Most people with inflammatory bowel disease do not carry one broken gene. They carry risk spread across many genes and regulatory regions at once, which is why a condition affecting millions still resists a simple genetic explanation. In the most recent episode of The Genetics Podcast, host Patrick Short speaks with Dr. Carl Anderson, head of the Human Genetics Programme at the Wellcome Sanger Institute and leader of its Genomics of Inflammation and Immunity Group, about the single-cell atlas his team built to resolve that complexity.
Carl is best known for building IBDverse, a single-cell genetic atlas of inflammatory bowel disease assembled with collaborators at Addenbrooke's Hospital. The conversation traces that work from a 2017 finding that first pointed his lab toward the idea to where the resource stands today: gene regulation mapped across 89 distinct cell types, a growing list of biological pathways implicated in IBD, and a plan to extend the same longitudinal, multi-omic approach to the sickest patients in the healthcare system.
For most of his career, Carl has worked to map the regions of the genome associated with immune-mediated diseases, chiefly IBD, and by now his group has helped identify hundreds of associated loci, the vast majority of them common, non-coding variants. The hard part was never finding these signals but interpreting them: understanding which of the many nearby genes each variant actually perturbs. A 2017 study offered a way in, when his team found IBD signals that shared the same causal variant as eQTLs, genetic variants known to control gene expression, which pointed directly to affected genes. That single finding became the seed of a much larger ambition, to build what Carl describes simply as the world's best map of genetic effects on gene regulation in IBD-relevant tissue.
Executing on that idea meant building a cohort from scratch. Working with Tim Raine, head of IBD at Addenbrooke's Hospital, Carl's team collected gut biopsies and blood from hundreds of Crohn's disease patients, alongside biopsies from hundreds of healthy controls undergoing colonoscopy for routine bowel cancer screening. Those samples were dissociated into single cells and run through single-cell RNA sequencing, which let the team map genetic effects not just on bulk gut tissue, as earlier studies had done, but separately across 89 distinct cell types, from T cells to gamma delta T cells and Tregs. Lined up against two decades of IBD GWAS hits, that map lets researchers identify, cell type by cell type, which genes a given signal is actually likely to be perturbing.
A small minority of IBD cases are monogenic, typically presenting in infancy in patients with mutations like an IL-10 receptor knockout. The overwhelming majority is common, complex disease, roughly 50% heritable, driven by an overreactive immune response to gut bacteria in genetically susceptible people. What has changed most in the fifteen years since Carl's early GWAS work is the treatment landscape itself. Anti-TNF therapies only began being prescribed at scale in the early 2010s, and in the UK at that time patients effectively had to earn access to them through severe disease. Anti-TNFs are now off-patent, available as cheap biosimilars, and often a frontline option, alongside a growing arsenal of newer drug classes that increasingly have genetic evidence behind their targets.
That shifting treatment landscape complicates the biology in ways that are easy to miss. Because management has changed so much over time, patients diagnosed in 2006 experienced a very different disease trajectory than someone diagnosed today and started on advanced therapy early, producing what Carl calls era effects that any longitudinal analysis has to account for rather than ignore. Disentangling treatment response from underlying disease severity is genuinely difficult: a patient who responds well to a drug may have looked severe if they had never received it. Real unmet need remains, too. About 30% of patients still go on to require surgery because they do not respond to available drugs, the order in which drugs are given can shape how a patient responds to the next one, and some patients' immune systems learn to recognize a biologic as foreign and clear it from their system before it can work.
A question Carl says he is asked constantly is whether GWAS, rare coding variants, and expression data actually point to the same underlying biology. In his experience, they do, and consistently so. Different genetic variants perturb the genes within a pathway to different degrees, but the pathways themselves tend to be the same regardless of which type of genetic evidence is used to find them. That consistency lets his team use genetics to test its own hypotheses: if a variant that raises expression of one gene increases IBD risk, the next gene down that pathway should show the opposite relationship, and so on, giving increasing confidence that the pathway itself, not just an isolated gene, is the right target.
ATG4C has become the team's clean illustration of this convergence: a coding mutation, an independent rare-variant burden, and an eQTL all separately implicate the same gene. Findings like this are what let the team move quickly from a raw GWAS signal to a specific, actionable biological pathway, rather than stopping at a region of the genome with no clear mechanism attached to it.
Carl doesn't expect this convergence to translate into single-gene, single-drug medicine for most patients. Thinking in terms of a liability threshold model, he expects most people with IBD to carry some degree of risk spread across most of the relevant pathways at once, rather than concentrated in any one of them. A smaller group of patients, NOD2 carriers being the clearest example, do carry an outsized share of their risk in a single pathway, but Carl treats that as the exception rather than the rule the field should design around.
