Inventing Liraglutide, a Glucagon-Like Peptide-1 Analogue, for the Treatment of Diabetes and Obesity

Abstract

Glucagon-like peptide-1 (GLP-1) has been in focus since the early 1980s as a long looked for incretin hormone, released from the gastrointestinal tract and with an important effect on glucose-dependent insulin secretion, providing efficient glucose lowering, with little risk for hypoglycemia. The enzyme dipeptidyl peptidase-4 (DPP-4) degrades GLP-1 very fast, and the remaining metabolite is cleared rapidly by the kidneys. Liraglutide is a fatty acid acylated analogue of GLP-1 that provides efficacy for 24 h/day. The mechanism of action for liraglutide is reviewed in detail with focus on pancreatic efficacy and safety, thyroid safety, and weight loss mechanism. Evolving science hypothesizes that GLP-1 has important effects on atherosclerosis, relevant for the cardiovascular benefit seen in the treatment of diabetes and obesity. Also, GLP-1 may be relevant in neurodegenerative diseases.

Introduction

Figure 1.

Lotte Bjerre Knudsen.

Discovery of GLP-1 and the Early Evidence for Its Relevance as a Diabetes Drug

The history of gut hormones tracks back to 1902 where Bayliss and Starling¹ discovered a “chemical substance which is formed in the mucous membrane of the upper parts of the small intestine under the influence of acid, and is carried thence by the blood-stream to the gland-cells of the pancreas”. The term incretin was introduced later, but it was not until the late 1960s when the introduction of more sophisticated technology led to the discovery of the first of the two main incretins we know today, gastric inhibitory polypeptide (GIP).² In the early 1980s, it was speculated that there was another incretin, as intestinal extracts still had insulinotropic activity after GIP was removed.³ The early incretin history is excellently described in a comprehensive review published in 1999 by Tim Kieffer and Joel Habener.⁴ The sequence of vertebrate GLP-1 was first discovered in 1983 by Graham Bell in Chicago who cloned hamster preproglucagon and found it to harbor potentially two more hormones that were named GLP-1 and GLP-2.^5,6 Initially, it was the extended form of GLP-1, GLP-1(1–37) that was proposed as the active hormone, but shortly thereafter, it was discovered that the GLP-1(1–37) was processed to GLP-1(7–37) and GLP-1(7–36)amide, which are the predominant active insulinotropic hormone forms. This discovery was made simultaneously by two groups, those of Joel Habener in Boston and Jens Juul Holst in Copenhagen.^7,8 One can say that we were lucky the GLP-1 field did not start out as a drug idea based on mouse genetics, because likely it would not have been considered important enough, as the GLP-1R^–/– mice have a limited phenotype when it comes to glucose metabolism and body weight homeostasis.⁹⁻¹¹ GLP-1 is certainly physiologically relevant, is involved in both glucose and body weight homeostasis, and affects the cardiovascular system in several ways, but it is not uniquely important when absent. We are thus lucky that brave people went from in vitro experiments and straight into clinical studies. In 1987, Kreymann et al. showed GLP-1 was an incretin in man.¹² In 1992, Gutniak et al. demonstrated a clear antidiabetogenic potential of GLP-1 not only in patients with type 2 diabetes but also in patients with type 1 diabetes.¹³ From 1993 and onward, Michael Nauck and colleagues in Germany and Jens Juul Holst and Sten Madsbad in Denmark published many studies documenting numerous aspects of the clinical potential of GLP-1. Importantly, these studies documented that GLP-1 did not induce hypoglycemia, as its effect on insulin secretion was glucose-dependent. GLP-1’s incretin effect is preserved in patients with type 2 diabetes, unlike GIP that is the most important incretin in people without diabetes, whereas in type 2 diabetes the effect of GIP is greatly diminished.¹⁴ Type 2 diabetes is characterized by a marked loss of incretin effect,¹⁵ and GLP-1, when dosed at pharmacological levels (but not at physiological levels), can restore glucose-induced insulin secretion in type 2 diabetes.^16,17 GLP-1 also has an important glucagonostatic effect. This effect was first described by Cathrine Ørskov and has later been shown to account for approximately 50% of the glucose lowering effect of GLP-1.^18,19 Importantly, the glucagonostatic effect of GLP-1 is also glucose-dependent and GLP-1 does not interfere with glucagon-induced rescue from hypoglycemia.²⁰ Reducing glucagon has been proposed as an independent mechanism for treating type 2 diabetes as early as the 1970s.^21,22 Thus, in the early 1990s, GLP-1 was positioned as a diabetes drug discovery candidate both addressing the pancreatic deficiency in type 2 diabetes with the incretin effect, promising insulin secretion with little risk of hypoglycemia, as well as addressing hepatic pathophysiology, promising lower hepatic glucose production via the glucagonostatic effect.

Personal Introduction

I graduated from university in 1989 with a degree in chemical engineering/biotechnology. At that point, I already had my eyes set on working for Novo, as it was called back then, but not in diabetes research. Being Danish, Novo was very visible in the scientific community, and I was a big admirer. I later got to know the CEO at that time, Mads Øvlisen, who was and still is very inspirational and had thoughts around ethics and corporate social responsibility that have been an essential part of why I always have loved my job—working in a company where I feel everybody tries to do the right thing. Of course, we do not always succeed but we try. I started working as a student in what is now Novozymes, who is a world leader in bioindustrial products and processes, and my first product on the market was a color detergent, or more specifically, my contribution together with my very good friend Brigitte Silau was to first purify and name the so-called 43 kDa cellulase that ended up being the specific component of many color detergents. What this does is to enzymatically remove the little strings of cotton from the surface of your clothes, making it look brighter. I also did some work trying to design a catalytic enzyme meant to bleach color in solution during washing, thus preventing transfer to lighter colored clothes.²³ I write about this experience because it was also my first learning of (unfair) competition in science, as we were never named inventors on the patent application for the 43kD cellulase. I did not know much of such things back then, just very eager to do something useful, and I moved to Diabetes Research at Novo Nordisk in 1991, where I became the youngest in the group looking to identify new treatments for type 2 diabetes. I set up assays to screen for various targets including the GLP-1 receptor (GLP-1R) where I transferred the assay from single tubes to 96-well plates, which saved us days of physical work back then, and I screened thousands and thousands of compounds as potential glucagon receptor antagonists.²⁴ I did not have a PhD degree, and still do not, and I was very happy to learn, from my colleagues back then, how one went about generating new ideas. Things changed rapidly, though, with lots of managerial turmoil in the company in 1992–94, and when I came back from maternity leave in 1994, the GLP-1 project was not in good shape and I was suddenly the only one left that knew anything about it. Mads Krogsgaard Thomsen had become Head of Research, a role he held until 2000, when he became (and still is) our uniquely inspiring CSO. Mads said to me “figure out what to do with GLP-1”, and there I was, 5 years of experience, only 3 of which were with diabetes, no PhD, and few colleagues left. I spent 3 months in my office debating with myself and the wall on what to do and developing my stamina to do it. I sometimes wondered back then why they even hired me, with no experience and no PhD. Maybe they saw what I later learned was an advantage for me; I do not think so much about what other people think of me, and I do not care too much about admitting if there is something I do not know or have been wrong. I also learned from my parents who did not have university degrees, but were entrepreneurs, to just start over if you fail or get no for an answer. All of those things have been important in my journey with liraglutide, which has been rocky at times.

