I say article of the week, but I have been lazy this summer (or maybe just consumed by other things). My last “article of the week” was in May and my last Plengegen blog post was over a month ago!
By now everyone knows the PCSK9 story. Human genetics identified the target; functional work in mouse and human cells led to a mechanistic understanding of PCSK9’s role in LDL receptor recycling; therapeutic modulation was shown to lower LDL cholesterol in clinical trials; and the FDA approved drugs based on LDL lowering, with outcome trials underway to demonstrate (presumably) cardiovascular benefit. What the story highlights is that a mechanistic understanding of causal pathways in human disease is key to the success of translating targets into therapies. Further, the PCSK9 story underscores the importance of a simple biomarker (LDL cholesterol) to measure a complex causal pathway in a clinical trial.
A recent study in the New England Journal of Medicine (NEJM) provides insight into a putative causal pathway in obesity, and thus a potentially a new mechanism for therapeutic modulation. The accompanying Editorial also provides a nice perspective.
[Disclaimer: I am a Merck/MSD employee. The opinions I am expressing are my own and do not necessarily represent the position of my employer.]
Previous studies have implicated at least two different types of human adipocytes in fuel storage and energy expenditure: white fat stores fuel as triglycerides and brown fat generates heat (thermogenesis). In contrast to white adipocytes, which contain a single lipid droplet, brown adipocytes contain numerous smaller droplets and a much higher number of mitochondria. A third type, beige fat (or “brown-like”), develops in white fat in response to various activators and has thermogenic properties that mimic brown fat (see Nature Medicine review article here). It has been hypothesized that therapeutic targeting of brown or beige fat may be useful in treating obesity.
But is the pathway causally related to human obesity, how would one design a drug discovery program to therapeutically modulate human adipocytes to promote weight loss, and what biomarker would one measure in a clinical trial?
Primary human adipose–derived progenitor cell cultures (“human preadipocytes”) were obtained from the subcutaneous adipose tissue of 100 healthy Europeans. Half were homozygous for the FTO risk allele and half homozygous for the non-risk allele. From these human preadipocytes, chromatin maps were generated and FTO risk allele-mediated differential gene expression measured via luciferase reporter constructs (Figure 1). The same human cells were used to show that the FTO risk allele also associated with differential gene expression of IRX3 and IRX5 (Figure 2). A different set of human adipocytes was used to identify genes with expression that was positively or negatively correlated with IRX3 and IRX5 expression. As shown in Figure 3, the pattern suggests that IRX3 and IRX5 influence energy dissipation and storage (mitochondrial genes were negatively correlated, lipid-metabolism genes were positively correlated). Further, IRX3 and IRX5 knockdown in human preadipocytes restored oxygen consumption and thermogenesis response. The mouse data, buried in the Supplementary data, provide mechanistic support for the human observations: Irx3 and Irx5 induce adipocyte lipid accumulation and repress thermogenesis in a cell-autonomous way. Taken together, the data indicate that IRX3 and IRX5 may repress thermogenesis in adipocytes through an FTO-mediated mechanism.
Next, the study went on to identify which of the 89 variants in linkage disequilibrium across the FTO locus is the most likely “causal allele”. Through a series of traditional molecular assays, they demonstrate that the FTO risk allele prevents the binding of repressor protein, ARID5B, thereby leading to increased enhancer activity at the FTO locus. They conclude that the most likely risk variant is rs1421085 (Figure 4A). They also demonstrate that ARID5B knock-down in human preadipocytes increased IRX3 and IRX5 expression from nonrisk-allele carriers to risk-allele levels; reduced basal oxygen consumption and lipolysis; and shifted expression patterns from mitochondrial to lipid markers (Supplementary data). As summarized in the manuscript: “These results suggest that the FTO obesity variant acts through disruption of ARID5B binding in the risk haplotype, leading to a loss of repression, a gain of enhancer activity, and increases in IRX3 and IRX5 expression (Fig. S6M in the Supplementary Appendix).”
