How do we improve early go/no-go decisions in bio/pharma discovery and development so as to reduce costly late-stage attrition?
A review of recent phase II and III attrition reports that 56% of failures were due to efficacy, compared to only 28% due to safety (including insufficient therapeutic index) [i]. Efficacy is therefore the greater problem, where efficacy is the product of potency and biological rationale. Yet, the medicinal chemistry field is advanced enough that developing small molecule potency is non-limiting. Affinity maturation producing potent monoclonal antibodies (mAb) is also non-limiting. Therefore the efficacy issue could be principally attributed to poor biological rationale, which can be improved by more detailed characterisation of mechanism of action, and also by better recapitulation of clinical biological context.
Both of these aspects are elucidated at the early stages of bio/pharma discovery: during target identification and validation, as well as during hit identification and validation (small molecule) or antibody selection (biological). Targets supporting drug discovery may be novel or established, where “established” relates to the availability of background information available on the target. This comprises scientific understanding, supported by lengthy publication history, on the function of the target in both normal physiology and human pathology. Where this functional information is not available, such as for novel targets, it is generated through the target validation process[ii].
There is significant discussion in academic literature relating the scarcity of well-validated targets to the duplicative efforts toward me-too and me-better drugs by BioPharma. Conversely, it was reported that novel targets represented approximately 74% of active drug projects in 2013 [iii]. Whatever the reality, there is much advocacy both within and without the BioPharma industry for accessing greater target space to support more innovative drug development, and in concurrence for improved target validation to help de-risk such development [iv].
This growing focus on novel target ID and validation provides a necessary escape from traditional targets, for which development efforts are highly competitive [v]. This is particularly true in oncology where validated targets are rare, and highly validated ones are extremely competitive [vi]. In conjunction, oncology and CNS indications account for the most phase II and III failures (30% and 14% respectively) [vii]. Therefore the improvement of pipeline quality for these indications is a prime opportunity.
This article reviews the early therapeutic development process as the context for such opportunity.
Analysis of Therapeutic Discovery Value Chain
1 Target Identification
Novel targets are sourced for biopharmaceutical development principally from academia, bioinformatic analyses, functional screens, and pathway analyses.
1.1 Academia
Whether a target is novel or established relates to the availability of background information. This comprises scientific understanding, supported by lengthy publication history, on the function of the target in both normal physiology and human pathology. Through literature and conference presentations, the academic literature is therefore a free source of candidate targets.
However, academic data suffer from poor quality. It was recently reported that in 60% of drug discovery projects based on academic information, inconsistencies between published and in-house data either considerably prolonged the duration of target validation or, in most cases, resulted in project termination because evidence was insufficient to justify further investment. Reasons include incorrect or inappropriate statistical analysis, negligence over the control or reporting of experimental conditions, and/or a bias towards publishing positive results .
The low quality of academic data therefore aggravates attrition in Target Validation. The cost of this attrition is an important message to communicate to bio/pharmaceutical developers and could be a benchmark to the value-add pricing of Target ID services by vendors.
1.2 Bioinformatic analyses
Bioinformatics can be used to systematically mine public and/or private datasets for differences in genome, transcripteome, and/or proteome between normal and diseased tissue. These differences suggest involvement in disease aetiology and are subsequently prioritised for additional investigation.
As a genetically driven disease, cancer is a good example and is the focus of several consortia generating genomic, transcriptomic and epigenomic information for a diversity of different cancer types and subtypes. In addition to defining the repertoire of cancer genes and other molecular abnormalities, another objective of these consortias’ profiling campaigns is to understand how cancer genes interact together in dynamic networks. This may identify further targets, prioritize certain target loci important in particular genetic systems and disease contexts, and also may reveal synergistic interactions such as synthetic lethal effects that can be exploited for drug development .
Whether public or private, these consortia and their activities should be monitored by vendors in the Target ID space. The information they disseminate represents competition, and may also raise the expectations of clients as to the degree of genetic characterisation accompanying oncology models. These consortia may also represent good opportunities for collaboration, for example to push standards in oncology models and their datasets that help to mainstream specific vendor products.
1.3 Functional screens
Functional screens involve multiplex modulation of gene expression in either non-transformed or established cell line models to correlate gene function to specific (patho)phenotypes. Such screens are offered as contract services by early discovery CROs.
