Serine Racemase Showdown: Clash of the Crystal Structures

Having been sitting on an X-ray crystal structure and half-finished manuscript for quite some time now, imagine my chagrin when during a casual perusal of the PDB, I found an extremely similar structure had just been deposited to the one I was aiming to publish.

The deposited structure (5X2L) is for wild-type human serine racemase (SR) and the associated paper focuses on in silico screening and medicinal chemistry, while my structure contains two cysteine-to-aspartate point mutations (C2D, C6D) with a (speculative) paper that explores the crystallography and biophysical side. As well as briefly summarising the findings of the study, I thought it would be fun to compare our two crystal structures in what I have dubbed…

Clash of the Crystals

Chloe 1

But firstly, some background on the contestant/s: 5X2L (published) vs. CRK1 (mine, unpublished).

In the blue corner…(and the red corner)…weighing in at 37.4 kDa, we have a pyridoxial-5’-phosphate (PLP)-dependent enzyme that catalyses the racemisation of L-serine to D-serine, as well as the α,β-elimination of water from L-serine. This forebrain-localised enzyme produces about 90% of D-serine in the brain, and because D-serine is a co-agonist of N-methyl-D-aspartate glutamate receptors (NMDARs), SR inhibition has been touted as an up-and-coming approach to indirectly modulate NMDAR activity. This is a potential game-changer for treating disorders underpinned by NMDAR overactivation, such as neuropathic pain, neurodegenerative disorders, and epileptic states.

So let’s hear it for…serine racemase! [Thunderous applause]

 Round 1: Paper summary

Anyone well versed in SR literature will know the paper in question, ‘Design, synthesis, and evaluation of novel inhibitors for wild-type human serine racemase’ by Takahara et al. (2018)1 is an additional chapter to an ongoing story. Several groups have previously tried to identify new SR inhibitors that are potent, selective, and structurally distinct from the countless amino acid analogue inhibitors that are already well-described2, and for many this has proved to be a challenging endeavour.

The status quo shifted when a series of dipeptide-like inhibitors with a clear structural motif and slow-binding kinetics was identified by Dixon et al. (2006)3, which later provided the query molecule for an in silico screen performed by the same group behind 5X2L4. The resulting inhibitors contained an essential central amide structure with a phenoxy substituent, and substitution of parts of the structure for heavier halogen atoms such as bromine and iodine produced derivatives with improved inhibitory activity (comparable to classical SR inhibitors), binding affinity, and ligand efficiency. The Mori group took their explorations even further by testing the most potent derivative in vivo to demonstrate the SR inhibitor suppressed neuronal activity-dependent Arc expression to regulate NMDAR overactivation5.

The current paper expands on these studies by firstly, solving the crystal structure of wild-type SR for molecular docking and in silico screening; secondly, using these methods to identify new SR inhibitors related to their previously described peptide compounds; and thirdly, testing these inhibitors in an in vitro assay. The team synthesised 15 derivatives, of which one showed relatively high inhibitory activity, making a nice addition to their growing rolodex of peptide SR inhibitors.
Chloe 2

Figure 1. Structure and binding pocket of the novel peptide SR inhibitor derivative identified by Takahara et al.

Round 2: Clash of the Crystals!

Both contestants were crystallised using the sitting-drop vapour diffusion method in very similar experimental conditions (Table 1). Both structures were determined to a highly respectable resolution of 1.8 Å, and organised into a large domain and small domain connected by a flexible loop region (Fig. 2). The PLP cofactor (Fig. 2; yellow sticks), on which SR is dependent for its catalytic activity, can be seen covalently linked to Lys56.

Table 1. Summary of key features of 5X2L and CRK1.

Feature 5X2L CRK1
Crystallisation conditions 10% PEG 8000, 5 mM MgCl2, 0.1 M Bis-Tris pH 6.0, 10% ethylene glycol, 20 °C 15% PEG 3350, 250 mM MgCl2, 0.1 M Bis-Tris pH 6.5, 20 °C
Resolution 1.8 Å 1.8 Å
Space group P212121 P21
a, b, c (Å)

α, β, γ (°)

80.1   112.6   88.0

90.0   90.0   90.0

69.0   53.8  79.4

90.0   106.1   90.0

Crystal system Orthorhombic Monoclinic
No. residues resolved 305/340 321/340
Ligands PLP, Mg2+ PLP, Mg2+


SR belongs to the fold-type II family of PLP-dependent enzymes, meaning it contains a β-sheet core surrounded by α-helices, with the active site located in a cleft between the two domains. Accordingly, both domains of 5X2L and CRK1 contain a parallel-stranded β-sheet surrounded by nine α-helices in the large domain and three in the small domain. A magnesium ion (pink sphere) that helps to stabilise protein folding and increase maximal activity6 is octahedrally coordinated by three water molecules, the acid groups of Glu210 and Asp216, and the carbonyl oxygen of Ala214.

