Toward a future of truly personalized cancer therapy

Targeted cancer therapies have produced substantial clinical results in the last decade, but most tumours eventually develop resistance to these drugs.

Benes, Engelman and colleagues have recently described in Science the development of a pharmacogenomic platform that facilitates rapid discovery of new drug combinations starting from human cancer samples.

This new approach could in the future help direct therapeutic choices for individual patients.

The study focused on non-small-cell lung cancers (NSCLCs) with activating mutations in epidermal growth factor receptor (EGFR) or anaplastic lymphoma kinase (ALK). This tumours are now routinely treated with specific tyrosine kinase inhibitors (TKIs), although the tumours eventually develops resistance within 1–2 years, through 2 main mechanisms:

gatekeeper’ mutations, which prevent target inhibition by the TKI;

bypass track’ mutations, which activate compensa­tory signalling pathways.

The authors generated cell lines directly from tumour biopsy samples using recent advances in cell culture methods. The cells were then subjected to a screen that combined the original TKI (against which cells had become resistant) with a panel of 76 drugs targeted at key regulators of cell proliferation and survival (Fig 1).


Fig 1. Schematic of the screen workflow

The pharmacological screen identified multiple effective drug combinations and a number of previously undescribed mechanisms of resistance. For example, a cell line derived from an ALK-mutated cancer was resensitised to ALK inhibitors when these were combined with a MET tyrosine kinase inhibitor. Furthermore, the screen identified resistance mecha­nisms that would have been difficult to discover by genetic analysis alone — for example, it was found that ALK-mutated NSCLCs often exhibit upregulated SRC tyrosine kinases signalling, without any evidence of mutations in SRC.

Several combination therapies identified in vitro were subsequently tested in xenograft models using the same cells and shown to be effective, indicating that the screen may indeed be predictive for in vivo activity.

Both the success rate of cell-line genera­tion from biopsy specimens (50% in this study) and the time scale for establishing cell lines (2–6 months) will need to be improved for this approach to become clinically useful. However, once optimized, it may be used not only for NSCLC but also for other types of cancer, allowing truly personalized cancer therapy.

Original paper:  Crystal, A. S. et al. Patient-derived models of acquired resistance can identify effective drug combinations for cancer. Science, 2014, 1480-1486.


Not all LogP’s are calculated equal: CLogP and other short stories

The partition coefficient (logP) of a material defines the ratio of its solubility in two immiscible solvents – although we normally use octanol : water, it could be any combination of immiscible fluids. This property is one of those chemical descriptors that pervades all aspects of ADMET and is used to filter out and define chemical space in which to work. Oddly, for such an important property, most projects and programs are built upon materials where the LogP has never been experimentally determined: relying on predicted values generated by software.

Recently, our DMPK scientist presented a series of predicted logP values vs some that he expertly determined in the lab. Whilst the correlation was good in many cases, there were some significant outliers, so he came to ask me, the computational chemist, to see if I might explain why the calculated logP was so different. There were some obvious structural features that can beguile certain methods of calculating logP – yes, there is more than one method of calculating logP – and other methods might closer predict the outlier values in our case.

Not All LogPs are Calculated Equal

When chemists talk about ClogP they are usually erroneously referring to “calculated” logP. To a CADD scientist, ClogP means something different – ClogP is a proprietary method (owned by BioByte Corp. / Pomona College) used to predict logP. Whilst there are a range of methods for prediction, there are three basic groups, and the vast majority of the current methods are flavours thereof:

Atomic (e.g. “AlogP”, ) & Enhanced Atomic / Hybrid (“XlogP”, “SlogP”)

Fragment / Compound (“ClogP”, KlogP, ACD/logP)

Property based methods (“MlogP”, “VlogP”, “MClogP”, “TlogP”)

Atomic logP considers that each atom has a contribution to the logP, and that the chemical entity’s final value is purely additive. Crippen et al. first proposed such a method in a series of papers in the late 80’s, with the refined version dubbed “AlogP”.1 The method is effectively a table look-up per atom, and there are plenty of free AlogP calculators available. It is suited to smaller molecules, particularly those with non-complex aromaticity or those which do not contain electronic systems that are known to have unexpected contributions to logP.

Enhanced Atomic or hybrid logP (XlogP, SlogP etc.) is a modification of the AlogP system – to try and address the shortcomings of atomistic approaches to larger systems, it takes the value of each atom type, as well as a contribution from its neighbours, as well as correction factors which help sidestep known deviances in purely atomistic methods.  This is an attempt to allow for larger electronic effects. It is fast, being a table look-up technique, and many free software use this too. The smarter hybrid algorithms know the state of each atom and thus how much of a contribution its neighbours add.