Inventing Liraglutide

At Novo Nordisk, we had one GLP-1 project that failed and several tracks of ideas before we started with liraglutide. We first tried to develop a sustained release formulation of native GLP-1, as we had acquired some intellectual property rights related to native GLP-1 from Massachusetts General Hospital (Joel Habener et al.) and we worked through a couple of collaborations with other companies on this. Technically, it was possible, but we never went into clinical trials because the formulation caused severe skin reactions. Lilly had the same experience and Roche went all the way through phase 3 with taspoglutide with a similar concept but never filed for approval due to 25 cases of allergic reactions including some cases of anaphylactic shock.²⁵⁻²⁷

Alongside the native GLP-1 project, we tried to develop our thinking around the relationship between the pharmacokinetics and pharmacodynamics, the so-called PK/PD relationship of GLP-1. We did a clinical study where we learned that GLP-1 has to be present for 24 h/day in order to have the best effect on glucose lowering.²⁸ Having this knowledge was an integral part of my thinking back then; I had to come up with an idea fulfilling that goal. Marketing colleagues were very focused on a simple product with once daily dosing, making it a greater challenge with 24 h/day activity, since also at that time it was becoming known that GLP-1 gets degraded very quickly by the DPP-4 enzyme and cleared rapidly by the kidneys.²⁹⁻³¹ The major metabolite of GLP-1, GLP-1(9–36)amide, may work as a low potency antagonist on the receptor, but like other peptide-based antagonists, it may also act as a partial agonist.^32,33 The team I worked in as a young scientist was a mixed group of chemists, biologists, and pharmacologists; we made detailed studies of the contribution of the individual amino acids in a so-called alanine-scan of the peptide, and we designed DPP-4 protected analogues.^34,35 None of these things however solved the problem, and it was alone in my office in the fall of 1994 and early 1995 that I decided to focus on the idea of using fatty acid acylation to achieve the goal of once daily administration and 24 h/day activity. We had tried it earlier but failed to purify the compounds, as they stuck to the purification columns due to the lipophilicity of the early compounds we made.

The project was set up in what would today be called working by LEAN principles. I had very few resources, and thus, setting up a focused plan was key for success. The selection criteria in the project were the following: (1) receptor potency, aiming for less than 500 pM on the human receptor; (2) half-life longer than 10 h; (3) metabolic stability better than GLP-1, against DPP-4 and other proteases; and (4) to be as close to human GLP-1 as possible. The last criterion was theoretical; we did not screen for it. I wanted to focus on it because I heard colleagues talk about problems around animal insulins and that it only took one amino acid potentially to cause antibody responses in patients. It turned out many years later to be important, as the human GLP-1R agonists (GLP-1RAs) have much fewer and very low titer antibodies in patients.^36,37

The concept around fatty acid acylation is related to albumin binding. Albumin is the most abundant protein in plasma (approximately 4%, or 0.6 mM), and it is known to bind many drugs.^38,39 Often this binding is described as undesirable, limiting drug efficacy due to problems with a low free fraction of drug. However, it can also be desirable by leading to increased stability and protraction of the therapeutic effect. Albumin is a large molecule that consists of three homologous helical proteins each consisting of an A and a B domain. Binding of fatty acids to albumin is important physiologically, and albumin thus has many binding sites for fatty acids. Eleven binding sites have today been identified by crystallographic analysis.^38,40−44 The binding sites differ in affinity for different fatty acids. The highest affinity binding sites are FA4 and FA5 located in albumin domain III.

The design of analogues resulting in liraglutide is described in two publications.^45,46 Briefly, our findings were that long fatty acids could be attached to a variety of positions on GLP-1 without loss of potency measured using the human GLP-1R. The positions for acylation were selected to be in the C-terminal end, furthest away from the N-terminal region that previous studies had shown to be important for receptor affinity and activation.³⁵ We identified several potent compounds with plasma half-lives longer than 10 h that made them suitable for once-daily administration. From these compounds, liraglutide was selected for clinical development. The fatty acids were optionally extended with a “spacer” between the ε-amino group of the lysine side chain and the carboxyl group of the fatty acid. The spacer was added to increase the charge of the molecule right next to the fatty acid to increase the affinity to albumin. It had been described previously that positively charged amino acid residues on albumin interacted with the carboxyl terminus of fatty acids.^47,48 The spacer was also meant to give extra water solubility simply due to the extra charge added to the molecule making it more hydrophilic. Pigs were used as the model to test pharmacokinetic properties, as the size and subcutaneous (s.c.) skin anatomy of pigs is very similar to humans, making them excellent models for testing s.c. injectable drugs. Liraglutide was selected as the compound most like native GLP-1, with the highest receptor potency. One amino acid in the backbone GLP-1 part of the molecule was exchanged from a lysine to an arginine, to be able to only attach a fatty acid and spacer to the one remaining lysine in the sequence of GLP-1. We found that, in contrast to what the literature suggested at that time, we did not need specific DPP-4 protection for the once daily profile. Liraglutide is thus the only marketed GLP-1RA that does not have complete resistance to enzymatic DPP-4 inactivation. The length of the fatty acid was very important for the protraction in vivo. The longer the fatty acid, the longer the half-life obtained. Also, the uptake from the subcutis was protracted, as seen by the longer time to maximum concentration for C16 and C18 fatty acids.

Eventually, clinical studies documented the pharmacokinetic properties of liraglutide in humans to be similar to those in pigs.^45,49,50 This was an important finding back then, as it was a huge uncertainty. Following that, we continued to the first phase 2 clinical study. However, we got that wrong, because we did not yet know of the need for titration to overcome the gastrointestinal side effects of GLP-1RAs, so we made a very nice injectable drug, using doses from 0.6 to 0.9 mg once daily, with efficacy comparable to a sulphonylurea and no weight loss.⁵¹ We then had to do another phase 2 study, but we only had enough good manufacturing principle (GMP) produced compound for a 5-week study, which is why what ended up being the clinical proof of concept study for liraglutide was only 5 weeks long and was based on fasting plasma glucose rather than Hba1c.⁵² Clinical pharmacology studies documented that liraglutide did not alter the pharmacokinetics of other albumin binding drugs,^53,54 and the liraglutide pharmacokinetic profile was not substantially different in patients with mild renal or hepatic impairment.^55,56 Albumin binding, route of clearance, and decreased gastric emptying are parameters that could give rise to altered profiles in patients. Liraglutide was shown to be fully degraded within the body,⁵⁷ as would be expected, since it is composed of a naturally occurring fatty acid, a spacer that is a natural amino acid, and a peptide made up of natural amino acids. I sometimes call it a steak; it really is, protein and fat.