Chloe 3

Figure 2. Overall X-ray crystal structure of the human SR holoenzyme CRK1. Residues are coloured from red to violet, N-terminus to C-terminus, and all helices are numbered 1–12 based on the order they occur in the polypeptide sequence. Each SR monomer comprises a large domain (helices 1–3 and 7–12) and small domain (helices 4–6) connected by a flexible loop region.


CRK1 boasts good ordering of residues, with only a few not well-defined: 1–3, 132–135, and 339–340. Aside from those at the C- and N-terminus, which are often poorly resolved during structure solution anyway, the only other undefined residues (132–135) were localised to the top of helix 5 in the highly-mobile small domain. 5X2L shows similarly undefined residues at the termini (1–2, 318–340) although in addition it is also missing residues of the flexible loop region (67–76) that connects the two domains.

Solvent-exposed loops are notorious for being tricky to model due to their high occupancy. The loop may be visible in CRK1 but not 5X2L because CRK1 has the help of its (symmetry) mates. By viewing the symmetry partners in the crystal lattice, the loop residues 69–73 are seen to be stabilised by a water-mediated interaction between Leu72 in one monomer and Lys221 in an adjacent monomer.

These favourable contacts may not occur for both structures because 5X2L crystallised in the orthorhombic space group P212121 while CRK1 crystallised in the monoclinic space group P21. Both possess 2-fold symmetry, but differences in molecular packing would have influenced whether the loop region would be suitably positioned to receive stabilising crystal lattice contacts.

Round 3: Best [Super]Pose!

A superposition of 5X2L and CRK1 (Fig. 3) revealed that, unsurprisingly, the two structures were well aligned with a Cα RMSD of 0.55 Å. Any remaining conformational differences are likely to result from the unresolved loop region, the missing helix and polypeptide strand that make up residues 318–340, and random structural variations. So hardly a ‘clash’ but at least it makes for a nice picture.

Chloe 4

Figure 3. Superposition of the X-ray crystal structures of 5X2L (blue) and CRK1 (red).

By now there is no doubt you are wondering who the champion is of the undeniably riveting Clash of the Crystals.

The answer is both, and neither, because any discovery that contributes to scientific advancement is a champion in my book J

Yes, even when said discovery beats me to the punch.

Blog written by Chloe Koulouris


  1. Takahara S, Nakagawa K, Uchiyama T, Yoshida T, Matsumoto K, Kawasumi Y et al. Design, synthesis, and evaluation of novel inhibitors for wild-type human serine racemase. Bioorg Med Chem Lett. 2017 Dec 13.
  2. Jirásková-Vaníčková J, Ettrich R, Vorlová B, Hoffman HE, Lepšík M, Jansa P et al. Inhibition of human serine racemase, an emerging target for medicinal chemistry. Curr Drug Targets. 2011 Jun; 12(7):1037-55.
  3. Dixon SM, Li P, Liu R, Wolosker H, Lam KS, Kurth MJ et al. Slow-binding human serine racemase inhibitors from high-throughput screening of combinatorial libraries. J Med Chem. 2006 Apr; 49(8):2388-97.
  4. Mori H, Wada R, Li J, Ishimoto T, Mizuguchi M, Obita T et al. In silico and pharmacological screenings identify novel serine racemase inhibitors. Bioorg Med Chem Lett. 2014 Aug; 24(16):3732-5.
  5. Mori H, Wada R, Takahara S, Horino Y, Izumi H, Ishimoto T et al. A novel serine racemase inhibitor suppresses neuronal over-activation in vivo. Bioorg Med Chem. 2017 Jul 15; 25(14):3736-45.
  6. De Miranda J, Panizzutti R, Foltyn VN, Wolosker H. Cofactors of serine racemase that physiologically stimulate the synthesis of the n-methyl-d-aspartate (nmda) receptor coagonist d-serine. Proc Natl Acad Sci U S A. 2002 Oct; 99(22):14542-7.



Large Scale Study Shows Antidepressants are More Effective than Placebo

This week an article in the Lancet has shed light on the controversy surrounding antidepressants.[1] Psychiatric disorders account for 22.8% of the global burden of disease, of which depression is the leading cause. In 2016 there were over 64 million prescriptions issued for antidepressants, which is more than double the amount issued ten years previously. Until now there has been much debate regarding the effectiveness of antidepressant drugs in treating this debilitating disorder. This study has been pivotal in providing evidence in addressing this controversy, as previous studies have not adequately examined the long term effects of antidepressants.

