The study, published in Clinical Cancer Bulletin, employed a technique called proteome-wide Mendelian randomization, an epidemiological approach that exploits the random shuffling of genes at conception as a natural experiment. Because genetic variants are inherited independently of lifestyle and environmental factors, they can act as proxies, or instrumental variables, for protein levels in the blood. If people who carry variants that raise the circulating concentration of a particular protein also tend to develop lung squamous cell carcinoma more often, that suggests, though it does not prove, that the protein itself plays a causal role in the disease rather than merely riding along as a byproduct of tumor growth.

To build this genetic proxy system, the researchers drew on two of the largest protein-genetic datasets ever assembled: the UK Biobank Pharma Proteomics Project, which profiled 2,958 proteins in 54,306 participants using the Olink platform, and the deCODE Health study, which measured 4,906 proteins in 35,559 Icelanders using the SomaScan platform. After filtering out proteins lacking usable genetic instruments and excluding the notoriously tangled major histocompatibility complex region of chromosome 6, the team arrived at 1,046 proteins with shared protein quantitative trait loci, genetic variants reliably associated with plasma protein levels.

For the disease side of the equation, the investigators used summary statistics from the International Lung Cancer Consortium, comprising 3,275 lung squamous cell carcinoma cases and 15,038 controls of European ancestry, as their discovery dataset. They then replicated findings in the FinnGen study, with 1,510 cases and 314,193 controls, and in a separate genome-wide association study led by McKay and colleagues involving 7,426 cases and 55,627 controls. Applying a stringent Bonferroni correction for the more than a thousand proteins tested, the analysis converged on six proteins with statistically robust associations: HSPA1L, PCSK7, POLI, SPINK2, TCL1A, and VARS.

Three of these proteins emerged as protective. Higher genetically predicted levels of HSPA1L were associated with roughly half the odds of developing lung squamous cell carcinoma, with an odds ratio of 0.47. SPINK2 showed an odds ratio of 0.68, and VARS an even more striking 0.44, meaning each standard deviation increase in circulating levels of this valyl-tRNA synthetase corresponded to a 56 percent reduction in disease odds. In the combined meta-analysis across all three disease datasets, HSPA1L, SPINK2, and VARS remained consistently protective, with combined odds ratios of 0.51, 0.71, and 0.48 respectively and p-values approaching ten to the minus ten, a level of statistical significance rarely seen in observational cancer epidemiology.

The other three proteins pointed in the opposite direction. PCSK7, a proprotein convertase previously implicated in activating matrix metalloproteinases that facilitate tumor invasion, carried an odds ratio of 1.37. POLI, a low-fidelity DNA polymerase from the Y-family known to promote migration and invasion of lung cancer cells, showed the largest effect of all, with an odds ratio of 4.50. TCL1A, an oncogene that modulates the Akt/mTOR signaling pathway crucial for cell growth and survival, was associated with an odds ratio of 1.72. However, the replication picture for these three was more complicated: PCSK7 and TCL1A displayed significant heterogeneity across datasets, with combined estimates that no longer reached statistical significance, while POLI yielded wide confidence intervals spanning both harm and neutrality.

To guard against false positives arising from linkage disequilibrium, the phenomenon in which nearby genetic variants are inherited together and can masquerade as causal signals, the researchers deployed a battery of complementary analyses. Summary-data-based Mendelian randomization and the heterogeneity in dependent instruments, or HEIDI, test helped distinguish genuine protein-disease relationships from shared genetic architecture. Bayesian colocalization analysis assigned posterior probabilities to competing hypotheses about whether a protein association and a disease association in the same genomic region stem from one causal variant or two distinct ones. Notably, TCL1A passed both the HEIDI test and the colocalization threshold, with a posterior probability of hypothesis four, or PPH4, of 0.817, indicating that the same variant likely drives both TCL1A protein levels and lung squamous cell carcinoma risk. Sensitivity analyses including Cochran’s Q test for heterogeneity, the MR-Egger intercept test for horizontal pleiotropy, and MR-PRESSO outlier detection found no evidence that the primary associations were artifacts of pleiotropic genetic effects.

The biological stories behind these proteins add plausibility to the statistical findings. HSPA1L, a heat shock protein, has been shown in laboratory studies to enhance cancer stem cell-like properties by activating the insulin-like growth factor 1 receptor beta and regulating beta-catenin transcription, pathways central to tumor progression and therapy resistance. SPINK2, a serine protease inhibitor, is known to influence protease activity and intercellular communication within the tumor microenvironment, and elevated expression has been linked to poor prognosis in acute myeloid leukemia. VARS belongs to the aminoacyl-tRNA synthetase family, enzymes best known for charging transfer RNAs with amino acids during protein synthesis but increasingly recognized for non-canonical roles in signal transduction and cellular homeostasis that can promote cancer cell survival and proliferation.

The authors are candid about the limitations of their approach. The Mendelian randomization framework rests on the assumption that genetic instruments influence disease only through the protein of interest, and while the team excluded trans-associated loci, applied Steiger filtering, and ran extensive sensitivity analyses, horizontal pleiotropy can never be fully eliminated. Low colocalization posterior probabilities for some proteins, such as HSPA1L and VARS, may reflect limited statistical power rather than true absence of a shared causal variant, and interpreting these values requires caution. Moreover, all data came from individuals of European ancestry, raising questions about generalizability to other populations, and the reliance on pre-existing databases rather than real-world cohorts leaves room for population-specific biases.

Nevertheless, the study represents the first systematic proteome-wide causal scan for lung squamous cell carcinoma, a cancer that accounts for roughly 30 percent of non-small cell lung cancers and carries a five-year survival rate below 20 percent for advanced stages. Unlike previous Mendelian randomization efforts that focused on lung adenocarcinoma or on mitochondrial proteins, this work delivers a subtype-specific map of circulating proteins that may shape disease risk. If future functional studies and diverse real-world cohorts validate these findings, HSPA1L, SPINK2, and VARS could inspire protective therapeutic strategies, while PCSK7, POLI, and TCL1A, particularly TCL1A with its strong colocalization evidence, could become targets for drugs aimed at intercepting the disease before it takes hold. For a cancer with so few targeted options, six genetically anchored leads represent a meaningful step forward.

Subject of Research: Proteome-wide Mendelian randomization of plasma proteins as causal risk factors for lung squamous cell carcinoma

Article Title: Causal associations of plasma proteins with lung squamous cell carcinoma risk: a proteome-wide Mendelian randomization and colocalization analysis

Article References: Wang, Q., Xue, X., Ling, X., Lang, Y., Wang, S., & Liu, G. (2024). Causal associations of plasma proteins with lung squamous cell carcinoma risk: a proteome-wide Mendelian randomization and colocalization analysis. Clinical Cancer Bulletin, 3(1), Article 18. https://doi.org/10.1007/s44272-024-00024-w

Image Credits: AI Generated

DOI: 10.1007/s44272-024-00024-w

Keywords: lung squamous cell carcinoma, Mendelian randomization, proteomics, plasma proteins, colocalization analysis, HSPA1L, SPINK2, VARS, TCL1A, PCSK7, POLI, pQTL