The Development of Rapastinel (Formerly GLYX-13); A Rapid Acting and Long Lasting Antidepressant.

BACKGROUND: Rapastinel (GLYX-13) is a NMDA receptor modulator with glycine-site partial agonist properties. It is a robust cognitive enhancer and shows rapid and long-lasting antidepressant properties in both animal models and in humans.
METHODS: Rapastinel was derived from a monoclonal antibody, B6B21, is a tetrapeptide (threonine-proline-proline-threonine-amide) obtained from amino acid sequence information obtained from sequencing one of the hypervariable regions of the light chain of B6B21. The in-vivo and in-vitro pharmacology of rapastinel was examined.
RESULTS: Rapastinel was found to be a robust cognitive enhancer in a variety of learning and memory paradigms and shows marked antidepressant-like properties in multiple models including the forced swim (Porsolt), learned helplessness and chronic unpredictable stress. Rapastinel's rapid-acting antidepressant properties appear to be mediated by its ability to activate NMDA receptors leading to enhancement in synaptic plasticity processes associated with learning and memory. This is further substantiated by the increase in mature dendritic spines found 24 hrs after rapastinel treatment in both the rat dentate gyrus and layer five of the medial prefrontal cortex. Moreover, ex vivo LTP studies showed that the effects of rapastinel persisted at least two weeks post-dosing.
CONCLUSION: These data suggest that rapastinel has significant effects on metaplasticity processes that may help explain the long lasting antidepressant effects of rapastinel seen in the human clinical trial results.

The discovery of B6B21

Rapastinel (formerly GLYX-13) is an amidated tetrapeptide, (threonine-proline-proline-threonine-amide) derived from cloning and sequencing the hypervariable regions of the heavy and light chains of a monoclonal antibody, B6B21. Fig. shows part of the amino acid sequence of one of the hypervariable regions of the light chain of B6B21 that contains the sequence TPPT. Based on NMR analysis, Fig. also shows that this molecule exists in a fairly rigid β-1 type turn, which is a common structural feature of many protein-protein interactions. Nuclear Overhauser effect analysis showed that rapastinel actually exists as a 3-ringed structure stabilized by hydrogen bonds. Thus the 3-dimensional structure of GLYX-13 is quite stable and complex for so simple a seeming small peptide [1, 2].

The monoclonal antibody (mAB), B6B21, was created in an attempt to generate a panel of monoclonal antibodies that could be used as tools to further dissect the molecular mechanisms that underlie synaptic plasticity associated with learning and memory. It was hypothesized that a judicious choice of immunogen and screening protocol could lead to the identification of a monoclonal AB that would facilitate learning and because of the exquisite specificity typically found for mABs, be useful to identify the molecular substrates underlying its learning and memory enhancing properties.

The immunogen chosen for these studies was dentate gyri from five-day old postnatal rats. Fresh, unfixed dentate gyri were immediately removed by the micropunch method of Palkovits and Brownstein [3], rapidly homogenized in PBS at 4°C, and injected intraperitoneally into Balb/c mice. An average of 50 mg wet weight of tissue was injected per mouse. This procedure was repeated 4 times over 2 months. Monoclonal antibodies were generated by conventional techniques using NS-1 as the parent myeloma cell line 
[4, 5].

The screening protocol chosen involved a sequence of steps based on the idea that the ideal mAB would recognize cell surface antigens expressed on living hippocampal neurons and that could be used to study synaptic plasticity in vitro before attempting to examine their learning and memory facilitation properties in vivo. Given that function (i.e., learning and memory enhancement in vivo) was the ultimate goal and not simply identification of novel ligands that modulate already known receptors, the key hurdle was to minimize the number of screens necessary without losing a key functional mAB along the way. In short, it was typical to obtain several hundred mABs from each immunization protocol but it was impossible to go directly to assaying each of them in an in vivo animal model of learning. The following is a brief summary of the screening protocol.

First, the mABs obtained from the dentate gyri immunogen were cloned at limiting dilution twice to insure monoclonality of each hybridoma obtained. Then individual hybridomas were grown in large enough quantities to be frozen for future use. Hybridomas growing in tissue culture were then combined into groups of 10; what was dubbed poly-hybridomas. This was a useful step because it greatly reduced the first screening step which was to evaluate histochemically the binding patterns of what would now be 10-20 poly-mABs to adult unfixed frozen sections of adult hippocampi. This reduced screening from several hundred immunohistochemical screens to approximately 15 followed by an additional 10 since each of the combined “poly-mABs” had been frozen as individual mABS (i.e., the hybridomas that secrete an individual mAB) and could readily be thawed and grown up for individual sub-screening.

