy and the k light constant regions, respectively (to reconstitute a human IgG1 antibody) and cloned in suitable eukaryotic expression vectors [42] that were used to co-transfect Chinese Hamster Ovary (CHO) mammalian cells.
ELISA assay was performed on CHO transfectant supernatants, to evaluate the binding of IgG1 hum-aD11 binding to mNGF and compare it to the parental mAb aD11, expressed in the chimeric immunoglobulin IgG1 format, IgG1 chim-aD11 (rat aD11 variable regions fused to the c1 heavy and the k light constant regions, respectively) [22]. The IgG1 chim-aD11 and the rat mAb aD11 were previously shown to have overlapping NGF binding curves [22]. The chim-aD11 and hum-aD11 in the human IgG1 format were transiently expressed in CHO cells, purified by Protein A-Sepharose and quantified by immunoblot. After normalization, serial dilutions were tested by ELISA ( Figure 6A). The results show that IgG1 hum-aD11 binds mNGF equally well as the IgG1 chim-aD11 (and the parental rat mAb aD11). Quite surprisingly, not only the NGF binding of the hum-aD11 IgG1 was not reduced, but on the contrary the binding affinity appeared to increase, in comparison to the chimeric IgG1 aD11 and to the parental rat mAb aD11 ( Figure 6A). This might be the result of a more favorable interaction at the variable/constant domains interface in the two antibodies. Thus, the single-cycle humanization procedure was sufficient to fully reconstitute the NGF binding strength of the parental antibody, even showing some binding improvement.
To confirm the likely affinity improvement of the IgG1 hum-aD11 by kinetic and quantitative affinity measurements, Surface Plasmon Resonance (SPR) studies were performed, to compare the NGF binding kinetics with respect to the parental and humanized antibodies, by exploiting two different configurations aimed to avoid any avidity effect due to the use of the IgG1 format as an analyte. Details of the parental and humanized aD11 antibodies binding curves to hNGF are reported in Figure 6B and Figure 6C, respectively. The humanized version of the antibody, indeed showed a significantly higher affinity for hNGF with respect to the parental version, K D values of 28.964.1 pM and 451645 pM, respectively ( Table 1).
Hum-aD11 fully preserves the in vitro NGF antagonistic activity and the in vivo analgesic properties of the parental mAb aD11 To verify that the IgG1 hum-aD11 maintained the ability of the parental one to inhibit NGF biological activity in vitro, two cellular models were employed.
First, in an NGF-induced neurite outgrowth bioassay on rat PC-12 cells [43] ( Figure 7A to Figure 7D) both the parental mAb and the IgG1 hum-aD11 exerted identical effects. Indeed, mNGF treated PC-12 cells ( Figure 7A), preincubated with either the parental mAb or the IgG1 hum-aD11, failed to show any neurite outgrowth ( Figure 7B and Figure 7C), as in the absence of mNGF ( Figure 7D). In control experiments NGF-induced differentiation occurs normally ( Figure 7A), even if mNGF was preincubated with a non relevant mAb or with the concentrated untransfected CHO cell supernatants (data not shown). The nuclear morphology of IgG1 hum-aD11 treated PC-12, stained with 49,6-diamidino-2-phenylindole (DAPI), under the different conditions of the assays, was normal (data not shown), underlining that the failure of cells treated with mNGF and IgG1 hum-aD11 to differentiate was not due to a non-specific toxic effect, but indeed to the neutralization of NGF binding and of the ensuing differentiation.
The ability of the parental mAb aD11 and the IgG1 hum-aD11 to inhibit the activation of human TrkA receptor by mNGF was compared in 3T3 TrkA cells (ectopically expressing human TrkA receptor, but not expressing the neurotrophin p75 NTR receptor). As shown by Western-blot analysis of TrkA phosphorylation at residue Y490 ( Figure 7E), no activation of human TrkA receptor could be detected in the cells treated with mNGF preincubated either with the parental mAb aD11 or the IgG1 hum-aD11, as well as in the absence of mNGF. In the control experiments, mNGF-induced human TrkA activation occurred normally, also in the presence of a non relevant antibody.
At the functional level, the potency of the mAb hum-aD11 to neutralize the activity of NGF from different species was also ascertained by the TF-1 cell-based proliferation bioassay [33]. The mAb hum-aD11 inhibits TF-1 cell proliferation with a similar potency (IC 50 value of ca. 30 ng/ml) in the presence of human, rat and mouse NGF, respectively (data not shown).
Finally the analgesic activity of the hum-aD11 antibody was tested in the formalin-induced inflammatory pain model in vivo.
Since previous experiments with the parental mAb aD11 showed that analgesic effect was greater for mAb aD11 in the Fab than in the whole IgG format ( Figure 2B), these experiments were performed with the Fab fragment of hum-aD11, expressed in the periplasm of bacterial cells. As shown in Figure 8, Fab hum-aD11, was able to reduce formalin-evoked pain both in the early and in the late phase (p,0.05) of the formalin test, with a stronger effect in the latter phase (which is related the inflammatory component of pain). The Fab hum-aD11 determined an identical analgesic response (halving of the pain response), at the same doses, as the parental rat-aD11 Fab fragment, and therefore it clearly retains the analgesic properties of the parental antibody.