The Pancreatic Challenge–Part 1

Early on, we did several studies characterizing liraglutide’s effects in various animal models, documenting effects on insulin, glucagon, and glucose lowering. Also, around then, this was the mid-1990s, there was high belief in GLP-1’s effect to increase beta-cell mass^58,59 and that there was potential for stopping beta-cell decline, leading to GLP-1 being sometimes called the holy grail of diabetes treatment. Reflecting back on it now, I think we fell into the trap that all new things do; first, there is this amazing enthusiasm, but then reality kicks in, and it turned out GLP-1’s effect on glucose lowering is great but it is not a cure for type 2 diabetes. The story went like this: As early as 1987, GLP-1 was shown to not only increase insulin secretion but also increase insulin biosynthesis.⁶⁰ A paper in Nature in 1993 described that GLP-1 rendered previously glucose-incompetent β-cells glucose-competent again.⁶¹ This ability is however not unique to GLP-1—glucagon and most likely GIP are also able to do this.^17,62 From 1999, papers started to appear that GLP-1 also had the capability to increase β-cell mass by increasing proliferation of existing β-cells and neogenesis from duct cells to new β-cells.^58,59 Neonatal administration of exendin-4 was shown to prevent development of diabetes up to 8 months post-treatment.⁶³ We showed liraglutide was able to increase β-cell proliferation and mass more than exendin-4, illustrating that also for this effect of GLP-1 continuous administration is better than intermittent.⁶⁴ However, these studies were all in rodent models of diabetes and all in young animals. Studies have documented that there is a strong age-dependent restriction of β-cell regeneration in diabetic mice and rats and also a complex relation to glucose, where initially there is a rapid increase in proliferation in response to hyperglycemia, but as the hyperglycemia increases along with age, the proliferative capacity markedly decreases.^65,66 We did a series of studies with liraglutide in Zucker diabetic fatty (ZDF) and pancreatectomized rats illustrating some of these findings.⁶⁷ The increase in β-cell proliferation and mass was highly time- and glucose-dependent. When liraglutide was dosed to normalize blood glucose, there was a decrease in both β-cell mass and proliferation, whereas there was an apparent increase in proliferation and mass with longer treatment duration (6 weeks) that was not able to keep glucose normalized. Later we did a study with 13 weeks of liraglutide or exenatide treatment where a baseline group was also included, showing that these animals have profound diabetes progression with loss of both β-cell mass and the ability of the cells to proliferate, suggesting that GLP-1 analogues actually delay disease progression but do not prevent it.⁶⁸ We also published a paper showing that liraglutide inhibited both fatty acid- and cytokine-induced β-cell apoptosis.⁶⁹ These effects on plasticity and cell mass regulation turned out to be less relevant in humans, though the antiapoptotic effect may be relevant given it has been validated in human islets.⁷⁰ It will, however, be hard to ever separate direct effects of GLP-1 on insulin secretion and biosynthesis and its other beneficial effects, like regulating the transcription of glucose transporters and glucokinase, from what may be an effect on apoptosis. The effect on proliferation is most likely not relevant in humans. Not even after partial pancreatectomy do adult human islets seem to have the capacity to proliferate.⁷¹ A study with liraglutide has shown that human β-cells may proliferate, but it is an extremely rare event.⁷² GLP-1 and exendin-4 have also been suggested to potentially induce differentiation of human duct cells to insulin-producing cells.⁷³⁻⁷⁵ However, even when using rodent models, there is great controversy if neogenesis has any importance at all.^76,77

We turned to the gerbil Psammomys obesus to study the effect on the pancreas further.⁷⁹ The model is considered to be of higher relevance to humans. It is not a model of chemically induced diabetes or based on missing leptin functionality, like in ZDF rats and ob/ob and db/db mice. This gerbil has adapted to live under difficult conditions in the desert, and when subjected to free availability of food, it becomes diabetic spontaneously. This model shows a progression to diabetes that is similar to humans, first adapting to the glycemic challenge with increased β-cell mass and only later developing irreversible β-cell destruction.⁷⁸ Our study with liraglutide perhaps illustrates the relative importance of functional effects on insulin secretion and biosynthesis and those of β-cell mass. Liraglutide increased insulin content and β-cell function, but β-cell mass was not increased.⁷⁹ Liraglutide (and other GLP-1RAs) has consistently been shown to improve the insulin/proinsulin ratio in patients, a measure of improved beta-cell health and insulin biosynthesis. As such, GLP-1RAs are better insulin secretagogues than sulfonylureas, but GLP-1 does not seem to prevent the decline in beta-cell function over time that we initially hoped.

There still may be an opportunity for using GLP-1RAs in connection with islet transplantation, inspired by this protective effect of GLP-1 on β-cells. The antiapoptotic effect seems desirable for protecting the transplanted cells and thus securing some endogenous insulin production in this vulnerable patient population. In collaboration with the Shapiro group in Edmonton, we described liraglutide’s effects in different transplantation models. The first study described liraglutide in a marginal mass transplant model in mice.⁸⁰ Liraglutide markedly improved time to normalization of blood glucose; this was also true when the immunosuppressive agent sirolimus that has been described to be partially detrimental to the transplanted islets was present. However, the study also showed liraglutide had to be present continuously. The next study showed the effect of liraglutide in a pig model of marginal mass islet transplantation.⁸¹ Few animals were included in the study, but there was a retained graft function even 18 months after a 2-month treatment with liraglutide. Liraglutide did not improve insulin independence rates or blood glucose levels after the transplant but did increase insulin secretion. The third paper described the effects of liraglutide on human islets in culture and after transplantation.⁸² Liraglutide preserved islets and increased the size of islets. Again, the in vivo data showed the need for continuous presence of liraglutide to preserve islet function.

For some time, we also pursued the idea that GLP-1 may prevent diabetes. GLP-1 may have a role in the interplay between lowering body weight and having beneficial effects on glucose homeostasis. While GLP-1 clearly has a role in lowering body weight in obese patients without diabetes, it may also help to delay or prevent diabetes in these nondiabetic patients via the collective effects on β-cells as described above—protecting them and making them more glucose competent. The effects of liraglutide on diabetes prevention have been investigated in multiple rodent models. While studies in ZDF rats showed a potential of liraglutide to delay the progression of diabetes, this was a study in ZDF rats, a monogenic model with impaired leptin function due to a mutation in the leptin receptor.⁶⁷ The UCD-T2D rat is a polygenic model in which diabetes develops more progressively and thus may be better suited for studies of diabetes prevention.⁸³ In this model, liraglutide had more of an effect on delaying diabetes than did food restriction, and liraglutide further lowered levels of insulin, glucagon, and triglycerides in contrast to food restriction where these markers were either unaltered or even increased.⁸⁴ In yet another model, liraglutide reduced weight gain and glucose intolerance induced by the atypical antipsychotic agent olanzapine.⁸⁵ These findings may be clinically interesting, as weight gain and diabetes is a clinical problem in treatment with atypical antipsychotic agents, especially olanzapine and clozapine, which are used for treating psychosis but also have a higher risk of obesity and diabetes.^86,87 A recent clinical study has shown the potential for managing metabolic syndrome arising in patients treated with atypical antipsychotics, and the studies with liraglutide in obesity document to some extent the potential to prevent diabetes.^88,89 Further clinical studies are needed to explore if GLP-1 can prevent diabetes or prevent loss of beta-cell function by treating newly diagnosed patients.

