Clear cell renal cell carcinoma (ccRCC) makes up about 80% of all kidney cancer cases. Even when caught early, nearly one in five patients already has metastatic disease at diagnosis, and 25 to 40% of patients with localized disease eventually experience recurrence after surgery. The 5-year survival rate for advanced ccRCC is critically low at 11.7%, making better tools for predicting outcomes and guiding treatment urgently needed.
One important but underappreciated driver of kidney cancer progression is lactate, a molecule that cancer cells produce in large amounts through a process called the Warburg effect. In this process, cancer cells consume far more glucose than normal and convert it to lactate even when oxygen is available. This excessive lactate production was historically dismissed as metabolic waste, but it is now understood to be an active signaling molecule that reshapes the environment surrounding the tumor.
Elevated lactate in the tumor microenvironment (TME) creates an acidic, immunosuppressive niche that helps cancer cells hide from and resist the immune system. High lactate levels directly suppress the activity of cytotoxic T cells that would normally kill cancer cells, promote immunosuppressive M2 macrophage formation, impair natural killer cell function, and expand myeloid-derived suppressor cells that block anti-tumor immunity. Essentially, lactate builds a protective shield around the tumor.
Understanding how lactate metabolism and immune cell activity interact in ccRCC could reveal new ways to predict which patients need more aggressive treatment and which patients are most likely to benefit from immunotherapy. This study built the LAC-TME classifier, a machine learning model that combines lactate-related gene signatures with tumor immune cell profiles to create a powerful new tool for patient stratification and treatment guidance.
The researchers used gene expression and clinical data from two ccRCC patient cohorts: the TCGA-KIRC training cohort (522 patients) and the E-MTAB-1980 validation cohort (101 patients). Comparing gene expression between tumor and normal kidney tissue identified 2,714 differentially expressed genes, of which 16 were also related to lactate metabolism. These 16 lactate-related genes formed the starting point for building the classifier.
To find the best way to combine these genes into a survival prediction score, the researchers tested 101 different machine learning algorithm combinations using the Mime1 package, including StepCox, Random Survival Forest (RSF), Lasso, CoxBoost, Gradient Boosting, and others. Model performance was measured using the concordance index (C-index), which measures how well the model correctly ranks patients by their survival time.
The immune component of the model was built separately. Using the xCell algorithm, the researchers measured the abundance of 64 different immune and stromal cell types within each patient's tumor. The same 101 machine learning algorithm framework was then applied to these cell type scores to identify which immune cell populations were most relevant for predicting survival. The StepCox combined with Random Survival Forest model produced the best C-index for both the lactate gene signature and the immune cell profile.
The final LAC-TME classifier was created by combining the lactate gene risk score and the immune TME risk score. Patients were first divided into Lactatehigh and Lactatelow groups and TMEhigh and TMElow groups. These were then combined into three prognostic subgroups: Lactatelow + TMElow (best prognosis), Lactatehigh + TMEhigh (worst prognosis), and a Mixed intermediate group.
The LAC-TME classifier achieved a C-index of 0.92 in the training set and 0.73 in the independent validation cohort. The area under the curve (AUC) for predicting 1-year survival was 0.88, rising to 0.90 for 3-year survival and 0.92 for 5-year survival, demonstrating that the classifier becomes more accurate the further into the future it predicts.
Patients in the Lactatehigh + TMEhigh subgroup had the worst survival outcomes across both the training and validation cohorts. This group was characterized by high tumor mutational burden, more frequent deletions of tumor suppressor genes (including CDKN2A and PTEN), and higher BAP1 mutation rates, all of which are associated with aggressive disease. In contrast, Lactatelow + TMElow patients had the longest survival and the fewest genomic disruptions.
Pathway analysis revealed striking biological differences between the two extreme groups. The Lactatehigh + TMEhigh group showed enrichment in cancer-promoting pathways including inflammation, cell cycle dysregulation, IL6-JAK-STAT3 signaling, and epithelial-mesenchymal transition (a process that allows cancer cells to become more mobile and invasive). Meanwhile, the Lactatelow + TMElow group showed enrichment in mitochondrial metabolic pathways associated with more normal cellular energy production.