Discussion
In this paper we describe the humanization of rat mAb aD11, which potently antagonizes the activity of NGF, a target of great clinical relevance for various pathological situations, including presently untreatable forms of chronic and inflammatory pain [8], [15]. The humanized form of aD11 recapitulates the remarkable affinity (which is even improved) and neutralizing potency properties of the parental mAb aD11, including its analgesic properties in rodent models of inflammatory pain. The results obtained unequivocally prospect hum-aD11 as a lead therapeutic candidate for human pathologies where antagonizing systemic or local NGF activity in the periphery would be of clinical benefit.
The structure-based humanization strategy of mAb aD11 was performed by a novel method, which overcomes the well known  limitations of the classical CDR grafting methods [44], [45], [46], whereby the CDRs of the parental rodent mAb are grafted onto human acceptor FWRs. The standard humanization procedure requires a laborious and time-consuming iterative procedure, to derive a humanized antibody with affinity and binding properties comparable to those of the parental one [47], [48], [49], [50]. Thus, the choice of the acceptor FWRs represents the critical point in the procedure of antibody humanization by CDRs grafting, leading most often to a significant binding affinity loss of the humanized antibody [51], [52], [53], due to distortions of the CDRs conformations by the human acceptor FWRs. This requires a trial and error iterative procedure, to correct structurally distorting residues and to reconstitute the original binding properties of the parental antibody. A universal combinatorial library of antibody FWRs suitable for humanizing exogenous antibodies by CDR-grafting, based on a bioinformatics approach, has been recently proposed [54] to address these issues. Even if computer modeling simulations and predictions can be employed to improve the outcome of humanization [44], [46], [55], the fidelity of antibody models to the experimental structures is rather low, especially concerning the CDR H3 loops, which are more variable in sequence, length and structure [56], [57], [58], [59] and more flexible [58], [60], [61] than the other CDRs. Moreover, the conformations of CDR H3 strongly depend on the neighbouring structural environment [58], and in a significant number of F v crystal structures they are not even structurally defined. Therefore, reliable modelling of the CDR H3 is still a real challenge [62]. Moreover the overall conformation of antigen binding site of antibodies further depends in a complex and unpredictable manner on movements at the V k /V H interface [63]. Thus, antibody humanization by CDRs grafting, as such, can, and has proved to be, an unpredictably daunting and laborious task. On the other hand, the alternative method of F v humanization by resurfacing, in which only solvent accessible residues of the nonhuman donor antibody are considered for substitution by homologous residues belonging to human FWRs regions [64], [65], [66], results in the presence of a higher number of nonhuman residues that might represent cryptic epitopes contributing to an immunogenic response in patients.
To overcome these drawbacks, we developed a new structurebased humanization strategy, to improve over traditional CDR grafting method. In the approach described here, we exploited the high resolution crystallographic structure of the Fab aD11 (PDB_ID: 1ZAN) [26], [27] in order to optimize the key step in the selection and choice of the human FWRs acceptors. This structure-based methodology allowed us to readily humanize the aD11 antibody in a single cycle, obtaining an engineered version whose binding and biological activity both in vitro and in vivo closely recapitulate those of the parental version. It is worth noting that the NGF binding affinity of the humanized hum-aD11 is not only maintained, but surprisingly and unpredictably improved, by an order of magnitude, over that of the parental antibody. The molecular underpinnings for such an unexpected affinity improvement will require a further structural investigation of the humanized antibody itself. With many antibodies of therapeutic  interest being derived from murine monoclonal antibodies, the need for their humanization for use in patients is mandatory. The single-cycle structure-based method is a considerable improvement over the standard humanization methods, which are intrinsically empirical and require several trial and error refinement cycles. The method described here may significantly accelerate the clinical development path of many monoclonal antibodies of therapeutical interest.
The second major conclusion of this study establishes the anti-NGF hum-aD11 as an effective lead candidate for clinical applications in a therapeutic area that represents a severe unmet medical need. Pain is the most common symptom for which patients seek medical assistance, and most forms of chronic and inflammatory pain are inadequately treated by present therapeutic options. Compelling evidence has accumulated, demonstrating that the NGF-TrkA system is a key modulator of inflammatory and nociceptive responses and a master control system for pain, functionally placed upstream in the hierarchy of the pain regulation process [6], [67]. There is, therefore, a strong rationale for the development of a new generation of analgesic therapeutic interventions, based on antagonising the NGF pathway via inhibition of its interactions to receptors [8], [15]. In this respect the large and rather flat ligand-receptor binding interfaces disclosed by the crystal structures of the complexes between NGF and its receptors p75 NTR and TrkA [29], [68], [69], make the rational design of high affinity and specific small-molecule antagonists of these interacting surfaces a daunting prospect. Moreover, small molecule NGF-TrkA antagonists are more likely than larger antibody molecules to cross the blood-brain barrier, leading to unwanted neurological side effects. For these reasons the use of function-blocking antibodies targeting the NGF ligand [27], [70] or its TrkA receptor [71], [72], [73] represent one obvious strategy to develop new pain therapeutics [8], [15].
The first antibody in this group reaching clinical evaluation in humans is tanezumab, a humanized anti-NGF monoclonal antibody [70], [74]. The therapeutic concept of blocking the activity of NGF for chronic and inflammatory pain received a very strong validation from the results of proof-of concept clinical trials showing that tanezumab can very effectively and persistently relieve joint pain and improve functions in moderate-to-severe osteoarthritis patients [75]. However, in a subsequent Phase 3 clinical study of tanezumab for osteoarthritis of the hip and knee, 16 treated subjects showed joint failure and required total joint replacement [16], [75]. This led the U.S Food and Drug Administration to put on hold the clinical programs for tanezumab, until more information on the incidence and causes of these adverse events is gained (http://clinicaltrials.gov/ct2/ results?term = ngf+antibody). While these events could be ascribed to causes to be determined, related to the NGF/TrkA system in vivo, antibody-specific causes cannot be excluded and should be considered. In this respect, comparative evaluations of different anti-NGF antibodies will be very informative.
Although several mAbs were raised against NGF (reviewed in [74]), the mAb aD11 deserves a special interest, not only because of its picomolar affinity [19], antagonistic potency and specificity, but also because its NGF neutralizing properties have been very extensively verified in a diverse set of in vivo situations [20], [21], [22], [23], [24], [25] and because it is the only anti-NGF monoclonal antibody for which a 3D structure has been derived [26], [27]. We now further show that the anti-NGF mAb aD11 displays very significant NGF antagonistic properties in relevant murine pain models, with a remarkable and surprising long-lasting analgesic property that may be of high clinical relevance. Although the mechanistic dissection of this long-term analgesic effect remain to be further characterized, it appears to be directly related to NGF-TrkA signaling in adult nociceptor neurons, since a similar long-lasting persistent analgesic effect was observed after administration of the neutralizing anti-TrkA MNAC13 antibody [73]. Thus, interrupting the activation of NGF-TrkA signaling in chronic and inflammatory pain states might induce a therapeutically beneficial feedback loop, prospecting anti-NGF ad anti-TrkA antibodies as analgesics with disease-modifying properties.
These results establish hum-aD11, the humanized counterpart of mAb aD11, as a lead candidate with a strong therapeutic potential not only in the pain arena, but for all those pathological states where an excess of NGF expression and/or activity is detrimental.