Stay tuned for part two of the pancreatic story of liraglutide, the one that we never saw coming, but to stay historically correct, you first must hear the thyroid story.

The Thyroid Challenge

In 2003, we first learned that liraglutide was associated with C-cell tumors in rats. This was a big surprise. A few older studies described GLP-1R expression on thyroid calcitonin producing cells (C-cells), but the articles did not give any clues to the physiological role of the receptor.⁹⁰⁻⁹² We originally designed the liraglutide carcinogenicity studies as a one-species program, with the thinking that, as liraglutide is a close analogue of GLP-1, we would not need to do two species (which is the normal regulatory requirement for chemical drugs). With these C-cell tumors, however, we quickly started new studies in mice which is the other standard species used to assess the carcinogenic potential in animals. We also started a large program with approximately 20 scientists working for three years to try and decipher the human relevance. The key findings are described here and all published in a joint nonclinical and clinical paper.⁹³ C-cells constitute only a small fraction of the cells in the thyroid, and their function in humans is questionable; perhaps calcitonin primarily has a role during pregnancy and lactation.⁹⁴ Perhaps as a direct consequence of their low importance in higher species, C-cell numbers were found to be lower in primates compared to rodents, and in humans, the C-cell number is 44 times lower than that in rodents—the relative ratio based on an indirect comparison with published human data.⁹⁵ The GLP-1R was expressed in a human C-cell line, but species differences were found where the human cell line was shown to express much fewer GLP-1Rs than what is seen in rodents, measured by in situ ligand binding (ISLB), and no significant detection of GLP-1R mRNA in human thyroid tissue by in situ hybridization. This point was supported by literature data.⁹⁶ Rodent C-cell lines expressed a functional GLP-1R that all ligands of the GLP-1R activated to the same extent, only with differing potency, whereas a human C-cell line showed no activation following any GLP-1 ligands. In vivo, rats responded to liraglutide with dose-dependent increases in calcitonin, and after prolonged (up to 2 years) exposure, there was a dose-dependent increase in hyperplasia and C-cell tumors. However, it is known from the literature that rats have a particular C-cell tumor propensity.⁹⁷ Mice do not have this age-dependent increase in C-cells and showed acute increases in calcitonin, followed by hyperplasia, with liraglutide as well as exendin-4 administration. As with other effects of GLP-1, there was no memory effect,⁹⁸ and continued presence of the analogue was required for full response. Thus, when exendin-4 was dosed once daily in mice, there was only a transient calcitonin release, whereas with continuous infusion of exendin-4 the calcitonin levels were elevated for 24 h/day. Since monkeys have a similar lower number of C-cells as observed in humans and the monkey receptor is highly homologous to the human,³³ the monkey was considered a good model to study events of relevance for humans. Monkeys showed no signs of C-cell activation or hyperplasia after up to 60-fold higher doses of liraglutide than maximally used in humans, and up to 87 weeks of dosing. Monkeys have previously been shown to respond to vitamin D with increased C-cell numbers,⁹⁹ indicating that they may be a good model for C-cell proliferation. Calcitonin, as shown in our and other’s studies, appears to be a marker of activation of C-cells by GLP-1 in rodents. Eventually, human studies showed no calcitonin increases following liraglutide dosing up to 2 years.¹⁰⁰ In summary, the evidence pointed to a potentially important species difference: C-cell numbers are much lower in primates than in rodents; GLP-1Rs are expressed at a much lower level in humans (if any at all) than in rodents; primates including human GLP-1Rs do not appear to mediate calcitonin release or proliferation; mice and rats respond acutely with calcitonin release to GLP-1 analogues; following prolonged exposure to GLP-1 analogues, mice and rats respond with hyperplasia and tumor formation; humans and monkeys do not show any activation of C-cells following prolonged exposure up to approximately 2 years.

After approval of liraglutide, we did a few more studies. We investigated if the hyperplasia and formation of tumors depended on the presence of the GLP-1R and showed that in GLP-1R^–/– mice no hyperplasia was found.¹⁰¹ Medullary thyroid cancers (MTCs) in humans are very rare and primarily associated with activating mutations in the REarranged during Transfection (RET) proto-oncogene as found in the multiple endocrine neoplasia (MEN) syndromes. RET is a receptor tyrosine kinase in the GNDF family (glial cell-line derived neurotropic factor), expressed in cells derived from the neural crest and the urogenital tract. Activation induces dimerization and tyrosine phosphorylation. Several specific phosphorylation sites have been implicated in the MEN syndromes, but especially the tyrosine 1062 phosphorylation has been linked to activation of major intracellular signaling pathways.^102−104 Our studies showed liraglutide-induced hyperplasia in mice was not associated with activation of tyrosine 1062 but with activation of the mTOR pathway which is also activated by incretins in pancreatic β-cells.^101,105 The molecular processes in human MTC and those induced by GLP-1 analogues in rodents seem fundamentally different. Today, we still have no cases of MTC in any randomized studies with liraglutide, despite a few cases in control groups. Studies published after we did this foundational work showed very limited, if any at all, expression of the GLP-1R in a study with more human thyroid tissue samples.¹⁰⁶ Studies with the GLP-1R^–/– mice point to a role for the rodent GLP-1R in bone resorption; the GLP-1R^–/– mice were shown to have higher bone fragility.¹⁰⁷

The discussions around the human relevance of these findings caused a small delay in the approval of liraglutide in the USA. Today, I think most agree there is limited human relevance of these findings. We are however still awaiting a registry study that all companies who market a GLP-1RA are part of (ClinicalTrials.gov identifier NCT01511393).

Clinical Efficacy

The phase 3 clinical studies used to support the registration for treatment of type 2 diabetes with liraglutide were reported in 2008–9.^108−112 The liraglutide studies were designed to address all stages of the type 2 diabetes treatment cascade, from monotherapy to add-on to one or more oral antidiabetic agents (OADs).