From this step, mABs were chosen for binding to cell surfaces of living hippocampal neurons in primary cultures [6]. This was the key next step for it would insure that mABs that successfully passed this screen would recognize both hippocampal antigens expressed on living neurons. Historically, mAB technology had been used as an immunocytochemical tool to identify staining patterns. Two excellent examples of this being McKay and Zipser [7] who identified distinct neuronal cell surface binding patterns in the leech nervous system and Trisler et al., [8] who identified a cell surface antigen expressed in a 40-fold gradient across the chick retina [9].

The next step was to screen these mABs in rat hippocampal slice preparations and look for the enhancement of long-term potentiation (LTP). This was chosen as the next critical screen among many other possibilities because it made it possible to get synaptic plasticity data from a fully functional model system. The hippocampal slice and LTP had emerged as an excellent model of learning and memory. Moreover, the daunting task of having to screen a large number of mABs was still present. To this end, a strategy had to be developed in which a single hippocampal slice could be used as both the experimental and control tissue. These studies led to the identification of a unique LTP inhibitory mAB [10] and later to the identification of B6B21 the mAB that was used for in vivo studies.

At this point, because of the robust enhancement of LTP found and the role of NMDA receptors, pharmacological studies were undertaken to directly assess B6B21 for possible NMDA receptor modulatory properties. Subsequently, B6B21, was found to be a glycine-site partial agonist at the NMDA receptor [11].

B6B21 was then evaluated in vivo using trace eyeblink conditioning, an NMDAR- and hippocampus dependent test of associative learning and memory. Given that mABs are too large to cross the blood brain barrier (BBB) but the third ventricle literally bathes the hippocampal formation, a hippocampus-dependent trace eyeblink paradigm was ideally suited since this was a true associative learning paradigm both well characterized in animal models as well as in humans [12]. Remarkably it was reported in 1991 [11] that B6B21 did significantly enhance trace eyeblink conditioning, approximately 8 years after the first mABs were generated [13]. And it was these results that led to the idea that B6B21 could be used as a template to create small molecules with therapeutic potential which led to the creation of rapastinel [14].

The development of rapastinel

The amino acids that comprise the hypervariable regions or CDRs of an antibody molecule determine its binding specificity. Synthetic peptides, derived from these amino acid sequences have been demonstrated to possess biological activity similar to that of the intact antibody [15-20]. To design B6B21 antibody mimetics, the hypervariable regions of the light chain of B6B21 were cloned using reverse transcriptase-polymerase chain reaction (RT-PCR) technology [21]. Upon cloning of both the heavy and light chains of the monoclonal antibody, B6B21, only one sequence, QQHYSTPPT (glutamine, glutamine, histidine, tyrosine, serine, threonine, proline, proline, threonine), found in the light chain, (see Fig. 1 for the complete light chain sequence) showed NMDA receptor binding activity. From this sequence a panel of peptides was synthesized and assessed as previously described [22]. Of these, GLYX-13, the TPPT-amide, was found to have the most robust binding activity [22]. GLYX-13 stimulated [3H]MK-801 binding to approximately 130% of control at 1 µM (Fig. ). TPPT-amide was later named GLYX-13 as it was the thirteenth peptide in the series showing the most activity in the MK-801 assay described above. Recently GLYX-13 has been renamed as rapastinel.

These results suggest that rapastinel was acting as a glycine-site partial agonist since it overrode the effects of a competitive glycine site antagonist, 7-chloro-kyneurinic acid and gave dose response curves similar to another partial agonist, DCS. The next study chosen was to establish if rapastinel crossed the blood brain barrier. BBB permeability was determined using methods previously described [23]. Using a brain uptake index of 100% for water, it can be seen that GLYX-13 is nearly as permeable as water with an index of approximately 80% (Fig. ). Interestingly, GLYX-13 crosses the BBB more efficiently than both heroin and codeine under similar conditions [14, 24].

From here a variety of behavioral pharmacology studies were undertaken (Fig. ). In the first series, rapastinel was tested in a variety of hippocampus learning and memory paradigms using young rats and cognitively impaired aging rats. The paradigms used were trace eyeblink conditioning, alternating T-maze, Morris water maze and positive emotional learning [14, 25, 26]. Robust cognitive enhancement in each paradigm was observed but noteworthy was the fact that rapastinel was able to reverse the cognitive deficits found in the learning impaired aging animals in each paradigm as well. The positive emotional learning paradigm was included in part because it has been established to be a medial prefrontal cortex-dependent process.