Materials and Methods
Construction and cloning of the cDNAs of the variable regions of hum-aD11 cDNAs of hum-aD11 V H and V k were obtained by gene synthesis using overlapping oligonucleotides (Table S1) according to Kolbinger et al. [76]. After overlap assembly PCR, fragments of the correct size were purified from agarose gel and directionally cloned in expression vectors for IgG1 expression [42], respectively BssHII/BstEII in V H Express vector for hum-aD11 V H and ApaLI/BglII in V k Express vector for hum-aD11V k .
In order to produce Fab hum-aD11 in the bacterial periplasm, the following two expression cassettes were cloned to express hum-aD11 V H and hum-aD11 V k in fusion with human constant regions of the heavy (hC H ) and the light (hC k ) chains. In details, hum-aD11 V k was directionally cloned (NcoI/HindIII) in frame with hC k (HindIII/NotI) in pET22b (Novagen) to obtain hum-aD11 light chain in pET22b. After inserting the PelB secretion sequence (NdeI/NcoI) in pET28 (Novagen), hum-aD11 V H was directionally cloned (NcoI/NheI) in frame with hC H (NheI/XhoI) and with C-terminal his-tag to obtain hum-aD11 heavy chain in pET28_ PelB.
At each cloning step, positive clones, isolated by PCR screening directly on bacterial colonies, were confirmed by DNA sequencing.