The study looked into 522 trials involving 116,477 patients and found that all antidepressants investigated were more effective than placebo. However, they weren’t all equally effective: It found that the drugs ranged from being a third more effective than a placebo to more than twice as effective. Interestingly, the findings showed that escitalopram, mirtazapine, paroxetine, agomelatine, and sertraline were the most efficacious in adult patients. On the other hand, reboxetine, trazodone, and fluvoxamine were found to be the least effective of the drugs tested, with effects diminishing over time. Many patients stop using antidepressants after only a few weeks, which may have contributed towards skewed results in previous investigations as many of the drugs tested are only effective after long term use. However, researchers noted that most of the data in the meta-analysis covered eight weeks of treatment, although it did not take into account problems that may emerge from longer term us of the drugs.

Overall, clinicians consider this study to be an important piece of evidence in encouraging patients to pursue treatment options for depression, including antidepressants where necessary. Hopefully, this seminal article will help remove the stigma surrounding depression and the use of antidepressants. More than ever, mental health is being acknowledged as a topic for serious discussion and in light of this article more awareness needs to be made about the treatment options for these disorders which affect one in four adults in the UK.[2]

Blog written by Rachael Besser


1 Cipriani, Andrea et al. Comparative efficacy and acceptability of 21 antidepressant drugs for the acute treatment of adults with major depressive disorder: a systematic review and network meta-analysis

2 McManus, S., Meltzer, H., Brugha, T. S., Bebbington, P. E., & Jenkins, R. (2009). Adult psychiatric morbidity in England, 2007: results of a household survey. The NHS Information Centre for health and social care.


Case placement in Emeryville

I have recently returned from my PhD Case placement at the Novartis Institutes for Tropical Disease (NITD) who recently relocated to Emeryville California. NITD drug discovery is focused in 3 areas; malaria, cryptosporidiosis and kinetoplastid diseases.

My PhD project has been to identify drug-like inhibitors that inhibit an enzyme that has been identified as a target for the kinetoplastid pathogen that causes African sleeping sickness, trypanosoma brucei. Whilst on my placement at NITD I was fortunate to work alongside scientists that have been involved with identifying drug candidates against the same pathogen, and have parasite assays in place to examine the effects of drug molecules against them. I was able to culture the live parasite and it was exciting to see that the compounds that I had designed and synthesised during my PhD inhibited the growth of the parasite.
During my time at NITD I was also able to work in the research chemistry laboratories, synthesising further analogues against my enzyme target, it was a great opportunity to work in a cutting edge research environment with a variety of instruments that facilitated day to day chemistry experiments. Over the course of my placement I was able to identify new chemical analogues that showed improved potency against the parasite, that I am currently following up back at the Sussex Drug Discovery Centre.

Ryan 1

At the weekends I was able to do a bit of travelling around California, some of the highlights were visiting Yosemite national park (above), cycling over the Golden Gate Bridge and visiting Alcatraz.

Blog written by Ryan West

Beware Greeks bearing gifts

The publication that I wanted to discuss was bought to my attention whilst reading Derek Lowe’s superb blog “in the Pipeline”. I would wholeheartedly recommend readers of this particular website to read the original blog discussion of the publication there. The link to in the Pipeline article is here and the link to the original article is as follows

In the publication a group at Dundee University were exploring a fragment based approach, against an enzyme which has a role in ubiquitin conjugating called Ube2T. The team had been successful in identifying a fragment hit in an earlier publication. In this first publication they had utilised differential scanning fluorimetry (otherwise known as a thermal shift assay) and followed up with biolayer interferometry, as an orthogonal technique to confirm primary hits from the DSF primary screen. To further generate confidence, ligand observed NMR was carried out.

From this work a specific fragment was chosen as the top hit and was then measured with 15N- labelled NMR and ITC measurements against which at this stage showed positive binding. At this point analogues were ordered around this fragment and this is where the project hit some rough water. None of the analogues seem to show any significant SAR, and the limited number of analogues that did bind were much weaker in potency than the original hit

In parallel, co-crystallization of the fragment against Ube2T was attempted. While this was unsuccessful in showing specific binding, it did highlight a rearrangement of the protein, and the presence of a metal ion close to the catalytic residue, this metal ion was then identified as Zn2+

As no zinc was present in the co-crystallization buffer system, the team investigated the original fragment hit compound, running a zincon colorimetric assay on the sample which was gave a positive result for the presence of zinc. The team re-ran the original ITC experiment with the fragment hit in the presence of EDTA (which should chelate all zinc present), this showed that all binding was lost. In summary, the fragment hit must have contained Zn2+ ions which was the cause of the activity.

Gareth 1

Figure A showing ITC binding experiment of Ube2T with the original fragment compound with and without EDTA present. ZnCl2 against Ube2T is also shown as a comparison.

Taken from: F. E. Morreale et al., Mind the Metal: A Fragment Library-Derived Zinc Impurity Binds the E2 Ubiquitin-Conjugating Enzyme Ube2T and Induces Structural Rearrangements. Journal of Medicinal Chemistry60, 8183–8191 (2017).