In cancer, gain-of-function screens involve cDNA libraries to discover aggravation or mitigation of pathophenotypes, suggesting oncogenic or tumour suppressor targets respectively. Conversely loss-of-function screens involve siRNA or shRNA libraries to discover aggravation or mitigation of pathophenotype suggesting tumour suppressor or oncogenic targets respectively. Tumour suppressor targets have historically been considered undruggable due to the difficulty of developing small molecule agonists. Currently however, there is a growing focus on identifying synergistic interactions, such as synthetic lethal effects, as a means to pharmacologically exploit tumour suppressor loss-of-function. A good example is the use of PARP inhibitors in BRCA mutant patients.
The large scale required of functional screens in search of novel targets requires the use of scalable technology for modulating gene expression. At present, only RNAi fulfils this requirement, however the recent development of CRISPR systems for precision genetic editing may offer a new avenue.
1.4 Pathway analyses
Tool antibodies and tool compounds, such as chemical probes or peptides, are used to modulate specific pathways of interest to reveal novel pathophenotype dependencies. Specific candidate targets within these pathways can then be identified by the use of reporters to elucidate modified cellular activity, and/or by the systematic modulation of molecular actors using RNAi or precision genetic editing.
Due to the high cost and technical complexity of developing and validating such tools (comparable to running HTS for hit discovery), their use is limited. There are currently very few commercial providers of these tools.
2 Target Validation
Target Validation is a critical step that establishes the confidence to initiate a hypothesis-driven therapeutic discovery program. This validation can be conceptualized into demonstrating disease dependence, demonstrating clinical relevance, and demonstrating druggability and development feasibility.
2.1 Demonstrating disease dependence
Demonstrating disease dependence entails a robust validation of the initial correlation between target function and pathophenotype suggested in target ID. The initial validation is in vitro but must subsequently also be validated in vivo.
2.1.1 In vitro validation
Most typically this involves genetic and transcriptional modulation of the target. The “gold standard” for this purpose, used virtually in every case, is a loss-of-function and rescue experiment performed as follows:
1. RNAi-mediated target suppression modulates the pathophenotype in a cellular model; demonstrating disease dependence.
2. An exogenous construct of RNAi-resistant target rescues the pathophenotype; demonstrating the on-target specificity of RNAi knockdown.
3. An exogenous construct of RNAi-resistant, activity-dead target does not rescue the pathophenotype; better mimicking small molecular inhibition and validating the specific target function-dependency of the pathophenotype. Rational target engineering in this third step helps to define the structure-activity relationship. This is critical as single proteins can have diverse biological functions, in particular targets harbouring more than one functional domain and/or interacting with several partners.
Optionally, a gain-of-function experiment can also be performed to demonstrate that expression/activation of the target is sufficient to drive a pathophenotype in normal cells. This typically involves cDNA over-expression or an inducible cell line. In cancer research, such experiments will measure aggravation in hallmarks of oncogenesis including growth rate, immortalization, loss of contact inhibition, two-dimensional clonogenic survival, or anchorage-independent growth in soft agar.
An alternative to genetic and transcriptional modulation is the use of tool compounds/antibodies to modulate target or pathway activity. These tools are much more predictive of the eventual small molecular or biological drug candidate, but as mentioned are not easily available in the validated form required for robust interpretation and are very costly to develop. However when available, their use is highly complementary to RNAi in being able to inhibit a specific function of the target protein rather than removing the whole protein, thus avoiding multiple function or scaffold effect issues; in giving an immediate inhibition rather than a delayed knockdown; and in providing greater control over the extent and kinetics of inhibition .
Whichever approach is used, In the case of cancer it is important to demonstrate target modulation leading to the expected effect in multiple cell lines with relevant genetic backgrounds. It is equally important that no or minimal effect is observed in cell lines with a different genetic background, and ideally also in normal, non-transformed cell lines from the same tissue, although the latter are challenging to work with in culture and to transduce .
Notably, although in vitro validation through gold standard RNAi suppression and rescue is used in virtually all of target validation destined for small molecular projects, the case is different for mAb projects. Antibodies are targeted either at the cell surface of tumour cells or at components of the tumour microenvironment; thus they have a greater need for the external cellular context to be recapitulated. Consequently, approximately 50% of mAb targets at the tumour cell surface and almost 100% of targets in the tumour microenvironment skip directly to in vivo validation studies.