Liraglutide was the first approved GLP-1 analogue to have a pharmacokinetic profile suitable for once-daily dosing (half-life of approximately 13 h).^49,50 Furthermore, the liraglutide profile is consistent with efficacy spanning a full 24 h/day, as even a low dose of liraglutide lowers blood glucose 24 h after the last dose.¹¹³ The same study showed that there is not really a substantial effect of GLP-1 analogues on fasting insulin and C-peptide, which is in agreement with a 6-week GLP-1 infusion study published a few years earlier.^113,114 Perhaps more importantly, glucose-induced insulin secretion is markedly stimulated by GLP-1, as documented in a glucose-clamp study for liraglutide.¹¹⁵

In phase 3 studies, liraglutide lowered HbA1c significantly better than glimepiride in monotherapy, but was not significantly better than glimepiride when added on to metformin, and significantly better than rosiglitazone and insulin glargine. HbA1c lowering was up to 1.5%, from a baseline of max. 8.6%, and weight loss was up to 2.8 kg from baseline over 26–52 weeks, with an increase for both glimepiride, rosiglitazone, and insulin glargine. These studies also documented a small but significant lowering of systolic blood pressure (SBP) of liraglutide, previously overlooked in studies with native GLP-1. Earlier literature on GLP-1 had shown a potent natriuretic and diuretic effect of GLP-1 that may partly explain the lowering of systolic blood pressure.^116,117 The liraglutide studies also documented a small increase in heart rate. This effect had also been seemingly overlooked in earlier studies with native GLP-1. Following the registration of liraglutide, two studies were published with exendin-4 and sitagliptin (DPP-4 inhibitor) as direct comparators.^118,119 These studies were the first randomized controlled studies to validate the importance of pharmacological vs supra-physiological levels of GLP-1, and of the pharmacokinetics and duration of action over 24 h. The liraglutide vs exendin-4 study documented that 24 h/day activation leads to a better effect on fasting blood glucose and Hb_A1C than two bolus injections of exendin-4.¹¹⁸ This is in agreement with the data on native GLP-1 showing 24 h infusion gave better glucose control than 16 h/day over a 7-day period.²⁸ The gastrointestinal tolerability was better for liraglutide. This may be explained by the difference in gastric emptying where exendin-4 lowers gastric emptying dramatically, whereas liraglutide has a much smaller effect.^120−122 Body weight change was similar to the two compounds, but an extension of the study showed that when switching patients from exendin-4 to liraglutide there was a small but significant lowering of body weight, FPG, H_bA1C. and systolic blood pressure, indicating that better weight loss can be achieved with full 24 h/day coverage of GLP-1.¹²³ One further explanation for the better efficacy of liraglutide once-daily over exendin-4 bidaily administered may be the existence of a higher incidence of neutralizing antiexenatide antibodies than that for liraglutide, a natural consequence of the higher human homology of liraglutide vs exendin-4 (97 vs 53%).³⁶ The liraglutide vs sitagliptin study documented that pharmacological GLP-1 concentrations lead to a substantially better glucose lowering (1.5 vs 0.9% H_bA1C) and weight loss (3.38 vs 0.96 kg over 26 weeks) than the levels obtained with the DPP-4 inhibitor.¹¹⁹ The data for β-cell function document that the presence of pharmacological GLP-1 levels 24 h/day gives better effects on fasting C-peptide and the proinsulin/insulin ratio, and HOMA-B (homeostasis model assessment of β-cell function) was markedly higher (28.70 vs 4.18% for liraglutide and sitagliptin, respectively, measured while still on drug).

Understanding the GLP-1R Structure and How It Led Us to Understanding GLP-1R Expression

The GLP-1R was cloned in 1992 by Bernard Thorens, who first cloned the rat and then the human receptor.^124,125 The receptor is very well preserved across mammalian species, along with the peptide, indicating the physiological importance of the hormone.^33,126 The GLP-1R is a G-protein-coupled receptor (GPCR) of the B family. Functionally, the receptor is coupled to the stimulatory G-protein G_s and adenylate cyclase activation. Numerous publications describe the intracellular signaling of GLP-1 in the β-cell, leading to glucose-dependent insulin secretion, insulin biosynthesis, and β-cell proliferation/apoptosis. An excellent review by Doyle and Egan describes these in detail.¹²⁷ New publications continue to further detail GLP-1-induced signaling and involve more complex pathways. Examples of seminal studies include the Wnt pathway involvement in mediating the effects of GLP-1 on proliferation and studies that show β-arrestin-1 mediated signaling to be involved in insulin secretion.^128,129 GLP-1-induced intracellular signaling is not well described for other cell types, but there are a few studies showing GLP-1-induced adenylate cyclase activation in, e.g., cardiomyocytes, endothelial cells, and neurons.^130−133 Downstream of cAMP, the signaling details vary in different cell types, but it seems that the GLP-1 receptor is coupled to adenylate cyclase activation in all cell types.

Since back in 1994, we have been interested in understanding the structure–activity relationship for GLP-1 and its receptor. We published the first alanine scan of GLP-1 and were also involved in creating the first GLP-1/glucagon chimera with Thue Schwartz.^35,134 Starting in 2005, I was part of starting a series of studies directed to understanding the structure of the GLP-1R.^135−139 These were primarily led by my then PhD student Steffen Runge, who has since relentlessly continued the work toward a full structure of the GLP-1R. Better understanding of the human receptor structure led Steffen to come up with an idea for generating specific antibodies for immunohistochemistry (IHC), namely, to use the extracellular domain of the human GLP-1R for immunization in GLP-1R^–/– mice. With that strategy, we generated what were likely the first specific monoclonal GLP-1R antibodies for IHC.¹⁴⁰ For many years, the work on GLP-1R expression methodology has been led by my colleague Charles Pyke who is an expert in both IHC, ISLB, and in situ hybridization. The first antibody was primate receptor specific, but later, we also generated one for use in rodents.¹⁴¹ They have both been made publicly available at University of Iowa’s NIH cofunded hybridoma bank (http://dshb.biology.uiowa.edu/). These antibodies became a great tool for us to better understand the pharmacology of GLP-1. Understanding the mechanism of action and potential safety issues of a drug starts with understanding which cell type and organs it primarily works on. Correct GLP-1R expression methodology unfortunately is, and remains, an issue in the literature.^142−145 We have all got IHC methodology wrong at some time, I think—at least I did. When we first tried to understand the pancreas pharmacology, we thought duct cells expressed the receptor, but later studies showed that was not the case. Instead, it turned out that the antibody we used was not specific, and we did not use multiple methods to make sure we had correct expression.¹⁴⁶ Also, when we started out investigating the rodent thyroid effects, we found GLP-1R expression in all species including humans with the same antibody, but in that work, we also did both in situ hybridization and in situ ligand binding and that was where we initially became suspicious of the antibody because the other methods did not confirm the findings and because GLP-1R^–/– mice also showed positive staining.⁹³ As also mentioned above in the thyroid section, later studies have clearly showed very limited expression in human thyroid.¹⁰⁶ The problems with IHC are not specific to the GLP-1R but rather are a challenge for all GPCRs.¹⁴⁷ Unfortunately, the problems persist and even some high-ranking journals fail to have strict criteria for evaluation of IHC results. For GPCRs, which are membrane-bound proteins, it is very easy to see the difference between staining that is clearly membrane associated and thus may be specific and diffuse staining spattered out all over a cell, which indicates the staining is not specific. Also, state-of-the-art methodology should always include validation by a second method when reporting a novel localization. This is especially important when the aim is to publish findings around the safety of drugs used in human patients, and authors and journals should be aware of these criteria. The next chapter in my story is unfortunately an example of how badly performed science almost destroyed an entire drug class, or actually two, the GLP-1RAs and the DPP-4 inhibitors.