The authors did point out that this hit was from a commercial library and one of the actions they undertook was to ask for the QC data from the supplier, at the point when the related analogues of the fragment compound did not show any activity. This obviously did not highlight the presence of the Zn2+

I did wonder when reading this publication if a different screening cascade may have identified this type of false positive, before having to get to the step of a very labour intensive co-crystallization process. For example a cell based assay might by more resilient to the presence of metal ions. However with a fragment based project, this may not be possible due to the poor affinity of fragments and also obtaining an effective concentration within the reduced DMSO tolerance of a cell based format compared to a biochemical assay. The only recommendation for this specific case, would be carry out further purification of samples, and then re-assay. It would also appear that the data from commercial suppliers is lacking and maybe that can be changed.

The authors should be commended for this work and putting it in the public domain. Drug discovery is a hard, long and quite often an unsuccessful process, and anything we can do to reduce the time following red herrings the better.

Blog written by Gareth Williams



Circular dichroism for protein characterisation

During my PhD project, I was working on a bacterial protein involved in iron transport. The secondary and tertiary information about this protein was limited and function was not well known at that time. In order to shed more light on this, I was looking for various techniques and came across ‘Circular Dichroism’ [CD], a widely used technique for analysing the secondary structure content of the protein of interest in solution. Circular dichroism measures the difference in the absorption of left and right circularly polarised light by chiral molecules. It is also a great technique to study the protein-ligand interactions.

The secondary structure composition is associated with characteristic spectra based on the absorption due to the peptide bond in the far ultraviolet [UV] CD region (180-260 nm) whereas tertiary structure features are influenced by the spectrum in the near UV region (260-320 nm) with absorption being due to aromatic residues [2]. In this category, synchrotron radiation circular dichroism spectroscopy [3] [SRCD] also became popular which provided more structural information with the available options for measuring the spectra to lower wavelengths, with improved signal to noise ratio levels.

I came across an article in Nature Scientific Reports indicating more advancement in this area by the introduction of High-throughput SRCD [HT-SRCD] using multi-well plates [1] which was interesting. The paper describes the features provided by Beamline B23 of Diamond Light Source, Oxford, UK for setting up the HT-SRCD using multi-well plates. It also describes examples for high-throughput measurements carried out using multi-well systems. In short, this HT-SRCD could be a potential resource for studying the protein folding, conformational changes in protein structure induced by ligands, buffers and other components as well as secondary structure determination on a high-throughput scale.

Blog written by Mohan Rajasekaran

  1. Hussain, R., Javorfi, T., Rudd, T. R., and Siligardi, G. (2016) High-throughput SRCD using multi-well plates and its applications. Scientific reports 6, 38028
  2. Kelly, S. M., Jess, T. J., and Price, N. C. (2005) How to study proteins by circular dichroism. Biochimica et Biophysica Acta (BBA) – Proteins and Proteomics 1751, 119-139
  3. Wallace, B. A., and Janes, R. W. (2010) Synchrotron radiation circular dichroism (SRCD) spectroscopy: an enhanced method for examining protein conformations and protein interactions. Biochemical Society transactions 38, 861-873





Citizen scientists

Tesh 1Many moons ago while I was an postgrad in Dundee, when a computer screen was the size of a medium to large beach ball, i.e. the late-90s, I came across a curious project over the then relatively nascent internet called SETI@home.

The project was conceived by the Berkley SETI Research Centre as an initiative to engage with the general public to help address a specific (and possibly the ultimate) question, “Is there anybody out there?”. The premise being that if there are other (extra-terrestrial) intelligent life forms out there in the universe, they may be broadcasting (knowingly or otherwise) their presence and thus should be detectable by simply listening in.

A vast amount of data was collected, consisting of recorded radio signals across a chunk of the spectrum, which then needed to be analysed. The scientific community could not fund/carry out the analysis and needed an “out of the box” solution. Step in SETI@home, where the idea was to make use of the processing power of the vast numbers of PCs that were starting to take up residence within the public domain. The strategy was not particularly interactive and solely relied in the ability of PCs to crunch through the data when not in use by their owner. All you had to do was register and packets of data were sent out to your PC (the University of Dundee’s in my case, ahem!) and your PC did the rest. There was something hypnotic, ethereal and satisfying as the screen presented how much data had been processed and how much more there was to go. Even as someone who only has a passing interest in space related matters, wondering if there were any “curious” signals in the data, how long it would take to get a positive answer and if enough bandwidth had been covered to account for a negative outcome, was exciting and alluring. Though it was not particularly interactive, it did capture the imagination of the wider public and made use of their individual resources to try and answer an important scientific question.

