Share
Last week, insitro’s work on IRS1 was recognized as a Poster of Distinction at The Liver Meeting 2025. But the real story isn’t about recognition. It’s about learning to see what was always there, hiding in plain sight. For the hundreds of millions living with MASLD worldwide, this discovery represents a fundamental shift in how we understand and approach their disease.
Hiding in Plain Sight
Imagine a disease affecting more than one in three people yet remaining functionally invisible to the very science designed to understand it. MASLD –– once called Non-Alcoholic Fatty Liver Disease (NAFLD), now renamed Metabolic Dysfunction-associated Steatotic Liver Disease in an effort to strip away stigma –– is one such condition. It advances quietly, accumulating fat in the liver while the body continues its daily routines, unaware that a crisis is building cell by cell, year by year.
MASLD affects 38% of the global population. For roughly 20% of those patients, the disease progresses to MASH, which is on track to become the leading indication for liver transplantation in the U.S. Fibrosis hardens into cirrhosis, an irreversible scarring of the liver that destroys its structure and function. From there, the paths diverge toward two catastrophic endpoints: complete liver failure, requiring a transplant, or hepatocellular carcinoma, a deadly form of liver cancer.
The Genomic Gap: A Puzzling Lack of Insight
For decades, MASLD presented medicine with a peculiar paradox: ubiquitous yet obscure. We knew genes played a significant role –– the disease clearly ran in families –– but traditional genetic studies had identified only a handful of culprits. PNPLA3, the most famous, explained just a fraction of risk.
The constraint was phenotyping. To confirm MASH progression requires a liver biopsy: invasive, expensive, and prone to sampling error. You can’t build genetic maps from hundreds of thousands of people when your measurement tool requires piercing organs with needles. The signal was there. We simply couldn’t hear it over the noise of our own methodological limitations.
This “genomic gap” wasn’t just academically frustrating. It meant drug development operated largely in the dark, targeting downstream consequences like inflammation and fibrosis rather than the primary cause of fat accumulation itself, a process that begins years or even decades before severe scarring.
The Machine Learning Leap: Generating Phenotypes at Scale
The path forward required reconceptualizing the measurement problem. Rather than waiting for decades of expensive biopsies to accumulate, we used machine learning to transform routine clinical data into high-quality phenotypes.
insitro’s ClinML platform analyzed 263,000 UK Biobank participants across three data modalities: MRI scans, DXA imaging (typically used for bone density but revealing fat distribution patterns through computational insight), and a combined biomarker stream integrating over 200 plasma metabolites with anthropometric and biochemical measures from standard blood panels. These data streams were integrated to produce a single, highly accurate estimate of liver fat content –– a “digital biopsy” that far outstripped the handful of patients who had received clinical diagnoses.
The results shifted what we could see. When we ran genome-wide association studies (GWAS) using this ML-derived phenotype, the genomic gap narrowed considerably. Instead of the 15 or so previously known loci, we identified 480 unique, independent genetic loci associated with liver fat accumulation. The comprehensive genetic architecture was always present; we simply needed better instruments to observe it.
Targeting the Root Cause: Finding IRS1 in the Noise
Among hundreds of implicated genes, Insulin Receptor Substrate 1 (IRS1), emerged with particular clarity. IRS1 acts as a crucial signal amplifier in the cascade triggered when insulin docks onto liver cells. Genetic evidence from the UK Biobank showed clear dose-response relationships: more deleterious IRS1 variants correlated with higher liver fat and elevated liver enzymes, including AST and ALT. But causal evidence from genetics, while providing crucial conviction, is only part of the story. Systematic validation of the biological mechanism is also critical.
Our CellML platform validated IRS1 causality in HepG2 cells and primary human hepatocytes using CRISPR-based genetic perturbations. CRISPR knockout of IRS1 resulted in significantly reduced lipid accumulation, while CRISPR activation increased lipid droplet formation. Fluorescence microscopy demonstrated clear effects across increasing doses of free fatty acids, glucose, and insulin. Hepatocytes with suppressed IRS1 showed sparse, small lipid droplets, while those with activated IRS1 exhibited dense, large lipid accumulation consistent with steatosis.