The Pancreatic Challenge–Part 2

Early on, many studies reported GLP-1 to have beneficial growth and plasticity related effects on the endocrine pancreas, including proliferation and antiapoptosis of β-cells and potentially also neogenesis of duct cells. Due to the proliferative effect, it was from the beginning a concern whether chronic GLP-1R stimulation would lead to pancreatic insulinomas over time. We did not find such a risk in life-long dosing studies of rats and mice with liraglutide. We dosed approximately 1200 animals for 2 years (the equal of a lifetime) with liraglutide and did not find any association between liraglutide and any form of pancreas cancer.¹⁴⁸ Some animals were found to have islet cell adenoma or hyperplasia, but these were evenly distributed between the control group and the drug groups.

In 2006 and 2008, case reports were published linking exenatide to cases of acute pancreatitis,^149,150 and pancreatitis and GLP-1 became heavily debated. Because pancreatitis had occurred at a very low rate in the clinical trials and has an increased incidence in diabetes, it was not possible right away to reach a conclusion. Large pharmacovigilance studies of health care databases did not reveal an association between exenatide and acute pancreatitis.^151−154 One study made an analysis of the FDA AERS (Adverse Event Reporting System) database and found a correlation with both pancreatitis and pancreas and thyroid cancers.¹⁵⁵ This study was, however, problematic, as the AERS database consists of not fully validated cases and the FDA specifically states it cannot be used for such analysis.¹⁵⁶ Another publication also warned against such analysis.¹⁵⁷ Since large clinical outcome studies were not available at the time, animal studies came into focus.

In 2009, a study using a human amylin transgenic rat model reported that sitagliptin induced ductal metaplasia and duct cell proliferation. Also, one case of pancreatitis was found among eight sitagliptin-dosed animals. A GLP-1 analogue was not dosed, nor was GLP-1 measured increased in the animals.¹⁵⁸ In 2010, a study indicated some changes to the pancreas following exendin-4 treatment in normal rats: increased inflammation and pyknotic nuclei.¹⁵⁹ The paper also reported a small increase in lipase activity but not in amylase activity. This paper was accompanied by a commentary, indicating that there was a big overlooked problem with GLP-1 and pancreatitis.¹⁶⁰ On the other hand, two articles using exendin-4 and/or liraglutide and several different rodents did not document any pancreatitis, no increased lipase or amylase activities was reported, and no worsening of caerulin-induced pancreatitis either.^161,162 Those studies reported what seemed to be a protection of the pancreas with up-regulation of anti-inflammatory proteins and a dose-dependent reduction of amylase and lipase levels following exenatide treatment. In 2012, we published our 2-year dosing carcinogenicity studies and a study using diabetic rats, which was part of the FDA negotiated postapproval commitment for liraglutide.^68,148 In the same issue of Diabetes as we published our carcinogenicity study, there was another report using a new model of chronic pancreatitis/early pancreatic cancer where PDX-1/Kras mice were used.¹⁶³ The papers were accompanied by two commentaries,^164,165 and again, especially one of these commentaries caused quite some attention.

Several things are important in elucidating the discrepancies between the studies that caused all of this attention. Most importantly, one must discern whether the problem to address is acute pancreatitis, chronic pancreatitis potentially leading to pancreatic cancer of any sort, or cancer in the endocine cells. Several of the studies mentioned above do address if there is actual pancreatitis. The studies with the most animals and the most well-defined protocol for examination of the pancreas following internationally established guidelines were (at the time) the industry studies from Amylin^161,166 and our studies from Novo Nordisk.^68,148,167 Such studies use very large cohorts of animals (typically around 50 per group) and use independent readings of sections by two trained pathologists and did not find any drug-induced pancreatitis. Some of the nonindustry studies used fewer animals, sometimes down to n = 5, had uneven and small group sizes, and used no trained pathologists.^158,159 While potential side effects should certainly be taken very seriously and any safety related finding should always get the benefit of the doubt, the authors of studies that address the safety of drugs marketed for patients must also take extra care to use well-validated methods, a reasonable number of animals, and detailed histological scoring of tissues. Of the studies using GLP-1 analogues, none reported increased pancreatitis. There was one affected rat in one of the papers, but that was treated with sitagliptin.¹⁵⁸ It was never clear what the mechanism might be, as the theories kept switching around between pancreatitis, pancreas cancer, acinar growth, insulinomas, or glucagonomas, but for quite some time, it was proposed that the GLP-1R expressed on duct cells was the problem. The data supporting this was however based on antibodies that do not really measure the GLP-1R; they have been repeatedly shown to be unspecific,^142,144 and the one most widely used (ab39072) has even later been withdrawn from the market by the company that sold it. Unfortunately, papers are still being published with such antibodies, even in highly ranked journals. The two studies we have published use complementary methods, and independent groups have reported similar results, that the human pancreas expresses the GLP-1R primarily in the beta-cell of the pancreatic islet and there is a lower level of GLP-1R expression in acinar cells.^140,168 Duct cells do not appear to express the receptor; occasionally one can find cells budding off from duct that have GLP-1R expression, but they also have insulin expression and appear to be newly formed endocrine cells.

We spent years investigating these concerns, and it is reasonable if there is any potential risk to patients then it should be given all possible consideration. We debated potential mechanisms for pancreatitis: was it direct or indirect? We meticulously went through all of our data again and discussed what else we could do. Apart from the focus on GLP-1R expression, we also published another toxicology study with nonhuman primates focused on pancreatic size.¹⁶⁷ The FDA and EMA have since also jointly published their view, highlighting that they have not seen any drug-induced adverse findings in the pancreas in over 250 studies with various incretin-based drugs.¹⁶⁹

When the pancreatic and thyroid safety was appropriately investigated and elucidated, I began spending a lot more time on the weight loss mechanism as we approached phase 3 clinical studies for the obesity indication.