Fragment / Compound logP is a method that uses a dataset from full compounds, or fragments, which are experimentally determined, and then modelled using QSPR or other regression techniques in small fragments rather than per atom. Fragment contributions are then added up, with correction factors. The rationale here is that sometimes atomistic approaches do not adequately model the nuances of electronic or intramolecular interactions, which may be better modelled by using whole fragments. This method tends to be better for systems with complex aromaticity, and larger molecules – on the condition that the molecule contains features that are similar to those from which the modelling was conducted. In the case of very obscure motifs in your molecules, then the model from which the prediction is made may not have a very good correlation.


Property based methods…
There are a whole host of methods for determining logP using properties, empirical approaches, 3-D structures (e.g. continuum solvation models, MD models, Molecular Lipophilicity potential etc…), and topological approaches. Most of these methods are reasonably computationally intense, and are buried in the world of informatics and stats, but one is worthy or particular note: Moriguchi’s method (or MlogP), which used the sum of liphophilic atoms, and sum of hydrophilic atoms as the two basic descriptors in a regression model that was able to explain nearly 75% of variance in experimentally determined LogP values of a dataset of 1230 compounds.2 The group later added 11 correction factors, and the model explained 91 % of variance. It is very fast, and so historically it was employed for large datasets, and was included in several property prediction software, such as Dragon, and ADMET Predictor (Simulations Plus, Inc.). Nowadays as computational speed has increased, MlogP is used less, as more accurate methods become manageable, even at large library sizes.

So, which method do you use?

Biovia’s Pipeline Pilot, and Discovery Studio sport a version of AlogP, and Knime has multiple free X and A logP calculator plug-ins. CCG’s MOE uses both an unpublished atomic model (Labute) and a hybrid SlogP. DataWarrior uses ClogP, Dotmatics / Vortex natively use XlogP, but you can patch in others. Cresset BMD’s offerings use SlogP and Optibrium’s StarDrop uses a fragment method. ChemAxon uses multiple methods (including hybrid (VG) and fragment e.g. KlogP), and if you have their InfoCom nodes in Knime, then you can use multiple methods and weight them according to your understanding, or better yet, you can do a quick correlation check across the methods with known data in your series (if your group has the resource to experimentally determine a few of your own LogPs), and then weight your model accordingly.

As a rule (to which there are exceptions):

Simple small molecules (e.g. fragment sized) – AlogP will probably perform just fine, but a hybrid method would be better.
Complex but standard small molecules (the normal development type med chemists love), then  fragment / compound logP methods will often be the most accurate. Hybrid methods are your second best option (but still reasonably good).

Complex, non-standard molecules (with rare motifs), then a hybrid system or fragment-based logP may be equally good (or bad), it depends on the model on which the fragment logP is based. You could also get your team to determine some experimentally and see if you can’t build yourself a model…

For statistical insight into many state-of-the-art and classical methods, and how well they perform across large experimentally determined sets, see Mannhold et al.’s thorough review.3

So, to conclude, not all logP prediction models are built equal and there will be times when some models exceed others in accuracy, depending on your chemistry. Hopefully now you’ll at least be able to explain in your group meetings why your predicted logPs were way off…


  • Ghose, A.K.; Crippen, G.M. Atomic physicochemical parameters for three-dimensional-Structure directed quantitative structure-activity relationships. 2. Modeling dispersive and hydrophobic interactions. J. Chem. Inf. Comput. Sci. 1987, 27, 21–35.
  • Moriguchi, L.; Hirono, S.; Liu, Q.; Nakagome, I.; Matsushita, Y. Simple method of calculating octanol/water partition coefficient. Chem. Pharm. Bull. 1992, 40, 127–130.
  • Mannhold, M. et al. Calculation of Molecular Lipophilicity: State-of-the-Art and Comparison of Log P Methods on more than 96,000 compounds. J. Pharm. Sci. 2009, 98, 861-893.