This cellular validation confirmed that IRS1 is not a passenger in MASLD pathology but a driver. In the context of MASH, the formation of new fats – known as de novo lipogenesis (DNL) can be an overactive, pathogenic process. In patients with insulin resistance, a hallmark of metabolic syndrome and a core driver of MASLD, the body struggles to process glucose efficiently. This leads to an excessive shunting of carbohydrates and other energy sources into the DNL pathway within the liver. The liver is essentially instructed to make fat at an accelerated rate, fueling the steatosis that defines the earliest stages of MASLD.
Genetic variations or specific perturbations of the IRS1 protein directly impact the activity of this DNL pathway. By understanding how IRS1 influences the decision of a hepatocyte to initiate fat synthesis, we have identified a primary, upstream control point for the entire disease cascade. This fat accumulation isn’t merely a consequence of poor diet; it is a genetically influenced, pathologically accelerated metabolic process controlled by specific proteins like IRS1.
From Mechanism to Medicine
Our therapeutic candidate, CTRO-1013, demonstrated efficacy in preclinical MASLD models. Histopathology analysis revealed CTRO-1013 treatment reduced hepatic steatosis from extensive macrovesicular fat accumulation (characteristic “Swiss cheese” appearance in vehicle controls) to minimal lipid presence. Quantitative endpoints included a 60% reduction in hepatic triglycerides, improved MASLD Activity Score (MAS), and decreased liver enzyme levels.
CTRO-1013 was designed with a tissue-selective targeting approach, maximizing therapeutic benefit in the liver while limiting systemic exposure critical for metabolic disease therapies where off-target effects can compromise efficacy or safety. CTRO-1013 is advancing toward First-in-Human studies and work remains, but we believe that our starting point provides us with a differentiated probability of success: genetically validated target, causality confirmed in human cells, demonstrated in vivo efficacy, precision design from inception.
Systematic Validation
This isn’t hypothesis generation from correlative genetics, hoping it translates. It’s building a continuous validation chain from human genetic evidence through mechanism confirmation to therapeutic candidates with demonstrated efficacy.
For years, drug development was hampered by insufficient genetic insight into disease mechanisms. MASLD affects 38% of the global population with limited treatment options and rising burden. What we’ve demonstrated is a path from correlation to causation to intervention –– systematic, validated, human-biology-first.
For the hundreds of millions of patients living with MASLD, this research offers more than just a new drug candidate. It validates a fundamental shift in the drug discovery paradigm. We are moving from an era of serendipitous findings to one of predictive design, where machine learning decodes the noisy complexity of human biology to reveal high-confidence targets. IRS1 is not just a highly compelling mechanism; it is a proof point that when we view disease through the lens of high-resolution data, we can engineer medicines with the precision that patients deserve.
What Comes Next
→ Advancing to the Clinic: We are progressing CTRO-1013 through IND-enabling studies and preparing for First-in-Human clinical trials.
→ Deepening Biological Insight: We continue to explore the underlying biology that IRS1 modulates to better understand its role in other critical aspects of liver pathology and metabolic disease.
→ Scaling the Platform: As we advance our pipeline across metabolism, neuroscience, and other therapeutic areas, insitro’s predictive AI models will continuously learn and improve.
Thank you to the many insitrocytes who contributed to this work, including:
Lead scientists: Tanaya Walimbe, PhD and Hari Somineni, PhD
Project conceived and led by: David Lloyd, PhD; Colm O’Dushlaine, PhD; Santhosh Satapati, PhD; Arijit Bhowmick, PhD; Chief Scientific Officer Philip Tagari, and Chief Medical Officer S. Mike Rothenberg, MD, PhD
We are proud of this interdisciplinary team of data scientists, geneticists, biologists, chemists, and drug developers whose collaboration makes this program possible.