A decade or so later I came across another project ( that pricked my curiosity and again it was a space related project. The premise was as simple as it was unimaginable – spot planets as they traverses across the face of their host star(s) in the line of site from Earth from up to ~1500 light years away. Tesh 2

This however was made possible by the recently launched Kepler space telescope by NASA, which is a space based observatory with the sole purpose of peering at a patch of the sky continuously and recording (every 30 minutes) the light emitted by the stars within its field. Every 3 months or so the data would be downloaded by NASA and distributed to its collaborators within the scientific community for analysis. This involved dealing with an incredible volume of data, for which data processing and analysis pipelines had to be set up for.

Though these were in themselves incredibly successful, it was recognised fairly early on that engaging with the public in a meaningful way would be advantageous, not only to prevent public apathy (which seems to haunt many of these large projects) but also in recognising that any analysis pipeline set up would ultimately be limited.

Analysis of any large data sets requires assumptions, which are usually applicable across the majority of instances but not where there are deviations from the “expected scenario”. The ability of the human eye to perceive subtle differences and patterns was seen as an advantage here, and which, with the right approach could potentially be tapped. Step in (hosted by and citizen scientists.

At its simplest level (below), the platform asks the public if there is a periodic signal within the light curve of a particular star. However, the level of analysis could be more complex if desired, e.g. identifying the star type, accessing the unprocessed/raw data and links to information about a stars’ age, its metallicity, a deep visual look into its neighbourhood and so on. There were forums set up to discuss the data in general, reoccurring glitches in the data, individual stars, analysis pipelines for larger bespoke batch analyses and much more.

Tesh 3So did these citizen scientists find anything? Yes, being the clear answer. To date there have been 10 peer reviewed publications and more will no doubt follow


These types of endeavours, where the public can be engaged in a meaningful way to answer specific and scientifically inspiring questions, are important on a number of levels:

  1. Access to a “free” and potentially vast resource.
  2. There are important (and sometimes unexpected) discoveries to be made.
  3. It prevents public apathy.
  4. Exposure.
  5. Funding (e.g. from the greater exposure of the project)

The question arises that in this age of big data, particular with the explosion in cell biology and disease related big data projects, why do they not also have such endeavours? For such well-funded scientific areas, for there to be only one such endeavour on (Etch a Cell) is if nothing else sad. Etch a Cell is an initiative led by the Francis Crick Institute where the aim is to engage with the public to help build 3D models of the nuclear envelope from electron micrographs. This is of interest to me and (possibly some) other people in my field of research but it hardly captures the ziet geist as Planethunters and SETI@home did and continues to do so.

Blog written by Tesh Patel

From lab to launch….

Drug discovery is a time consuming and expensive undertaking. Currently it commonly takes between 10 to 15 years to get from the start of the discovery process through to launch (if you make it !) and the cost can range from $2 billion to $5 billion depending upon whose statistics you reference. The preclinical discovery phase tends to be short and relatively cost effective, the output of which then runs the gauntlet of a battery of toxicity and safety tests before being allowed to enter testing in humans. After what is usually a relatively short efficacy study in a small number of patients provides a suggestion that the drug may work, a series of larger and longer clinical studies ensure that the drug is both safe and effective. If all goes well in these studies the regulatory authorities, usually FDA or the EMEA, will review the extensive data package and if opinion is positive, give approval for the drug to be launched for use in routine clinical practice.

We all want this process to be faster and more effective but without any compromise on safety and determination of efficacy – and by ‘we’ I mean drug discovers, patients, the pharmaceutical industry and regulators. We also want the process ideally to be more efficient and predictive with less chance of failure, particularly in late stage, expensive studies (one of the major reasons the costs above are in the billions – if every drug that entered clinical studies worked the cost of discovering a new drug would be ~$350 million). A focus on orphan diseases where patients are more homogeneous and we have strong understanding of the genetic basis of their disease is delivering higher success rates. Perhaps the poster child for this approach has been cystic fibrosis and the introduction of therapies that effectively address the genetic defect by repairing the defective protein – in CF it’s the cystic fibrosis transmembrane conductance regulator (CFTR), a chloride ion channel. Vertex Pharmaceuticals introduced the first of its expanding portfolio of CFTR repair therapies in 2012 – this was the CFTR potentiator, ivacaftor (tradename Kalydeco). Ivacaftor demonstrated impressive clinical effects in patients with a specific CFTR mutation (G551D) and has demonstrated efficacy in a number of follow on trials in CF patients with mutations which are biophysically similar to G551D (a CFTR protein that makes it to the cell membrane but is loathe to open). G551D is the third most common CF disease causing mutation that accounts for somewhere between 2 – 5% of the CF population so relatively rare as there are estimated to be ~70,000 patients worldwide.