Peptide therapeutics: current status and future directions

By Keld Fosgerau and Torsten Hoffmann (Drug Discovery Today ahead of print)

In recent years peptide therapeutics have emerged as a novel therapeutics due to good safety, tolerability and efficacy. Also the production cost of peptides is lower than other technologies, like small molecules or protein production. However, there are some pitfalls in peptide therapeutics such as reduced choice of route of administration, poor high-life in plasma or higher degradation rate by oral route. Consequently, there is an important focus on new approaches to improve the use of peptides in pharmaceutical research. In this paper the authors give a summary of the most important new approaches and developments in peptide therapeutics.
In the last ten years, some companies have successfully launched peptides as new drugs, such as Lupron™ from Abbot Laboratories (prostate cancer treatment), gaining worldwide sales of more than US$2.3 billion in 2011. Sanofi with the drug Lantus™ (Diabetes treatment) reached US$7.9 billion in 2013. By last year the FDA had approved more than 60 peptides drugs on the market and currently there are more than one hundred peptides drugs currently in clinical trials, predicted to reach sales of US$25.4 billion in 2018. The main disease areas in which peptides are used are metabolic diseases and oncology.
In order to obtain better peptide therapeutics it is necessary to make a rational peptide drug design to improve the physicochemical properties of natural peptides. The strategies include substitution of amino acids (alanine scan), small focused libraries and structure-activity relation to identify essential amino acids and possible substitution (Figure 1).

In recent years peptide therapeutics have emerged as a novel therapeutics due to good safety, tolerability and efficacy. Also the production cost of peptides is lower than other technologies, like small molecules or protein production. However, there are some pitfalls in peptide therapeutics such as reduced choice of route of administration, poor high-life in plasma or higher degradation rate by oral route. Consequently, there is an important focus on new approaches to improve the use of peptides in pharmaceutical research. In this paper the authors give a summary of the most important new approaches and developments in peptide therapeutics.
In the last ten years, some companies have successfully launched peptides as new drugs, such as Lupron™ from Abbot Laboratories (prostate cancer treatment), gaining worldwide sales of more than US$2.3 billion in 2011. Sanofi with the drug Lantus™ (Diabetes treatment) reached US$7.9 billion in 2013. By last year the FDA had approved more than 60 peptides drugs on the market and currently there are more than one hundred peptides drugs currently in clinical trials, predicted to reach sales of US$25.4 billion in 2018. The main disease areas in which peptides are used are metabolic diseases and oncology.
In order to obtain better peptide therapeutics it is necessary to make a rational peptide drug design to improve the physicochemical properties of natural peptides. The strategies include substitution of amino acids (alanine scan), small focused libraries and structure-activity relation to identify essential amino acids and possible substitution (Figure 1).


Figure 1

Figure 2 highlights the advantages and issues associated with peptide based drug discovery


Figure 2

The new approaches currently used in development for peptides are:
Multifunctional and cell penetrating peptides
Peptide drug conjugates
Alternative routes of administration
The majority of current peptide drugs are injectables with a lesser number of oral drugs, such as cyclosporine (Neoral™) and desmopressin (Minirin™). The development of oral delivery is expected to increase due to it being more suitable for the patient. However, one of the difficulties of oral peptides is acidic/enzymatic degradation by the gastrointestinal tract and intestinal mucosa. New approaches for oral peptides to avoid degradation includes; stabilizing the secondary structures by stapled peptides, hydrophobic faces, cyclization, N-methylation, and intramolecular bonds. Some companies working in this research are; Ra Pharma, Peptidream, Cyclogenix and Bycycle Therapeutics.
Current multifunctional peptides are antimicrobial peptide drugs that have additional functions like immune stimulation and wound healing. The most focused research in multifunctional peptides is the glucagon-like peptide-1 (GLP-1) agonist with several currently in clinical trials.
Another major strategy for novel peptide therapeutics in the field of oncology, is the conjugation of peptides with a higher efficacy and safety properties. Nowadays there are more than 20 peptide conjugates in clinical trials.
Some examples of modifying peptides using a rational design to help improve the physicochemical properties are; the introduction of stabilizing α-helixes, salt bridge formations or other chemical modifications, such as lactam bridges. It is really important to avoid the enzymatic degradation of peptides due to the short half-life in plasma. Some important biotechnologies companies (Pepscan and Aileron Therapeutics) are working towards protecting peptides against enzymatic cleavage by different approaches, such as insertion of a structure inducing probe (SIP)-tail, lactam bridges, stapling/cipling peptide sequences or by cyclization.
One common strategy for extending half-life of peptides is binding them to albumin as a vehicle, to obtain a peptide requiring less frequent administration. Another approach for increasing plasma half-life is using Polyethylene glycol (PEG)-ylation to limit globular filtration and reducing the elimination of peptides; however, PEG had some concerns related with safety and tolerability. A new “smart” formulation technology by Intarcia is an implantable device that delivers peptide from a dry reservoir using an osmotic pump system. The last example gives a new and innovative way of delivering peptide therapeutics in a better, easy and comfortable method for the patient.
With the potential sale market, excellent safety, tolerability and efficacy in humans peptide therapeutics is an attractive approach for drug research and development.