Finally, as shown in Figure 4C-D and Figure 5, they use genome engineering via CRISPR–Cas9 in human preadipocytes to “rescue” the ARID5B motif disruption in rs1421085 risk-allele carriers, followed by assessment of changes in (a) gene expression levels and (b) cellular signatures of obesity. The results were consistent with rs1421085 as the causal variant: (a) differential regulation of IRX3 / IRX5, thermogenesis regulators (ADRB3, DIO2, PGC1A, and UCP1), mitochondrial markers (NDUFA10, COX7A, and CPT1), and lipid storage / lipolytic genes; and (b) alterations in basal metabolic rate, oxygen consumption, and thermogenesis. As summarized in the manuscript and succinctly illustrated in Figure 5D: “These results indicate that the rs1421085 T-to- C single-nucleotide alteration underlies the association between FTO and obesity by disrupting ARID5B-mediated repression of IRX3 and IRX5. This disruption leads to a developmental shift from browning to whitening programs and loss of mitochondrial thermogenesis.”
Why this is important for drug discovery:
First, the study provides causal support for mechanisms related to brown and beige fat in human obesity and generates new therapeutic hypotheses based on regulatory networks in the early differentiation of human preadipocytes. Previous genetic studies have implicated the central nervous system in obesity, including those related to synaptic function, glutamate signalling and energy metabolism (see Nature manuscript published earlier in 2015 by Liz Speliotes, Joel Hirschhorn and colleagues here). However, these same genetic studies also point to pathways related to lipid biology, insulin secretion/action and adipogenesis. Together with the detailed functional data from Claussnitzer and colleagues, these studies provide support for adipogenesis and the ARID5B–FTO–IRX3/IRX5 regulatory network as causally linked to human obesity.
Second, the study provides a starting point for a drug discovery program (see brief slide deck here). One approach would be therapeutic perturbation of FTO directly, or therapeutic modulation of other genes / protein products in the ARID5B–FTO–IRX3/IRX5 pathway. For example, it is possible that drugs that perturb IRX3/IRX5, downstream thermogenesis regulators (e.g., ADRB3, DIO2, PGC1A, UCP1) or downstream mitochondrial markers (e.g., NDUFA10, COX7A, CPT1) may favorably shift energy storage and expenditure in a brain-independent and tissue-autonomous way in humans. Another approach would be to use various adipocyte readouts described in the Claussnitzer manuscript as a starting point for a phenotypic screen. As stated in the Editorial: “shifting adipocytes from energy storage to energy expenditure with pharmacologic and nonpharmacologic measures may become more feasible as the ARID5B–FTO–IRX3/IRX5 regulatory network becomes fully defined”.
Third, the study is an elegant example of “humans as the model organism”, and the role of animal models to confirm mechanistic implications derived from humans. The study (a) started with a genetic observation in humans from GWAS; (b) fine-mapped the risk locus using 1000 Genomes data; (c) utilized primary cells derived from humans with different risk genotypes to understand biological pathways via co-regulated gene expression; (d) incorporated Epigenetic Roadmap data from humans; and (e) engineered human DNA with CRISPR/Cas9. What is striking to me is that none of these resources was available just ten years ago! Think about that: even though we had a single reference human genome sequence in 2000, we did not have the tools necessary to understand genetic differences that lead to disease. Today we do, and these tools – coupled with traditional molecular biology tools such as gel shift assays, luciferase reporter constructs and knockout mice – are critical to translate genetic findings into therapeutic hypotheses for drug discovery.
And fourth, it is a humbling reminder of the challenges of understanding human biology based on human genetic associations, and the steps towards an effective therapeutic. Mark McCarthy, Tim Frayling and colleagues published the original FTO genetic association study in 2007 by (PubMed Science link here). Nearly a decade later, we have a better, albeit still incomplete understanding of how FTO exerts its effect on BMI.
Building on this last point, it is important to state what the study does not provide. For drug discovery, robust biomarkers are essential for the translation of complex biological pathways into simple readouts that can be measured in clinical trials. Thus, even if a putative pharmacological molecule were to be developed, it would need to be tested in a clinical trial. Ideally, an early clinical trial would test proof-of-mechanism that modulating a target has the desired effect on human biology. And although it is not yet clear what molecular or imaging biomarker would be used to measure thermogenesis in humans, the determination of energy expenditure in man has been routinely used in the past. Furthermore, significant advancements to image the mass and activity of metabolically active adipose tissue in humans have been made in the past few years by combining PET and CT.
In addition, the study focuses on the ARID5B–FTO–IRX3/IRX5 regulatory network in the context of preadipocyte biology. These same genes likely influence other biological processes, which may have unexpected safety implications Towards this end, phenome-wide association studies (or PheWAS) serve as a powerful tool to explore the pleiotropic effects of functional genetic variants such as rs1421085. Such data could be used to understand the potential safety effects of perturbing a biological pathway that is likely tightly controlled by homeostatic mechanisms.