Understanding Weight Loss

It was always the idea to develop liraglutide as a diabetes drug as well as an antiobesity drug. Liraglutide was advanced for clinical development in 1997, and by that time, we did have proof-of-concept in animals for the obesity indication, but it was not until 2014 that we obtained the license to market liraglutide for obesity (the diabetes license was obtained in 2009). We started addressing weight loss in rodent studies back in 1995, on the basis of knowledge from our colleague Ole Madsen who had a rat strain with tumors that produced glucagon, GLP-1, and GLP-2. Those rats simply starved themselves to death.¹⁷⁰ We found that our analogues had potent effects on food intake in a liquid diet intake model (data not published), and we continued planning clinical studies for both indications. When the first papers published in 1996 showing that native GLP-1 had effects on food intake when injected directly into the rodent brain,^171,172 we already knew this effect could also be found with peripheral dosing. We published our first study focused on the weight loss effect in 2001, showing weight loss in both normal and obese rodents but also a marked effect to increase diuresis and decrease water intake.¹⁷³ Water homeostasis was initially severely affected, but tachyphylaxis to or compensation for these effects led to normalization of water homeostasis within a few days. This effect of GLP-1 is clinically relevant but has never received much attention, likely because there is such profound difference between acute and chronic dosing. There was much skepticism to GLP-1 as a satiety agent in those early days, and it was suggested GLP-1 was merely an agent that caused taste aversion.¹⁷⁴ We published two new studies the following years, in rodents and higher species such as pigs, documenting weight loss and effects to change food choice in a chocolate-fed rat model.^175,176

Initially, we thought (probably like many others) that the effects of GLP-1 were mediated through the vagus nerve and gastric emptying, and in the New Drug Application (NDA) for liraglutide, we wrote (in 2008) that we did not see uptake in the brain. After the approval for diabetes, there was more time to think about the mechanism for weight loss and we started doing more studies. The first realization that we might be wrong about the brain was when we realized that, while GLP-1 physiologically and acutely has a powerful effect on gastric emptying (GE), there is rapid tachyphylaxis to this effect and as a result liraglutide has only a minor effect on GE in patients.^122,177 We did a study using rats comparing GE of exendin-4 and liraglutide directly and found the same pattern,¹²⁰ thus validating the translational value of rats for the clinical findings. With exendin-4, like GLP-1, there is a marked lowering of GE, but with liraglutide, there is much less (but still a significant lowering, though). Since both compounds lowered body weight to the same extent, then GE was concluded not to be the likely mechanism for the effect on weight lowering. GLP-1 lowers energy intake under fasting conditions, again making it unlikely that GE is an important mediator.¹⁷⁸ It was clear from early human studies that GLP-1, as well as liraglutide, primarily lowers energy intake and has little effect on energy expenditure,^122,179,180 and we started to take a deeper interest in understanding the brain effects of liraglutide. In 2014, we published a paper that documented that liraglutide was able to access the brain in a GLP-1R-dependent manner.¹⁸¹ It does not as such cross the blood–brain barrier, as that does not really make sense for a peptide or a protein, but it has access to circumventricular organs like the median eminence below the third ventricle and the area postrema in the hindbrain, as well as some others. Liraglutide also accesses a few specific sites in the hypothalamus, namely, the arcuate nucleus and the paraventricular hypothalamus. We documented a molecular mechanism in the arcuate with internalization into the POMC/CART neurons and indirect inhibition of NPY/AGRP neurons. Liraglutide was approved for the treatment of obesity in 2014. The label text for the obesity indication for liraglutide includes some of the data around brain uptake and effects in the brain.

In later studies, we have tried to understand the hindbrain effects better, and it seems that, through the hindbrain, liraglutide signals to the lateral parabrachial nucleus and from there to the amygdala and other parts of the brain.¹⁸² We also published two detailed mapping studies of GLP-1R expression in the brain. More than 50 neuronal populations express the GLP-1R.^141,183 Only about 10 of them are directly accessible by liraglutide (at the detection level we have), and in another eight populations, there is cFOS activation without measurable compound access.¹⁸² We still have a lot to learn about the specific neuronal mechanisms that are activated in the brain by peripherally dosed GLP-1RAs. We need to understand the interconnections between the many different GLP-1R positive cell populations that have been shown in individual studies to be involved in appetite regulation and food intake. The rapidly evolving fields of bioinformatics and data science will likely help develop methods that can help us understand how, and in what order, the signals pass through the brain. The next chapter in my story builds on our initial focus on the brain and how that led us to think more about other effects in the brain.

Going to the Brain

Inspired by trying to better understand the antiobesity effects and following the literature where the first paper published in 2002 indicated that GLP-1 could be more broadly relevant for neurodegenerative diseases,^184,185 I started paying attention to other effects in the brain and took an interest in understanding the potential benefits of GLP-1 in neurodegenerative diseases like Alzheimer’s. Most drug development efforts have pursued the antiamyloid target strategy with various antibodies aimed at clearing amyloid fibrils from cerebrospinal fluid. GLP-1 is something entirely different. While the mechanism is not fully understood, there are potentially multiple effects that could be beneficial, and numerous publications describe an effect in animal models of Alzheimer’s disease. The GLP-1R^–/– mice provide additional support for the physiological importance of GLP-1 in cognition, as these mice have a phenotype characterized by a learning deficit that is restored after hippocampal Glp1r gene transfer. In addition, rats overexpressing GLP-1R in the hippocampus show improved learning and memory.¹⁸⁶ While the earliest data suggested this hippocampal mechanism for the effect on memory and learning, it may well only be part of the story that involves several effects including the following: reduced neuroinflammation, improved brain insulin sensitivity, blood flow, glucose metabolism, and all resulting eventually in less amyloid plaques and Tau fibrils. We found positive effects of liraglutide in two animal models, a natural aged model, the SAMP8 mice, and a Tau model.^187,188 We did not find any effect in the widely used APP/PS1 model, and because there are several such reports in the literature, we did two studies with 5 months of treatment and we could not find any effects.¹⁸⁹ A small clinical study (N = 38) with liraglutide has shown an increase in neuronal glucose metabolism but no significant effects on cognition,¹⁹⁰ and a large, placebo-controlled, multicenter study is ongoing with 100 patients/arm (ClinicalTrials.gov identifier NCT01843075). There is also some evidence for the relevance of GLP-1 in Parkinson’s disease; here two clinical studies have indicated a protective effect on motor dysfunction.^191,192 There is no measurable GLP-1R expression in substantia nigra dopaminergic neurons which are lost in Parkinson’s, but those neurons could be modulated by synapsing cells that express GLP-1Rs, or effects could be indirect through modulation of other neurotransmitter pathways. A recent publication in Nature Medicine suggests a new mechanism whereby activation of GLP-1Rs on microglia can prevent conversion to toxic astrocyte phenotypes which can lead to neuronal death.¹⁹³ This mechanism appears very relevant for current thinking in the Parkinson’s disease field and may be potentially relevant in Alzheimer’s as well as other neurodegenerative disorders like depression and stroke. It may also explain why we did not see any effects with liraglutide in a Parkinson’s mouse model with complete chemical knockdown of dopamine neurons.¹⁹⁴

Reports also implicate benefits of GLP-1 across a range of more classical CNS indications (i.e., depression, stroke, seizures, traumatic brain injury, addiction) as well as more limited data for rare CNS diseases (i.e., Huntington’s Disease, multiple sclerosis, amyotrophic lateral sclerosis, and narcolepsy). Given the well-established safety with proven cardiovascular risk reduction of long-acting GLP-1R agonists, it seems prudent to consider GLP-1 for several other indications where scientific evidence indicates it may be useful.