Highlights from Society of Medicines Research Meeting 4th December 2014

The Society for Medicines Research was hosting the “Recent Disclosures of Clinical Candidates” meeting on Thursday 4th December. The talks covered an array of novel molecular therapeutics across several target classes and therapeutic areas, including oncology (inhibitors of BTK and Mdm2), pain (Nav1.7 modulation), CNS (highly selective M1 agonists and BACE1 inhibitors), bacterial infection (inhibitors of C. diff.) and allergic inflammation (intranasal TLR7 agonists). Below are a couple of highlights from the meeting.

Keith Biggadike (GSK) gave a talk on the discovery and early discovery of an intranasal TLR7 agonist.

GSK2245035 is a novel small molecule TLR7 agonist in development as an immunomodulatory, intranasal treatment for allergic airways disease aiming to reduce Th2 and enhance Th1/Treg responses to aeroallergens via the local induction of type1 interferons (IFN). Although the structure was not disclosed, a series of close analogues were described.

GSK2245035 was reported as a highly potent TLR7 agonist inducing IFNa production in human in vitro assays (pEC50 blood 9.5) whilst showing selectivity over the pro-inflammatory cytokine TNFa  (pEC50 blood 5.0). The compound was also reported to be selective over the other Toll-Like Receptors (and in particular 3, 8 & 9), had low oral bioavailability and high clearance.


1% oral bioavailability, T1/2 (IV) 1.4h, Vss 2.7l/kg, PPB 73%

 pEC50 blood 9.5 IFNa, pEC50 blood 5.0 TFNa

GSK2245035 showed comparable IFNa potency and IFNa/TNFa selectivity in Cynomolgus monkey and human blood cultures. Biomarkers of target engagement with IP-10 in Cynomolgus monkey were identified to support the dose selection in clinical studies.

Ian Storer (Pfizer Neusentis) made the first disclosure of one of Pfizer’s NaV1.7 modulator for the treatment of pain. The channel subtype NaV 1.7 has been shown to be essential for the normal sensation of pain. Ian’s presentation covered the identification and optimisation of an acyl sulphonamide series as the first generation of Nav1.7

The initial selectivity of the lead compound PF-5241328 came from an initial file screening and believed to bind to a novel voltage sensor domain. However, the initial series suffered from Cyp inhibition and poor PK. Further SAR optimisation led to acidic isosteres in the form of acyl sulphonamides


The selected compounds showed good in vitro ADME but had an invitro/invivo mismatch due to active transporters linked to the acidic series. High dose in man was also predicted due to the low clearance (1450 to 3150 ml/min/kg). Overall, 4 compounds were advanced to human microdose studies and showed poor PK which led to the series to be halted. Further disclosures on the follow up series are due to be presented mid 2015.

Sharing Libraries, not Structures

This article refers to the paper:
“Sharing Chemical Relationships does not Reveal Structures”
M. Matlock and S. J. Swamidass, J Chem. Inf. Model. 2014, 54, 37-48.


Sharing data, projects and programmes between organisations is becoming more and more common for some companies looking to find more value in expert-centric models. There are stumbling blocks in terms of trust and sharing, however, and a prime example of this is the unwillingness of some organisations to share structural information in proprietary screening libraries at the very early stages of collaboration. Blind screening, where a company will screen your target and then report back hits under agreement, or where they send you a blinded diverse representation of their library on plates, which – if there are any hits – you feed back to them to pick out similar materials, is gaining pace as a way of accessing large, well curated libraries for screening, and likewise for the library owners; new targets.