Expression of hum-aD11 in the IgG1 and Fab formats
The CHO cells (The European Collection of Cell Cultures, Sigma-Aldrich, Product nu 85050302), grown in 90mm dishes, were transfected with 3 mg IgG1 hum-aD11 expression plasmids (1.5 mg of hum-aD11 V H in V H Express and 1.5 mg of hum-aD11 V k in V k Express) and with 3 mg chimeric aD11 expression plasmids (1.5 mg of rat aD11V H in V H Express and 1.5 mg of rat aD11 V k in V k Express) [22], using FuGENE (Roche) according to the manufacturer's protocols. Conditioned medium was collected 72 hrs after transfection.
To select double transfectant clones expressing IgG1 hum-aD11, CHO cells were cotransfected as reported above, using the same amount of EcoRI linearized plasmids. 36 hrs after transfection, micophenolic acid (50 mg/ml) and G418 (500 mg/ml) were added. Conditioned media were concentrated to 0.2 ml final volume using Microcon concentrators (Millipore) or purified using Protein A-Sepharose (GE Healthcare).
Protein concentrations were estimated by immunoblot. In details 5 ml of each purified sample or concentrated conditioned medium were spotted on a nitrocellulose membrane. After blocking with PBS (5% non fat dry milk), the membrane was incubated at first with an anti-human polyclonal antibody (Pierce) diluted 1:500 in PBS (5% non fat dry milk), acting as primary Ab, followed by the anti-goat Ab, coupled to horseradish peroxidase (Dako), diluted 1:1000. The Horseradish Peroxidase (HRP) conjugates were detected with the electrochemiluminescence protocol developed by Amersham Corp.
The amounts of human IgG in the different samples were normalized after densitometric scanning and standards of purified human IgG (Sigma) were used to determine absolute protein levels; as negative control for concentrated conditioned media, concentrated CHO supernatant of untransfected cells was used.
In order to express hum-aD11 Fab, hum-aD11 heavy chain in pET28_PelB and hum-aD11 light chain in pET22b were cotransformed in BL21(DE3) E.coli cells. Cells were grown at 30uC in M9CA broth supplied with kanamycin (50 mg/ml) and ampicillin (100 mg/ml), induced at an OD 600 nm of 0.9 by adding 0.5 mM IPTG and left to grow for 15 hrs. The cell pellet resuspended in buffer A (50 mM Na phosphate pH 8, 0.5 M NaCl, 5 mM MgCl 2, 10 mM imidazole, 5% glycerol) was incubated with 1 mM PMSF, 100 mg/ml lysozyme and 5 mg/ml Dnase I, for 30 min and then sonicated on ice. Soluble extract was loaded on Ni-NTA Superflow resin (Qiagen) equilibrated with buffer A and Fab hum-aD11 was subsequently eluted with 250 mM imidazole. The collected fractions were then pooled and loaded on a HiLoad 16/60 Superdex 75 (GE Healthcare) preequilibrated with 20 mM Tris-HCl pH 8.5, 0.2 M NaCl, 5 mM MgCl 2, 5% glycerol. The fractions corresponding to the Fab hum-aD11 were concentrated using an Amicon Ultra-15 centrifugal filter unit (Millipore) with a membrane cut-off of 10 kDa.

ELISA assays
ELISA assays were performed according to Covaceuszach et al. [72] with the following modifications.
In order to compare mAb aD11 binding towards human and mouse NGF, mNGF (Alomone) and hNGF obtained according to Covaceuszach et al. [77], were coated; followed by first serial dilutions of mAb aD11 (1:2 dilutions in the concentration range between 20 mg/ml and 18 ng/ml), acting as primary antibody, and finally by the anti-rat antibody peroxidase conjugated (Dako) acting as secondary antibody.
ELISA assay to compare chimeric and IgG1 hum-aD11 binding towards mNGF (Alomone) was performed with the following modifications: serial dilutions of chimeric and IgG1 hum-aD11 were incubated and the anti-human polyclonal antibody (Pierce) was used as primary antibody and the anti-goat antibody peroxidase conjugated as secondary antibody (Dako). All the experiments were done in duplicate.