Understanding CV Benefits

When the LEADER study published showing CV benefits in 2016,¹⁹⁵ most people were surprised. I cannot of course say I had fully expected the results, but there were many animal publications indicating beneficial effects, so I was hopeful.^130,196,197 Immediately—in the space of hours—after the LEADER publication, Dan Drucker published a review citing over 100 primarily animal scientific papers about GLP-1 and cardiovascular effects.¹⁹⁸ It is easy to say there is no translation, of course, and many people have limited views on animal pharmacology. In my opinion, animal pharmacology should never be used as proof but may be useful to understand potential mechanisms when human outcomes have been established. Richard Simpson and Anthony Dear in Australia designed the first study with liraglutide in ApoE^–/– mice published in 2013, showing reduced atherosclerotic burden,^199,200 and Dan Drucker and Mansoor Husain in Toronto published the first study showing direct effects on the heart in 2009.¹³⁰ The LEADER study was published in 2016 and showed consistent effects on all three components of the MACE endpoint, leading to the hypothesis that an underlying effect on atherosclerosis is the most likely clinical mechanism for the benefit.¹⁹⁵ At Novo Nordisk, two of my colleagues (Bidda Rolin and Gunaj Rakipovski) relentlessly pursued setting up various models. We have been able to confirm a small effect of reduced aortic atherosclerotic burden in both ApoE^–/– and LDLr^–/– and consistently reduced anti-inflammatory footprints in the atherosclerotic tissue, which has no direct GLP-1R expression.²⁰¹ The effect is independent from both glucose lowering and weight loss and may be mediated through reduced systemic inflammation. GLP-1 may reduce systemic inflammation by improving the gut barrier function, either via GLP-1R-induced duodenal Brunner’s gland mediated mucous secretion improving the intestinal barrier function or via GLP-1R expressing intraepithelial intestinal lymphocytes.^202,203

Reflections

Here ends my story of liraglutide. Several reflections come to mind.

GLP-1R^–/– mice have relatively boring phenotypes. If GLP-1 had been an idea starting out on mouse genetics, then likely there never would have been a GLP-1RA drug class. Brave clinicians like Michael Nauck, Suad Efendic, and Steve Bloom went into patients on the basis of limited data. It is worth reflecting upon, considering the present hype around essential and required receptors based on mouse genetics. Leptin is one such idea with elegant genetics, which sadly did not work in treating general obesity. There are no easy solutions; sometimes animal pharmacology is predictive, and sometimes not. Collaboration is essential, for basic scientists, clinicians, and the drug industry to work together to take the best ideas to clinical testing.

Going oral? I spent a considerable amount of time in the early 2000s together with many colleagues at Novo Nordisk and Alanex, a small biotech in La Jolla that does not exist anymore, on discovering the first small molecule GLP-1RA,^204,205 a program led by my favorite chemist colleague Jesper Lau, who is also one of the lead inventors of semaglutide.²⁰⁶ The GLP-1RA small molecules were never drug-like enough to advance to clinical trials, but they displayed a very interesting mechanism of positive allosteric agonism. In the last 10 years at Novo Nordisk, we have developed an oral formulation of semaglutide, a unique tablet formulation where we have shown that the absorption takes place in the stomach, and not in the intestine as most small molecules.²⁰⁷ With liraglutide, it is not possible, likely because the fatty acid used for acylation of liraglutide is a monoacid that can serve as a membrane anchor, something suggested by one of the reviewers in the very first paper we published with the liraglutide compound chemistry. That paper was rejected at first, but we resubmitted it to the same journal, and given arguments that we were in phase 1 clinical studies, it was accepted.⁴⁵ I did not have any data to address the question of membrane anchoring, so I answered with this sentence which is in the paper: “However, the selective activation of the GLP-1R evidences that this phenomenon is not important for these compounds”. I am not particularly proud of that answer, which does not really address the question, as they can be specific and still have partial membrane anchoring, and I have over the years thought about how we could prove this but have not come up with a clear answer.

Studying the mechanism(s) of action for drugs is important. Without such understanding, prescribers and patients are poorly informed. Also, without knowledge on the mechanism of action, it is not easy to understand potential side effects.

Many drug classes may work across several diseases, and GLP-1 is one such drug class. There are several other examples like that, especially various anti-inflammatory drugs. I was greatly inspired by reading a Harvard Journal of Law and Technology review a few years back that my lawyer/tech nerdy daughter sent to me²⁰⁸ and which was recently published in a shorter form in Science Translational Medicine.²⁰⁹ This paper discusses how one could imagine a future where all information on individual patients, diseases, and drugs is used to create algorithms that can diagnose a patient and decide what drug is best for the individual patient. The article discusses the various legal, financial, and regulatory problems around such algorithms.

With me, a very young and female scientist as project lead, I have to say the project could have been closed many times. I think it is safe to say that we still have some way to go for female scientists to be recognized similarly to men in research, whether in academia or industry. My lack of a formal PhD made some people look down on me, and I was fortunate to be introduced to Jens Juul Holst early on. Jens was my guru for many years, guiding me until I could stand my own ground. So many people at Novo Nordisk or externally did not believe in GLP-1, and I could have given up several times when it was just tough to keep ignoring that very few people wanted to work with me. Having Jens to talk to and knowing how respected he was made me continue to champion liraglutide at Novo Nordisk. Of course, also very importantly, I always had the support of Mads Krogsgaard Thomsen, Head of Research from 1994 and CSO from 2000. Peter Kristensen headed up the development of liraglutide for many years, and later the entire clinical development at Novo Nordisk. He was immensely important for the successful development of liraglutide, and for the high quality of our clinical trials. Later, when I started to look more into understanding effects in the thyroid, in the brain, or on the cardiovascular system, Randy Seeley and Dan Drucker became my gurus and still are. I can never be an expert the same way they are. Randy taught me a lot about brain circuits, and we have had many debates on necessary and sufficient GLP-1R positive neuronal populations, versus GLP-1R populations that are just involved but not fully mediating an effect. Dan Drucker is an icon in so many ways and should be an example to all for his diligence in doing things properly. Dan taught me a lot about so many things.

In 2014, I defended a doctoral thesis in Scientific Medicine at Copenhagen University that is based on my work with GLP-1 and liraglutide, and in 2015, I was evaluated and approved as adjunct Professor in Translational Medicine at Aarhus University, also based on my work with GLP-1 and liraglutide. Part of my thesis has been rewritten for the use in this paper.

Lastly, I want to pay tribute to Heather Millage who sadly died from brain cancer in 2018. Heather was responsible for the marketing of liraglutide, and she was a wonderful inspiring person, always looking out for other people and causes she could help, always with a bright smile and a positive everything-is-possible attitude. She is dearly missed, and I keep her picture on my office wall looking at me every day. When in doubt of what to do, I look up at Heather and think what she would have done, and then I do that.

The author declares the following competing financial interest(s): I am a full time employee of Novo Nordisk who markets liraglutide for the treatment of diabetes and obesity, and I am a named inventor of that drug.