In a second series of studies, rapastinel was assessed for its antidepressant properties. As shown in Figs. -, rapastinel displayed marked antidepressant-like properties in a variety of models including forced swim, learned helplessness, novelty-induced hypophagia. Further studies were performed using a mild chronic unpredictable stress model (CUS) to induce a depressive-like state. This model has been well characterized and used to study many of the neurobiological underpinnings of depression [28]. Rapastinel was found to rapidly and markedly reverse the prodepressive effects of CUS in the same models above as well as the sucrose preference test. Likewise (Fig. ) the cognitive defects in positive emotional learning, induced by CUS, were also reversed by rapastinel.

Recently, based in part on the fact that rapastinel showed significant cognitive enhancing properties as well as antidepressant properties, studies were undertaken to evaluate it as a possible therapeutic candidate for post-traumatic stress disorder. The same CUS treatment used above was also used as part of a contextual fear extinction paradigm. As shown in Fig. , rapastinel facilitated contextual fear extinction and prevented fear extinction and re-consolidation in rats exposed to CUS.

And finally a series of studies were performed to evaluate the long-term antidepressant and cognitive enhancing effects of rapastinel. A single dose of rapastinel produced marked antidepressant-like and cognitive enhancing effects one week post-dosing (Fig. ). Rapastinel exerted its cognitive enhancing antidepressant-like, CUS reversal, and CFE effects at the same dose. CPP, an NMDA receptor antagonist, blocked the antidepressant-like effects of rapastinel demonstrating that receptor activation is required for its effects. Rapastinel, 24 hrs post-dosing was also found to significantly enhance the formation of LTP in both hippocampal and MPFC slices in vitro and led to a significant increase in mature dendritic spine formation in the dentate gyrus and layer 5 of the MPFC in rats (Fig. ). Thus it appears that rapastinel exerts its cognitive-enhancing and antidepressant properties by activating NMDA receptors leading to a form of NMDA receptor dependent plasticity akin to LTP.

Mechanism of action

Rapastinel was created from the amino acid sequence information derived from a hypervariable region of a mAB that was shown to act as a NMDA receptor glycine site partial agonist, suggesting that rapastinel would act in a similar fashion. As shown in Fig. , several lines of evidence supported this in fact was the case: (1) rapastinel was able to compete with the glycine site competitive antagonist, 7-CK, in a dose dependent fashion to activate NMDA receptors in membrane preparations derived from rat hippocampal synaptosomes; (2) Electrophysiological studies using murine NMDA receptors expressed in oocytes showed rapastinel 
to have glycine site partial agonist properties; (3) electrophysiology studies using rat hippocampus slices showed rapastinel to be a robust NMDA receptor-dependent LTP-enhancer; (4) partial agonist studies using the Stephenson method [31] using either NMDA receptor-specific current readout or D-serine-induced [3H]MK-801 binding as the readout also showed that rapastinel displayed

partial agonist properties. Taken together these data make a compelling case for rapastinel as a glycine site partial agonist. However each of the assays employed measured a functional readout and as such rapastinel has been described as a functional glycine site partial agonist since no direct ligand binding studies or competition studies with radiolabeled glycine or an analog have been reported. Thus, at face value it could be argued that rapastinel works by another mechanism which could include, among others, binding to an allosteric site, interfering with NMDA receptor protein-protein interactions necessary for optimal function such as those of the post-synaptic density complex, or binding to a NMDA receptor modulating protein directly. And one cannot rule out that an effective direct ligand binding assay simply has yet to be developed. In lieu of such an assay, studies are in progress using in silico methods of recently published NMDA receptor X-ray crystallographic data [33] to guide a program of creating point mutations aimed directly at inhibiting rapastinel’s ability to effect NMDA receptor function when these receptors are expressed in model systems such as human embryonic kidney (HEK) cells.

Summary and Conclusions

Rapastinel has been shown in two Phase II clinical trials to have significant rapid acting and long lasting antidepressant effects [34, 35]. Rapastinel was created using a monoclonal antibody that enhanced learning and acted as a NMDA receptor glycine site partial agonist as a template by cloning and sequencing its heavy and light chains, creating peptides from this information and screening them for glycine site modulation properties using an in vitro membrane assay. The therapeutic potential of rapastinel was assessed by establishing that it readily crossed the BBB, showed no obvious side effects, was a robust cognitive enhancer and displayed rapid and long lasting antidepressant-like effects in a variety of models. Mechanism studies have shown that rapastinel modulates the glycine site of the NMDA receptor but are based on functional and thus indirect readouts. As such it remains unclear as to the exact mechanism of rapastinel’s action although it does seem clear that it happens via glycine site modulation. Finally, it is proposed that rapastinel exerts its cognitive-enhancing and antidepressant effects via an NMDA receptor triggered LTP-like synaptic plasticity mechanism associated with learning and memory (Fig. 10).