The classifier maintained its prognostic accuracy across patient subgroups defined by age, sex, tumor grade, and tumor stage, indicating that the LAC-TME classification captures biologically meaningful information independent of conventional clinical factors. This is important for clinical utility, as a tool that works across diverse patient populations is more likely to be broadly applicable in real-world hospital settings.
One of the most clinically important findings is that the Lactatehigh + TMEhigh subgroup, despite having more immune cells present in the tumor (higher TME score), is associated with immunosuppression and a poor predicted response to immunotherapy. This paradox occurs because the lactate-driven immune environment is not an active, cancer-fighting immune response but rather a dysfunctional one where immune cells are present but suppressed.
Using the TIDE algorithm, which estimates how likely a patient's immune system is to respond to immunotherapy, the researchers found that Lactatelow + TMElow patients were more likely to be immunotherapy responders. Knowing this before treatment begins could help oncologists avoid exposing immunotherapy-resistant patients to the significant toxicities of these drugs while prioritizing them for alternative treatment approaches.
Drug sensitivity analysis using the GDSC database predicted differential responses to several standard ccRCC therapies. The study found differences in predicted sensitivity to Sunitinib, Temsirolimus, Rapamycin, and Sorafenib across the three LAC-TME subgroups. These predictions suggest that the LAC-TME classifier could help guide not just immunotherapy decisions but also selection among targeted therapy drugs.
The core lactate-related gene driving the classifier, LGALS1 (also known as Galectin-1), showed the highest importance score in the model. LGALS1 is known to promote immune evasion and tumor progression in multiple cancer types. Laboratory validation confirmed that when LGALS1 was silenced in kidney cancer cells, both cell proliferation and migration were significantly reduced, directly linking the classifier's top gene to cancer cell behavior.
To validate the biological significance of the top gene in the LAC-TME model, the researchers silenced LGALS1 in two ccRCC cell lines (786-O and 769-P) using small interfering RNA (siRNA). This approach reduces the amount of functional LGALS1 protein in cancer cells, mimicking what would happen if a drug successfully inhibited this gene's activity.
When LGALS1 was silenced, kidney cancer cells showed a significant reduction in their ability to proliferate, as measured by CCK-8 viability assays over four consecutive measurements. Fewer cells survived and continued dividing, confirming that LGALS1 is necessary for active kidney cancer cell growth.
LGALS1 silencing also significantly reduced cancer cell migration and invasion, as measured by Transwell migration assays, scratch wound assays, and Matrigel invasion assays. These functional tests specifically measure the cancer cells' ability to move through barriers and invade surrounding tissue, the cellular processes responsible for cancer spread. Reduced invasion is a particularly meaningful finding because metastasis is what makes kidney cancer life-threatening.
These laboratory findings provide direct biological support for the computational findings: LGALS1 is not merely a statistical predictor of survival but is an active contributor to cancer cell growth and invasive behavior. This makes LGALS1 a compelling target for future drug development, and its role as the anchor gene in the LAC-TME classifier strengthens the biological credibility of the entire prognostic model.
The LAC-TME classifier represents an important advance in kidney cancer precision medicine. By integrating two distinct biological dimensions of tumor biology, lactate metabolism and immune microenvironment composition, into a single machine learning model, it captures more clinically relevant information than either component could provide alone. The result is a robust prognostic tool that works across independent patient cohorts and diverse clinical subgroups.
For patients, the most meaningful implication is that the LAC-TME classifier could help doctors choose better treatments from the start. Rather than trying immunotherapy on all advanced ccRCC patients and waiting to see who responds, this classifier could identify likely non-responders upfront, allowing oncologists to direct those patients toward alternative therapeutic strategies that are more likely to work for their specific tumor biology.
The identification and functional validation of LGALS1 as the central lactate-related gene in the model opens new therapeutic possibilities. LGALS1 inhibitors are already being investigated in preclinical cancer research, and these findings provide a rationale for exploring LGALS1-targeted therapy specifically in ccRCC, particularly in the high-risk Lactatehigh + TMEhigh patient subgroup.
While prospective clinical validation will be needed before the LAC-TME classifier can be formally adopted in patient care, this study demonstrates that combining metabolic gene signatures with immune profiling through advanced machine learning is a productive approach for building clinically useful cancer prognostic tools. The framework could also be adapted for other cancer types where lactate-driven immune suppression is thought to play a significant role.