2.1.2 In vivo validation
Subsequent to the in vitro demonstration of disease-target dependence, in vivo data is required to validate the functional relationship in a more clinically-predictive context. At this stage, the quantitative consequences of target modulation must be shown to be sufficient to deliver a therapeutically meaningful biological effect in relevant experimental models. Cancer research typically employs xenografts on nude mice, and less frequently genetically engineered mouse models, in conjunction with either shRNA suppression or tool antibodies/compounds where available.
2.2 Demonstrating clinical relevance
In initial target validation, the prevalence of the target pathogenotype within the patient population must be sufficient to justify developing a complementary therapeutic. Often this data is already available, especially in the case of cancer. More robust demonstration of clinical relevance is subsequently achieved 1.5-2 years later when candidate molecules are assayed in clinical samples.
However if prevalence data of the target pathogenotype is not available upfront, it must be assessed across a large panel of histotypes in order to validate the actual clinically relevant population associated with the envisioned target. The epidemiological information obtained from the analysis of clinical samples can then be applied to the overall validation and drug discovery strategy with the aim of closely matching later experimental systems in vitro (e.g., cell lines) and in vivo (e.g., GEMMs; primary xenograft) with the envisioned relevant clinical indication.
2.3 Demonstrating druggability and development feasibility
This is easier in target classes known to be druggable (e.g. kinases, GPCRs), for which druggability may be inferred by association, for example when sequence homology is high to kinases with existing inhibitors. Publicly available databases, such as canSAR , allow searching for homologous targets based on protein sequence and structure and also report published chemical hit matter. Thus an added benefit is that inhibitors of the related target can inform small molecular design.
However the druggability of a protein can never be predicted with absolute accuracy, especially if little information on the target is available. There is also a ‘druggability gap’, where many targets with very promising disease linkage, (e.g. mutated RAS proteins or transcription factors like c-MYC or hypoxia-inducible factor) are considered technically undruggable or extremely challenging to target with small molecules. However, it is recognised in industry that evaluations of druggability are biased by the restricted target space that the pharmaceutical industry has historically tackled and is limited by the static nature of protein structure snapshots, which do not allow appreciating protein plasticity and dynamism.
In addition to druggability, the feasibility of development must be considered. At this stage this refers to the availability of assays for the biological test cascade involved in a drug discovery campaign. Transcription factor targets are again a good example, being not easily amenable to high-throughput screening because it is difficult to develop assays that identify compounds interfering with protein–DNA or protein–protein interactions. Thus alternative approaches to small molecule discovery for these targets are needed.
3 Target-driven small molecular discovery
Once a target is validated, the focus progresses to its successful modulation, for which a therapeutic agent must be developed. For the development of small molecular therapeutics, the target is incorporated into a biochemical assay for high throughput screening of libraries of chemical compounds. This activity is a mixture of design and screening elements.
Traditional high throughput screening has lost popularity in recent years in favor of increasing the design element, which depends on more detailed biophysical target characterization. This enables techniques such as virtual drug design, fragment-based screening, and the use of more focused compound libraries.
Once generated, hit matter must undergo retesting, orthogonal validation, and secondary profiling. Hit clusters will then be prioritized based on the output of these steps, including the following outputs of cell-based assays :
• Orthogonal validation of target engagement and significant biological activity in a cellular assay.
• Satisfactory cell membrane permeability (ADME)
• Slow metabolisation (ADME)
• Lack of cytotoxicity (Tox)
• Selectivity versus other related targets
At this stage, the cell-based assays used do not specifically aim to recapitulate the clinical context of disease. Reasons include reliance on the validation of existing assays, the need to keep costs low at this scale of throughput, and the expectation that the downstream use of more accurate clinical models is sufficient. However as part of the growing trend of frontloading improved clinical relevance to reduce attrition, there is an open attitude to considering more accurate cell models that remain cost-effective.
4 Monoclonal antibody discovery
Validated targets may be the focus of monoclonal antibody discovery instead of small molecular. In this case, pathway-based target discovery is a particularly relevant approach to expanding the existing target space since a large number of signaling pathways are initiated at cell surface receptors and mAbs are currently limited to targeting cell surface or extracellular targets. Industry-academic collaborations are also very significant in mAb target discovery.