So what happens if you have a medicine which you believe will deliver benefit to additional patients but they are few and far between, with not enough to undertake a robust phase 3 trial ? This was the conundrum facing Vertex when looking to expand the labelling for Ivacaftor. In what is the first of its kind the FDA granted expanded approval to Vertex for Ivacaftor based upon in vitro data only. This could be a landmark step and the FDA has acknowledged that this approach could have implication for other drugs that have a well understood safety profile and address well characterised diseases. With Ivacaftor Vertex have a drug with a robust safety package, a strong understanding of its mechanism of action and have put considerable effort into assessing the correlation between preclinical cellular assays, clinical biomarkers and registerable endpoints. To support the request for expanded labelling Vertex expressed ~50 mutations in Fisher rat thyroid cells, a cell system widely used by the CF field as it has low expression of background chloride channels and can be used in a variety of assays (including Ussing chamber ion transport). Mutations that delivered a 10% increase in chloride transport when treated with Ivacaftor were considered responsive. This wasn’t an arbitrarily selected figure but one borne out by Vertex’s clinical experience with Ivacaftor and other compounds from their developing CFTR repair portfolio. Of those tested 23 mutations have been added to Ivacaftors labelling (26 failed to meet the criteria).

In real terms this means that ~900 CF patients in the US alone will now have the opportunity to access this breakthrough medicine – my congratulations to Vertex for pioneering the approach and my congratulations to the FDA for entertaining it….let’s hope it can be pursued for many other diseases.

Martin 1

Image source:

Blog written by Martin Gosling


Durmowicz A.G et al (2018) The FDA’s experience with Icavaftor in cystic fibrosis: establishing efficacy using in vitro data in lieu of a clinical trial. Ann Am Thorac Soc. 2018 Jan;15(1):1-2. doi: 10.1513/AnnalsATS.201708-668PS.

Kingwell K (2017) FDA Oks first in vitro route to expanded approval. Nature Reviews Drug Discovery; doi:10.1038/nrd.2017.140

AI – a cure for the ROI?

Marcus 1

The face of the Pharmaceutical industry has changed beyond recognition over the past 20 years with many of the major players passing through multiple rounds of M&A, calving off large swathes of their portfolios, synergising, repurposing drugs ………all ultimately to improve Return On Investment (ROI). It is no secret the sector has had a rough ride with blockbuster drugs becoming increasingly rare, only ≈10% of drug candidates in phase I reaching approval in the years 2006-2015 (1) , increased payer pressure to cut prices, company revenues taking a hit as patent cliffs pass by and a dearth of innovative medicines being brought to the market.

How long will companies be able to sustain the significant cost of R&D whilst still turning enough of a profit to satisfy shareholders? The morality of industrial drug discovery (DD), long questioned in any case by those outside the industry, will be under serious scrutiny – not hard to see why with companies like Gilead charging upwards of $80,000 for a full course of the hepatitis C drug Sovaldi (2).  Take Ebola for instance.  Prior to the 2014 outbreak in West Africa (3) Ebola may have been viewed as an African problem and perhaps not an attractive area for investment.  After the outbreak and associated hysteria in certain corners of Western society all of a sudden this had the potential to be a little more than just an African problem.  Ebola was discovered in 1976 yet a study by the University of Edinburgh in 2015 estimated around half of all funding for Ebola research occurred in the 2014-2015 period after the outbreak.

Even the philanthropy of the Wellcome Trust is driven by ROI which unlike VC funding, is perhaps based more on intellectual than financial value, but here too we see a move away from traditional DD as demonstrated by the demise of the Seeding Drug Discovery Initiative. Granted this change in focus may be designed to generate new targets or technologies but the sentiment is clear; as traditional DD becomes more difficult with target patient populations potentially dwindling as a result of increased personalised/specialised therapies or peripheral areas of unmet need, where is the motivation for investment?

We are in desperate need of new anti-bacterials, as the population becomes older the prevalence of dementia is on the rise and alongside the ever present spectre of cancer all represent substantial investment if we are to have a meaningful impact in the development of effective treatments for these indications. It is clear that with the patient heterogeneity and the variable aetiology of these conditions that the model of drug discovery to date needs a significant change in prosecution.  In an effort to speed up the DD process we have seen a recent spate of AI-Large pharma collaborations (4/5/6) but are these tools merely ways to speed up old methods or will they genuinely result in the generating novel targets which might otherwise remain undiscovered by conventional means of investigation?

Regression analysis such as QSAR has long been used within DD to correlate physiochemical and functional parameters to guide chemical synthesis programs. Correlations between various ‘attributes’ (derived from principal component analysis) are used to generate a model which can be used to predict the behaviour of new compounds.  Of course the model is only as good as the number of data points and parameters used to define it and, once defined, remains unchanged.  The essential difference is that AI (specifically Deep learning) generates self-adapting models using a multi layered network approach that wasn’t really possible before the development of GPUs which allowed the parallel processing of vast amounts of data.  The data is assessed according to any one of its ‘attributes’ in the ‘top layer’ before being passed to the ‘second’ layer and processed according to another attribute etc. and as this is an unguided approach the systems needs a sufficient volume of data and many iterations to generate a reliable model of correlation.  After each iteration the algorithms used to generate the correlations can then be altered as the network ‘learns’ from the previous iteration.  Like any other computer model generating system though, it is liable to the ‘garbage in, garbage out’ (GIGO) concept.