The problem of sharing data in this way is that project leads and chemists would like to be able to find similar materials within the library themselves to help prioritise initial hits, and also reduce the time delays involved with going back to the library provider and requesting additional materials. As a result, there have been several papers attempting to address how to “blind” structures but still leave in information that will help project leaders pick follow-up materials without sharing structural information. One such paper is that of Joshua Swamidass and Matthew Matlock, (Loc. cit.), which details an interesting way of blinding a library whilst also empowering project decision makers to pick similar materials for hits using relationship metrics without relying on the external library providers.
Imagine that for a given screen a company gives you a blinded plate of materials as a diversity sub-set of its library. You screen these and some of the wells are hits. If, with the plate, you were given data sheet which showed for each well the serial number of similar materials (along with the similarity metric) within the library that the company had given you, you could then pick out those that you wished to follow up – saving the company time, but also allowing you to weight your selection (e.g. Hit One appears far more potent than Hit Two… so can we have all the similars to Hit One, and a couple of the similars to Hit Two).
Given that most chemoinformatics systems used for picking follow-up hits convert structural data to relationship data (e.g. similarities) in order to pick the next round of materials, this kind of information is very useful, even when structurally blinded.

Swamidass and Matlock detail several approaches that attempt to allow secure transmission of chemical relationships, such as Similarity Neighbours, Scaffold trees and Networks (allowing for sub-structural similarity distances and not just have/have-not functionality metrics), and R-Group networks (see figure below). They assess the information density and how secure each method is to prevent the reverse engineering of these data to provide structural insight.


They conclude that it is possible to communicate useful chemical information without sending structural data, and that similarity data is in fact one of the key data that is used to help select follow up materials. The authors also conclude that materials with very simple structures, and high symmetry are more vulnerable from reverse engineering from some relational data systems than more complex molecular motifs, but sharing relationship data dispels much of the insecurity of sharing structural descriptors.

This paper, and those similar, re-ignite the debate on making curated chemical libraries items to share and collaborate with, rather than shield and hide. Sharing libraries in this way also enables project leads to feel more empowered and informed when screening external libraries. Though clearly in its infancy, deployment of such secure library systems could open doors to easier and faster collaborative efforts between organisations, which clearly has benefits across many chemical domains.

In Silico Prediction of Off Target Activities of Chemical Probes


Chemical probes are frequently described as important tools in the validation of new drug targets, yet many of them are of little value due a number of important factors frequently including poor selectivity. This issue was highlighted by Frye in 2010 (Frye, Nature Chemical Biology 2010, 6(3), 159-161) where he illustrated the problem with the case of staurosporine, a promiscuous kinase inhibitor and the subject of 8,000 publications, many of which make conclusions which are not valid due to the compounds promiscuity. Frye defines a set of criteria which a probe should have in order for it to be a valuable tool to the drug discovery scientist. One key criterion is selectivity and he states that it is important that ‘a quality chemical probe has sufficient in vitro potency and selectivity data to confidently associate its in vitro profile to its cellular or in vivo profile,’ Which is valid suggestion.
Determination of selectivity is difficult as the majority of groups including pharmaceutical companies only have access to a limited range of selectivity assays and establishing selectivity against targets related to the protein of interest is relatively straightforward; however assessing broader selectivity is an issue both with respect to choosing targets to screen against and accessing suitable assays.
A recent publication highlights the importance of assessing broad selectivity and describes an in silico approach capable of predicting off target liabilities (Antolín and Mestres, ACS Chemical Biology 2014, ahead of print). The authors apply their software to the full set of chemical probes from the NIH Molecular Libraries Program (MLP) with some very interesting findings, illustrated by the examples below.
Table 1 shows the activity at the canonical (original target) of four probes in the collection together with off target activity identified through in silico testing and subsequently demonstrated in a biological assay. In each case the probes are more potent at the newly discovered target than the original published target.


ML006 was published as a weak probe for the S1P3 shingosine receptor but has micromolar potency for mTOR kinase. ML124 and ML204 were originally identified as probes for TRP channels, TRPML3/2 and TRPC4/5 respectively but show off target activity as acetylcholinesterase inhibitors (ML204) and sigma receptor ligands (ML124). ML141 was identified as a highly potent carbonic anhydrase inhibitor having originally been described as a selective inhibitor of Cdc42 GTPase. The latter discovery regarding ML141 should possibly have been identified earlier as the template in question is a well documented carbonic anhydrase inhibitor chemotype (Weber et al J. Med. Chem. 2004, 47(3),550-7). These results thus pose significant questions as to the usefulness of this set of compounds as probes.
The publication from Antolin et al not only highlights the caution that workers need to exercise when using chemical probes but also offer a potential strategy for checking the validity of the probe before drawing conclusions regarding experiments in which the probes have been used.