Three approaches are dominant in therapeutic mAb discovery: wild-type animal immunization followed by antibody humanization, transgenic animal immunization, and surface display methodologies.
The only use of cell-based screening assays in mAb discovery is in surface display methods, whose key feature is the direct linkage between the antibody variant and the complementary genetic code . Surface display involves high-throughput in vitro screening of antibody fragment libraries displayed on bacteriophages or cells.
Cell-based antibody display and screening has certain advantages over bacteriophage-based methods, notably including quality control machineries for correct protein folding and post-translational modifications. However, current cell display technology is focused on yeast . Mammalian cell surface display is much more recent and still too nascent for use in industry.
Notably however, a significant challenge to the use of immunization-based antibody discovery is that many novel mAb targets are multispanning membrane proteins. Conventional biochemistry in preparing soluble protein as immunogens does not work well for such targets . This may therefore favor surface display going forward, at least for this target class, and drive interest in developing mammalian cell display technologies.
Overall however, the market opportunity for engineered cell lines is slim in mAb discovery, and very highly specialized in contrast to general applications of precision genetic editing.
5 Target-agnostic phenotypic screening
An alternative to target-based hit generation is the use phenotypic screening, which tests the impact of small molecules directly in a cellular system. This target-agnostic approach ‘prevalidates’ the small molecule and its (initially unknown) target as an effective means of perturbing the biological process or disease model under study. However, the major limitation of phenotypic screening is the difficulty in subsequently elucidating a molecular drug target . Although not required by regulatory authorities, this information is invaluable for the subsequent optimization of small molecular candidates. This problem is less severe when a hypothesis-driven project employs a pathway-specific phenotypic assay, however the majority of phenotypic screening is pathway-agnostic.
A further significant complication is that small molecule–induced phenotypes observed in cell culture may represent the superposition of effects on multiple targets. However, successfully identifying both the therapeutic target and other targets that might cause unwanted side effects enables optimization of small-molecule selectivity . Conversely, there is a rising interest in polypharmacology where multiple small-molecule effects may be leveraged to gain maximal effect. This interest is supported by analyses of 890 approved drugs indicating that 788 share molecular targets with at least one other drug, and that on average each drug interacts with six known molecular targets anyway . To this end, phenotypic screening provides a means to agnostically interrogate multiple, biologically relevant pathways.
Recombinant cell lines and immortalized primary cells are commonly used in phenotypic screening, largely because they rapidly proliferate to support the large quantities of cells needed. However, primary human cells and patient derived cells are more desirable because of their greater biological insight and disease relevance. A compromise may be the use of embryonic stem (ES) cells and induced pluripotent stem (iPS) cells, which can be differentiated to many types of mature cells such as neurons, cardiomyocytes and hepatocytes for drug screens. While the methods for stem cell differentiation including the differentiation efficiency, scale-up, reproducibility and cost effectiveness are still being improved, several pilot compound screens using stem cell differentiated progenitor cells have been recently reported.
Hit matter generated from phenotypic screening are orthogonally validated to remove generally cellular toxicities and assay-dependent effects. Hits are then prioritized through secondary profiling, such as on the basis of preliminary ADME and successfully establishing SARs. Target deconvolution is then attempted only for top ranking candidates due to the significant investment in time and resources required.
6 Target Deconvolution
Target deconvolution relies on two distinct approaches, chemical proteomics and functional genomics, supported by bioinformatics-enabled comparison with existing public and private datasets. These are generally used in complementary fashion to triangulate the target and therefore developments in either field may help to boost the use of phenotypic screening in drug discovery, and the associated market opportunities in functional genomics for vendors of related technologies.
6.1 Chemical Proteomics
Chemical proteomics is the most prevalently used method. Unlike genetics- and bioinfomatics-based deconvolution, chemical proteomics does not rely on inference. It demonstrates direct interaction between compound and cell lysate components. These interactions with putative targets, off-targets, and their complexes help inform mechanisms of action for efficacy and toxicity, as well as potentially allow evaluation of polypharmacology.
The greatest drawback is the significant background noise caused by high-abundance, low-affinity proteins. Increasing the stringency of affinity purification can ameliorate this but will also eliminate protein complexes involving the target, as well as secondary, lower affinity targets that contribute to the overall effect. Another difficulty is finding a suitable, inactive analog for use as a control. Yet another difficulty is that preparing immobilized affinity reagents that retain cellular activity is technically difficult and time consuming.