It is easy to see how AI can be and is effective in, for instance, developing candidate compound libraries generated from well-characterised protein and protein/ligand crystal structures and suggest routes of synthesis (7) but the question of how AI can truly revolutionise an ailing industry is a long way off being answered. The regression analysis used in AI is the same as that used in QSAR for years, essentially it’s just the volume of data and the learning aspects that are different.  The hope, however, is that AI will generate unique correlations that have thus far eluded us or only revealed themselves serendipitously.  Pfizer’s hopes to quickly analyse and test hypotheses from “massive volumes of disparate data sources” (including more than 30 million sources of laboratory and data reports as well as medical literature)(8) seems, to the untrained eye (mine!) to be fraught with danger regarding curation of the input data.  Even in the simplest instance of a standard compound IC50, how would un-curated inter-institution variations affect a blind, self-determining analysis?  Perhaps, conversely to the GIGO scenario, considering the volume and disparate nature of data used (literature, experimental, predicted) and correct application of principal component analysis in a given enquiry, AI may actually be resilient to these small variations.

With regard to mental health we only have to look at our efforts to provide an objective definition of a subjective experience in the reconceptualization/re-categorisation/inclusion/elimination of mental disorders in subsequent editions of the ‘Diagnostic and Statistical Manual of Mental Disorders’ (9) to know that our understanding of these disorders is in a constant state of refinement.  AI assessment of a potentially novel pathway/target based on the prevailing definitions of a given condition superposed on the inherently variable subjective clinical data would, it seems, yield different answers from one year to the next.

AI has been hampered not necessarily by the development of algorithms but by the availability of sufficiently broad, curated training data sets and the development of both GPUs and adequate storage (10). With the advent of ‘-omics’ technologies able to acquire vast amounts of data, only relatively recently have the means been developed by which we can effectively interrogate this huge repository of information.  It would seem then that a standardised curation of this data is of primary importance if the industry is going to rely heavily on AI to effectively generate new medicines……notwithstanding the importance of generating clinically verified biomarkers in parallel…..but that’s for another blog!!

We only have to look at GenBank as an example. From its inception in 1982 it took until 2003 for the first release of the curated RefSeq collection.  I remember trying to identify novel splice variants in the late 90’s only to be frustrated by poorly annotated and simply incorrect sequences.  Contemporary parallels can be drawn with the Protein Data Bank (PDB) especially in relation to a) the Structural genomics program where un-curated, non-peer reviewed, homology based structures are being submitted to the database (11), and b) inaccurate Protein-Ligand co-crystal structures (12).

It is clear that AI can/will be/is a benefit during every step of drug discovery and that algorithm refinement is an ongoing, iterative process but what is not currently clear is whether AI will deliver where, for instance, HTS failed in dramatically impacting on the inefficiencies of the DD process (13). I have no doubt that very soon AI will become a fundamental part of all aspects of health care and drug discovery but I wonder whether this will actually precede the demise in scale of small molecule drug design and highlight the need to pursue other avenues (e.g. Gene therapy/Biologics) more vigorously.   In any case as the complexity of both the diseases/unmet need and the required solutions increase it will be interesting to see how ROI will be maintained and how much more Big Pharma consolidation we will see over the coming years.

Blog written by Marcus Hanley


  3. Fitchett et al. (2016) Ebola research funding: a systematic analysis 1997–2015, Journal of Global Health. Available from:
  9. The DSM-5: Classification and criteria changes. World Psychiatry. 2013 Jun; 12(2): 92–98. Published online 2013 Jun 4.
  11. Domagalski et al. (2014) The Quality and Validation of Structures from Structural Genomics MinorMethods Mol Biol. 2014 ; 1091: 297–314
  12. Reynolds, Charles H (2014) Protein-ligand cocrystal structures: we can do better. ACS medicinal chemistry letters, 10 July 2014, Vol.5(7), pp.727-9





Using Biochemical Light-switches to Illuminate Ion Channel Activity

There’s no doubting that optogenetics is an important recent development within the field of neuroscience. Channelrhodopsins, non-mammalian proteins that conduct ions in response to specific wavelengths of light, have now been inserted into various neuronal pathways to demonstrate the use of light to control modalities as diverse as vision, hearing, pain and motor control. In November 2017, The Scientist reported on progress in human clinical trials using channelrhodopsins in combination with viral vectors to restore a degree of function in damaged sensory neurons in response to light1. In a study conducted by Allergan, patients suffering from retinitis pigmentosa were injected with virus carrying the genetic signal to express channelrhodopsins specifically in retinal ganglion cells, bypassing the damaged light-detecting cells of the retina, enabling a rudimentary sense of light-detection in patients that were previously totally blind. Although primarily a safety study, it has shown promising progress in the field which may soon see developments towards treating hearing loss and chronic pain using a similar approach.