That distinction is exactly what a standard polygenic risk score obscures today: two patients can land on an identical overall score while carrying completely different combinations of variants and pathways underneath it, which Carl sees as a real loss of clinically useful information. Pathway-specific polygenic risk scores, once the underlying genetic maps are complete enough to build them, could instead point toward which drug or combination of drugs a given patient is likely to need. Even so, Carl doesn't expect genetics alone to get there. The best predictive models, he argues, will need to sit alongside clinical records and multi-omic data such as the proteome, metabolome, and methylome, since IBD's roughly 50% heritability leaves a large environmental component that genetics alone cannot capture.
IBDverse was the first single-cell sequencing project Carl's lab had ever run, a striking leap for a group that, until then, had been entirely computational. Being at Sanger helped: colleagues like Sarah Teichmann, Muzlifah Haniffa, and Roser Vento-Tormo had already built deep single-cell expertise the team could draw on, along with institutional familiarity with the relevant technologies and vendors. The bigger structural change was bringing in Rebecca McIntyre as senior staff scientist to build and lead a wet lab within the group, essentially a group within the group, that shaped the project from the ground up.
One early technical problem stood out: epithelial cells are built to undergo programmed cell death, anoikis, once they detach from the basement membrane, which is exactly what happens to them during standard single-cell dissociation. McIntyre and colleague Mena Geraba had to work out a protocol that got usable, disease-relevant transcriptomes out of cells that are biologically primed to die in the process. Looking back, Carl says he never expected his lab to have a wet lab of this scale, and can no longer imagine it without one. His advice to early-career scientists facing a similar leap is to deliberately recruit people whose skills diverge from, and often exceed, their own, then give them the license and space to shape the work.
Looking five years out, Carl describes the Human Genetics Programme moving from mapping the genetic basis of disease toward predicting it, built around longitudinal multi-omics rather than genetics alone. Genetics keeps its central role in that shift precisely because of its unique ability to separate cause from effect: in models built to understand disease progression and drug response, not just susceptibility, genetic data is what helps identify which of many correlated factors are likely to be causal rather than incidental.
The rest of that 50% of IBD risk that isn't explained by genetics will need to come from elsewhere. Carl sees proteomics and methylomics, in particular, as more promising for building durable predictors than single-cell RNA sequencing, both because their signal has a longer half-life and because clinical medicine already has a more established path for translating protein measurements into practice. Single-cell RNA sequencing remains his tool of choice for understanding disease biology in detail, but building predictors at scale, he suspects, will lean more heavily on proteome and methylome data than on the kind of single-cell atlas IBDverse represents.
Resources like UK Biobank and Our Future Health have been transformative, and Carl's lab relies on them constantly, but their volunteer-based recruitment skews toward healthier participants: people who smoke and drink less, have lower BMIs, higher educational attainment, and less multimorbidity. That is close to the opposite of who Carl thinks needs precision medicine the most, patients who are passed from pillar to post through the healthcare system on repeated diagnostic odysseys with complex, debilitating disease.
Some groups are effectively excluded from these cohorts altogether: very sick elderly patients, where clinicians are weighing whether someone is well enough to be discharged, and children and infants. Carl's plan is to build longitudinal multi-omics cohorts sampled at the specific points where these patients actually interact with the healthcare system, in conversations already underway with clinicians at Addenbrooke's and other NIHR biomedical research centers. Notably, he wants to organize these cohorts around clinical populations facing real challenges, such as frail elderly patients, rather than starting from a single named disease, since groups like that are often heterogeneous by diagnosis but share the same urgent need for better data.
Carl Anderson's conversation traces IBDverse from a single 2017 eQTL finding to a single-cell atlas spanning 89 cell types and roughly 300 people, and from there to a broader argument that the genetics of complex disease converges on shared pathways rather than isolated genes. That convergence reframes what precision medicine for IBD could look like: not a single drug matched to a single mutation, but pathway-specific risk scores that combine genetics with clinical and multi-omic data to guide which drug, or combination, a given patient actually needs.
The next chapter of that work, Carl says, is extending the same longitudinal, multi-omic approach beyond IBD to the sickest, most complex patients that volunteer biobanks systematically miss, built around real points of contact with the healthcare system rather than around any single disease label. He is currently recruiting faculty and researchers at the Wellcome Sanger Institute to help build it, and is especially keen to hear from statisticians, machine learning researchers, clinicians, and omics scientists, not only human geneticists.
Listen to the full episode below.