Notably, chemical proteomics has also been used to discover new targets for known drugs, allowing expansion into new disease indications. Therefore the growing interest in drug repurposing may support the development of this field and of phenotypic screening in general.
6.2 Functional genomics
Functional genomic target deconvolution involves systematic gene modulation to reveal impacts on drug activity. The most prevalent approach is the use of yeast knock-out libraries to explore dependencies of drug sensitivity through drug-induced haploinsufficient profiling, homozygous profiling, and multicopy suppression profiling. Integrating the results of these complementary assays helps to triangulate likely target pathways.
The use of yeast currently dominates due to its genetic tractability, however its limited translatability to human biology is driving alternative approaches in mammalian systems. In this context, parallel genome-wide RNAi screening can be related to small molecular phenotypes in an attempt to connect target pathways through similar phenotypes. Any existing mechanistic information, such as derived from chemical proteomics, can be used to focus the set of RNAi reagents. Alternatively, screening with RNAi or cDNA libraries can be used to mimic the yeast assays described above in investigating drug sensitivity. Functional genomics therefore is an interesting opportunity for vendors of RNAi or CRISPR technology solutions.
These functional genomic approaches are particularly suitable for oncology-relevant targets because growth inhibition and cell death are easy outcomes to measure. Other endpoints, such as reporter activity needing to be quantified at the well level, are more technically difficult and significantly limiting to functional genomic target deconvolution. Some of these limitations are obviated by the application of high content sceening (HCS), which can examine multiple biological or phenotypic parameters at the single cell level as well as quantitatively evaluate spatially-derived cellular changes and cell population heterogeneity. The current developments in HCS are therefore likely to enable broader applications of phenotypic screening through functional genomics, including through facilitated expansion of both endpoint- and live cell-based studies into multiwell plate formats suitable for drug or siRNA/shRNA screening applications .
7 Concluding remarks
The use of RNAi technology for screening is prevalent in Target ID & Validation, as well as in Target Deconvolution, but not in Hit Discovery (small molecular) or Antibody Selection (biological). The decision by pharmaceutical firms to perform hit discovery in a target-driven or -agnostic manner will determine whether Target ID & Validation or Target Deconvolution is required, and therefore understanding the relevant trends will be crucial to the positioning of vendors. Notable examples include the effectiveness of fragment-based screening in rationale drug design, the enablement of phenotypic screening by HCS, and the co-dependence of functional genomics and chemical proteomics in Target Deconvolution.
As regards biopharmaceutical firms, because Antibody Selection is intrinsically target-driven the rising prevalence of mAb therapeutics will strengthen the demand for Target ID & Validation services over Target Deconvolution. A caveat to this is that the localization of mAb targets (cell surface or extracellular) weakens the utility of in vitro validation in simple cell line models, and consequently the use of RNAi or genetic editing technologies therein. This suggests that expertise in applying such technologies in vivo is more relevant to the development of mAb therapeutics.
Precision genetic editing technologies have applications in all segments analysed for generating cell lines that better recapitulate the clinical context of disease. Applications in phenotypic screening are particularly strong due to the underlying logic of recapitulating the clinical context. By contrast applications in Antibody Selection are limited and highly specialized. Notably, in steps where RNAi screening is also applied, the two technologies may be beneficially packaged together by vendors.
[i] Arrowsmith et al. (2013) “Trial Watch: Phase II and Phase III attrition rates 2011–2012” Nature Reviews Drug Discovery 12:526
[ii] Kyla notes on drug discovery
[iii] Agarwal et al. (2013) “Novelty in the target landscape of the pharmaceutical industry”, Nature reviews Drug discovery Vol. 12:575
[iv] 2012 Discovery of small molecule cancer drugs successes, challenges
[v] Agarwal et al. (2013) “Novelty in the target landscape of the pharmaceutical industry” Nature Reviews Drug Discovery 12:575
[vi] Moll et al. (2012) “Target identification and Validation in drug discovery”, Springer methods and protocols, Humana Press
[vii] Arrowsmith et al. (2013) “Trial Watch: Phase II and Phase III attrition rates 2011–2012” Nature Reviews Drug Discovery 12:526
Kommentare