Compound contaminants; a story of false positives

Have you ever had that feeling your assays are picking up multiple false positives, spikey SAR repeating itself through the screening cascade. Well it may not be the assays causing the problems. There has been some discussion recently about inorganic contaminants in screening decks causing false positives. Last year a group from Roche published a paper (1) ( from their experience that shines a little more light on this problem.
This group were running a project for Pad4 and their initial screen (a high throughput enzyme assay) produced a number of hit series. These were confirmed in the conformational screening using an ELISA (enzyme-linked immunosorbant assay) and binding was also demonstrated in a ForteBio and Biacore assay with IC50 and Kd values being in the low µM range. The trouble was all three series ‘lacked conclusive SAR’ and upon re-synthesis all displayed varying activities depending upon batch. The group then did a little detective work and found that the different routes of synthesis had a strong impact on the resulting compound potency. In short those batches that used zinc in their synthesis were positive in the downstream assays, those that didn’t were negative. This hypothesis was tested by screening ZnCl2, which was found to have an IC50 of 1 µM and a Kd of 1 µM.
There are regularly false positives in screening cascades, but these are commonly identified in orthogonal assays. What is interesting in this case is the hits were positive in three separate screens, demonstrating simply screening compounds in multiple assays is no guarantee of identifying false positives due to contamination.
To see how common zinc contamination was as a source of false positives the Roche group looked back at 175 past HTS campaigns. 41 showed a high hit rate of zinc-probing compounds (>25%), with the expected hit rate being <0.01%, demonstrating the extent of the problem. The important thing to note is they were only looking at zinc contamination; it is likely the total hit-rate for other inorganic containing compounds would be much greater.
The highest hit rate observed was in a fragment screen, which was performed at 250 µM. The group postulated since fragment screens are performed at higher concentrations they are likely to be a source of high false-positives due to inorganic contamination.
The trouble with these impurities is they are not flagged by purity checks on organic material and, as demonstrated by the Roche group, they can often maintain activity in downstream assays. It may be possible to run screening assays in the presence and absence of a non-selective chelator such as EDTA, however, the only way to be fully confident of the hits would be to re-purify, or re-synthesise hits. This is especially important when compounds are obtained from external sources and the route of synthesis is unknown.
Finally it is important to congratulate the Roche group for not only the investigation, but for publishing the results. It is something many of us will have come across, but mostly we will just move on rather than fully explore the frequency of the issue and share the information with the rest of the scientific community.

1. Hermann JC, Chen Y, Wartchow C, Menke J, Gao L, Gleason SK, Haynes N, Scott N, Petersen A, Gabriel S, Vu B, George KM, Narayanan A, Li SH, Qian H, Beatini N, Niu L, Gan Q. Metal Impurities Cause False Positives in High-Throughput Screening Campaigns. ACS Med. Chem. Lett. Chem. Lett. 4: 197–200, 2012.

‘Alzheimer’s in a dish’ – A 3D cell culture model of Alzheimer’s disease.

The Amyloid Hypothesis is a proposed model  describing the development of Alzheimer’s disease. The deposition of excess amyloid- β peptide leads to the formation of amyloid plaques which then go on to form neurofibrillary tangles. These tangles are further composed of  hyperphosphorylated Tau. However, there has been doubt surrounding this hypothesis as some mouse models with mutations in Familial Alzheimer’s disease (FAD) genes do not show the same development of the disease as humans, forming amyloid plaques but not forming neurofibrillary tangles. Studies using human neurons from Alzheimer’s patients have also shown that these cells display increases in toxic amyloid-β species and phosphorylated Tau but without the formation of amyloid plaques or neurofibrillary tangles.

This month a group in the US have published an elegant study using a 3D cell culture system to recapitulate the development of ‘Alzheimer’s disease in  a dish’, providing evidence that the 30 year old Amyloid Hypothesis may well be correct and raising hopes for the future of Alzheimer’s research.

The group overexpressed human beta-amyloid precursor protein (APP) and presenilin (PSEN1) constructs with FAD mutations in human neural progenitor cells (ReN cells). These cells then differentiated into neuronal and glial cells within three weeks and exhibited an increase in neuronal marker genes as well as increased levels of  Tau and amyloid-β isoforms. They then transferred these cells to a 3D culture system using BD Matrigel containing high levels of brain extracellular matrix proteins. They observed that this 3D culture method promoted more neuronal and glial differentiation than equivalent 2D cultures, together with increased Tau isoform expression.