However, alongside the use of non-mammalian light-activated proteins to control neuronal activity, an alternative light-based approach has been developing which has direct and immediate usefulness as a tool in the field of drug discovery. The use of light to reversibly deliver ligand to a native protein receptor or ion channel, or ‘optopharmacology’, is the subject of an interesting recent mini-review by Bregestovski, Maleeva and Gorostiza2. For drug discovery, the use of these chemical photo-switches enables the rapid, and most importantly, reversible activation of ion channel function in response to light. Ligand-gated channels are the subject of this particular review, but much work with voltage-gated channels and G-protein-coupled receptors has now also been published. This approach provides a huge step forward from the previous use of caged-compounds in flash photolysis – often used, for example, to study synaptic transmission, but would leave the preparation awash with ligand that was slow to clear by re-uptake/diffusion, meaning difficult and slow ‘one-hit’ experimentation.

At the Ion Channel Modulation Symposium in Cambridge last year (2016), Dirk Trauner spoke about the development of these tools and demonstrated their use in research conducted both within his group and in conjunction with collaborators around the world, showing examples (see figures 1 & 2 below) of the use of photochromic ligands in both their soluble (PCL) and tethered (PTL) forms.

These two forms are defined by the nature of their interaction with target proteins – PCLs are designed to mimic the ligand of a specific receptor but are freely diffusible and may not exhibit sub-type specificity within a tissue. PTLs, as their named suggests, become covalently tethered to their target, usually via naturally-occurring or genetically modified cysteine residues, conferring a much higher degree of selectivity. On exposure to specific wavelengths of light, the molecules photoisomerize between cis-and trans- states, enabling ligand-receptor interaction that triggers either activation or deactivation of the target protein.

Of the two types, PCLs are the simplest to use. Figure 1 shows the impressive degree of temporal control of gained over the function of the capsaicin receptor TRPV1 in the presence of 1uM of a ‘photolipid’ PCL (here, an azobenzene combining the vanilloid head-group of capsaicin with photoswitch-containing fatty-acid chain (AZCA derivatives)) in response to simply varying light stimulation between 350 and 450 nm.

Figure 1: reproduced from Frank et al (2015)3

Sarah 1

Having proved the principle in TRPV1-expressing HEK cells, optical switching in the presence of this PCL was applied to isolated mouse DRG neurones, and c-fibre nociceptive neurones in saphenous nerve preparations, both of which contain native TRPV1 receptors. All showed rapidly reversible non-selective cation conductance in response to the shorter wavelengths of light, translating into nerve depolarisation and action potential firing in native neurones – responses which were absent in TRPV1-/- knockout mice.

Trauner also presented work using PTLs, and show-cased developments made in collaboration with US colleagues to extend the chemical tether to reinforce its chemical stability and reduce the chance of off-target attachment. Their new ‘PORTL’ (photoswitchable orthogonal remotely tethered ligand), shown in figure 2a, was used to demonstrate dual optical control of mGluR2 and GluK2 expressed in the same cell, differentially responding to the specified wavelengths of light (figure 2b).

Figure 2: reproduced from Broichhagen et al (2015)4                 Sarah 2

This elegant chemistry is a powerful tool for studying ion channel-mediated physiology. Its use, for example, to selectively activate or silence particular neurones, or sub-populations of heteromeric channels containing a common tagged subunit, with a degree of spacial and temporal control unachievable with perfusion, enables more qualitative assessment of the interaction between  new possible therapeutic compounds and their target proteins. And like channelrhodopsins, the use of these photochromic ligands as therapies in their own right is also possible and currently being investigated.

Blog by: Sarah Lilley


1 The Scientist Nov 16 2017 article by Shawna Williams

2Bregestovski, P., Maleeva, G., and Gorostiza, P. (2017) Light-induced regulation of ligand-gated channel activity. British J Pharm, doi: 10.1111/bph.14022.

3Frank, JA., Moroni, M., Moshourab, R., Sumser, M., Lewin, GR., and Trauner, D. (2015) Photoswitchable fatty acids enable optical control of TRPV1. Nature Comms 6, doi:10.1038/ncomms8118

 4Broichhagen, J., Damijonaitis, A., Levitz, J., Sokol, KR., Leippe, P., Konrad, D., Isacoff, EY., and Trauner, D. (2015) Orthogonal optical control of a G protein-coupled receptor with a SNAP-tethered photochromic ligand. ACS Cent Sci 1 383-393, doi: 10.1021/acscentsci.5b00260