Surface Plasmon Resonance
SPR measurements were performed with a BIACore instrument (BIACore AB, Uppsala, Sweden) in triplicate.
Parental rat mAb aD11 was immobilized on three CM5 sensor chips by cross-linking the amine groups according to the manufacturer's instructions, obtaining SPR signals, after completion of the chip regeneration cycles, respectively of 1025, 1040 and 1280 resonance units (RU). The binding kinetics for rat mAb aD11 were determined by injection on each surface of serial dilutions (in the 0.29 nM to 37.5 nM concentration range) of hNGF in PBS buffer with addition of 0.005% v/v Surfactant P20 at a flow rate of 30 ml/min.
The binding kinetics for Fab hum-aD11 were determined by immobilizing hNGF on three CM5 sensor chips by cross-linking the amine groups according to the manufacturer's instructions, obtaining SPR signals, after completion of the chip regeneration cycle, respectively of 89, 93 and 100 resonance units. Serial dilutions (in the 0.39 nM to 25 nM concentration range) of Fab hum-aD11 (kindly provided by PanGenetics UK Ltd.) in HBSEP (10 mM HEPES, 150 mM NaCl, 3 mM EDTA and 0.005% Surfactant P20, pH 7.4) containing 100 mg/ml bovine serum albumin, were injected on each surface at a flow rate of 50 ml/min. Data were analyzed using the BIAevaluation 3.0 package (GE Healthcare) to yield the apparent equilibrium constant K D (defined as the k a /k d ratio) and K A (defined as the k d /k a ratio).

NGF bioassay with PC-12 cells
Rat adrenal gland phaeochromocytoma PC-12 cells (Sigma-Aldrich, Product nu 88022401) [43] were maintained in RPMI 1640 medium (Life Technologies, Milano, Italy), supplemented with 10% fetal calf serum. PC-12 cells were primed with 50 ng/ml mNGF (Alomone) on collagen-coated (type I, BD Biosciences) 35 mm Petri dishes at a density of 0,25610 5 cells per dish for the second day after seeding onward. Priming was carried out for 5-6 days, with mNGF added every 3 days, then cells were detached and plated on collagen-coated 35 mm Petri dishes. For differentiation assays, cells were incubated with 100 ng/ml mNGF for 2-4 days in the presence or absence of parental mAb or IgG1 hum-aD11 (5 mg/ml; 1 hour preincubation).

TrkA Phosphorylation-Inhibition Assay
TrkA Phosphorylation-Inhibition Assay was performed according to Ugolini et al. [73], with the following modification. Prior treatment, 100 ng/ml mNGF was preincubated in the presence or absence of parental mAb aD11, IgG1 hum-aD11 or irrelevant mAb SV5 (2.5 mg/ml) in serum-free medium supplemented with 0.05% BSA for 1 hour.
The human premyeloid cell line TF-1 was purchased from the American Type Culture Collection (ATCC, Product nu CRL-2003).

Inflammatory and neuropathic pain models
Formalin test and neuropathic pain tests were performed according to Ugolini et al. [73].

Ethics Statement
All animal work has been approved by Italian Ministry of Health (order N.34/2008-B) and have been conducted according to the Italian National law (DL116/92, Application of the European Communities Council Directive 86/609/EEC) on care and handling of the animals and with the guidelines of the Committee for Research and Ethical Issues of International Association for the Study of Pain [78]. Figure S1 Primary and tertiary structural comparison. 3D structural and sequence comparisons between the crystal structure of the Fab rat aD11 and the crystal structures of human or humanized antibodies, Fabs, IgGs or of their complexes with antigens indentified in the PDB database (release #101, July 2002). The plotted variables are: The skeleton C a r.m.s.d , the % of C a atoms considered in the r.m.s.d calculations and A) % of sequence homology on both the variable domains B) % of sequence identity on both the variable domains C) % of sequence homology on the FWRs D) % of sequence identity on the FWRs. (TIF)

Supporting Information
Table S1 Oligonucleotides sequences used in the synthesis of the CDRs grafted hum-aD11 V k (A) and V H (B) regions by overlap-assembly PCR. (DOC)