After a further 2-6 weeks of differentiation, an increase in extracellular amyloid-β deposits was observed in the FAD ReN cells compared to the controls.  The level of deposition could be decreased by treatment with β-secretase inhibitor IV or γ- secretase inhibitors DAPT and SGSM41. The FAD ReN cells also exhibited accumulation of insoluble amyloid-β aggregates, which again could be decreased by treatment with γ- secretase inhibitors.


Amyloid-β deposits in control (ReN-G) and FAD ReN (ReN-mGAP)

The group also analysed levels of phosphorylated Tau in the FAD ReN cells and detected a large increase in levels of both phosphorylated and total Tau. The level of phosphorylated Tau in the FAD ReN cells could also be decreased by treatment with β- or γ- secretase inhibitors, suggesting that phosphorylated Tau accumulation in these cells is a consequence of amyloid-β accumulation. They further observed phosphorylated Tau aggregates and filamentous structures that were strikingly similar to those structures observed in Alzheimer’s patient brains. Treatment with GSK3β inhibitors 1-azakenpaullone and SB415286 reduced levels of phosphorylated Tau without affecting total Tau or amyloid-β levels, suggesting a role for GSK3β in tauopathy downstream of deposition of amyloid-β.

This important work has not only provided experimental evidence supporting the Amyloid Hypothesis of Alzheimer’s disease but also generated a model system that can be used for studying the mechanisms behind the pathology of the disease. The 3D culture system could also be used for screening of drugs and could potentially be modified to develop models for other neurodegenerative disorders.


Choi SH, Kim YH, Hebisch M, Sliwinski C, Lee S, D’Avanzo C, Chen H, Hooli B, Asselin C, Muffat J, Klee JB, Zhang C, Wainger BJ, Peitz M, Kovacs DM, Woolf CJ, Wagner SL, Tanzi RE, Kim DY., A three-dimensional human neural cell culture model of Alzheimer’s disease., Nature. 2014 Oct 12. doi: 10.1038/nature13800. [Epub ahead of print]

Can we predict compound precipitation in DMSO stocks?

Collections of compounds used within drug discovery screening projects have to be tested in a variety of different assay types and therefore are stored in a liquid form. The most widely used solvent for this purpose is DMSO (Dimethyl sulfoxide).

However some compounds do not remain in solution and fall out forming a precipitate. A team at GlaxoSmithKline investigated this issue, in this publication – (Ioana Popa-Burke and John Russell, “Compound Precipitation in High-Concentration DMSO Solutions.,” Journal of biomolecular screening, 19 (2014), 1302–8 .

In the article the team noted that from one library of compounds – the “Tox Set” stored at 100mM they measured a 15.17% precipitation rate (by means of visual inspection). This was compared to a collection of fragment based compounds (at 100mM concentration) which had a 4.76% precipitation observed. It should be noted the there was a difference in the total number of theses set of compounds 422 versus 1995 respectively, which may explain some differences. The team also investigated other fragment and diversity compound collections with a higher number of members but at a lower concentration (a 40mM Fragment set with 7137 members and 10mM diversity set of 38,360 members). These both gave similar observed precipitates of 3.45% and 3.11% respectively.
The water content of all the DMSO samples was measured using Echo 555 acoustic dispenser, and this was similar across the samples ranging from 90.9% to 92.0% in all collections apart from the fragments at 100mM which had 85.8% DMSO. It was therefore assumed that % water content was not to a cause for precipitation. As all these compounds had been prepared and solubilised in the same manner, it was concluded that a chemical property must be the driver of precipitation of these compounds.

To determine if a physicochemical property could be identified as a correlating factor for increased precipitation rate in the Tox set compared to the fragment set , the MW, clogP, fsp3 (fraction of sp3 carbons) and TPSA (total polar surface area) were analysed with three different data mining techniques however no relationship was uncovered.
The identified precipitating compounds soluble concentration and purity was determined using LC-UV-MS-ELSD system. This revealed that concentration was, as to be expected lower in the free solution of precipitated samples, however the level of purity was similar for precipitating compounds as fully soluble compounds, suggesting that impurities are not the causative factor in the production of precipitates.

To determine if number of freeze thaw cycles would increase the number of precipitates observed, the team took the identified 110 precipitating compounds from the 100mM tox and 100mM Fragment set and made fresh solution in DMSO at different concentrations. These samples were then exposed to 1, 2, 5, and 10 freeze thaw cycles, and the number of precipitates was observed.
Number of Freeze thaw cycles did not have a significant increase in the number of compounds that precipitated however there was a correlation with compound concentration.
A key result was that from this set of precipitating compounds was 86% were unable to go into a solution at 100mM initially before any freeze thaw cycle.

So to answer the question – no, from this latest publication the authors cannot predict which compounds will precipitate from a DMSO solution. Compound concentration does have an effect on the number of precipitates observed, which is important to remember when thinking about compound library composition and stored concentration.

Fragment Screening: False Positives – (2,5-Dimethyl-1H-pyrrol-1-yl)benzoic acid

After a recent screening campaign in the Translational Drug Discovery Group (TDDG), using a well-known fragment library, we were looking into confirming our hits by repurchasing them and purifying these samples. When searching for 4-(2,5-dimethyl-1H-pyrrol-1-yl)benzoic acid (3) we came across this paper by Rolf Hartmann et al( Chem. Eur. J. 2013, 19, 8397 – 8400). in which they describe their elucidation of a screening artefact contained within their sample of 3.
Hartmann was screening fragments in the search for novel bacterial RNA polymerase (RNAP) inhibitors and came across pyrrol 1 as a positive hit from their screening library. Upon further reading of the literature they found that structurally related compounds 2-4 had also previously been described as RNAP inhibitors. However, once compounds 1-4 had been resynthesized and tested no activity was observed. Hartmann did notice, as had we, that after purification these compounds are colourless but quickly develop a red tone when a solution of them is left open to the air on the bench.

lewis 1

Hartmann postulated that the observed activity that he and others had seen from these compounds was coming from a decomposition impurity rather than the parent pyrrol. To accelerate their decomposition the purified inactive parent pyrrol 1 was heated to 50°C in DMSO for 10 days and the resulting HPLC trace and NMR spectra can be seen in figure 2.

lewis 2

As the C-H protons from the pyrrol ring (5.7 ppm) of 1 had disappeared and a broad new signal was observed in the aromatic region (7.0-8.5 ppm) it was thought that the decomposition product was a polymer. The decomposition sample was subjected to ultrafiltration (cut-off 3.5kDa) and it was shown that the active component was of a high molecular weight which they named P1. Figure S3 shows a broad molecular weight distribution of P1 by gel permeation chromatography (GPC) with a weight-average molecular weight of 40kDa

lewis 3

The UV/Vis spectroscopy of P1 shows an absorption peak at 498 nm (2.49 eV) which is associated with a π-π^* transition indicating a well conjugated π electron system in the backbone (figure S4). This absorption at 498 nm was also seen in the commercial starting material but not in the purified pyrrol 1.

lewis 4

The decomposition of compounds 2-5 were carried out and the corresponding polymers P2-P5 were isolated. A carboxylic acid signal was observed by IR for P1-P4 which was supported that they could be easily dissolved in water at basic pH in contrast to P5. The polyanionic structure of P1 was confirmed by gel electrophoresis. At pH 3.6 P1 and P5 remained at the starting point but at pH 7.8 P1 migrated to the anode and P5 remained at the starting point.
Using all of the above analysis Hartmann has postulated that the structure of P1 (figure 3).

lewis 5

Compounds P1-P5 were tested against E. coli RNAP in an in vitro transcription assay (table 1). None of the monomers affected transcription but all of the carboxylic acid containing polymers displayed a concentration dependant inhibition.

lewis 6

Figure 4 shows the positively charged DNA binding channel which was used to explain why P5 is inactive against RNAP but the other polyanionic P1-P4 show inhibitory activity. Further experiments were conducted that all supported this hypothesised mechanism of P1 RNAP inhibition.

lewis 7

When P1 was tested against 8 other RNA polymerases and only 1 (bacteriophage T7) was strongly inhibited. Unfortunately P1 showed no inhibition in growth of E. coli and Pseudomonas aeruginosa. This lack of in vivo activity has been put down to P1 being to hydrophilic for passive diffusion and too large to permeate the porins. Hartmann was hopeful that if he could reduce the average size of the P1 polymers then he would able to achieve in vivo activity by permeating these porins.
Hartmann has taken the time in this paper to thoroughly investigation a false positive from their own work and also confirmed that the other close analogues that had previously been described as RNAP inhibitors were also false positives. Not only is this an interesting paper as Hartmann has been able to identify the previously unknown inhibitor but it also helps the wider community potentially avoid wasting resources pursuing these Pan Assay Interference Compounds (PAINS).
Key words
Fragment screening, false positives, PAINS