Acute lymphoblastic leukemia (ALL) is the most common cancer in children, accounting for roughly 30% of all childhood cancers and about 80% of all childhood leukemias. It originates from immature T- or B-lymphoid cells in the bone marrow that multiply uncontrollably instead of developing into normal immune cells.
Modern treatment protocols have achieved impressive results: approximately 85% of children with ALL are cured. However, more than 10% still experience an unfavorable outcome, and 1-3% of pediatric ALL deaths are caused not by the disease itself but by the toxic side effects of the treatment - a sobering statistic that drives the search for better, more targeted therapies.
Personalized medicine aims to tailor treatment to each individual patient based on their unique genetic profile. Rather than using one-size-fits-all dosing, doctors could one day adjust which drugs are given, at what doses, and for how long - based on a patient's DNA. This is the promise that pharmacogenomics and pharmacotranscriptomics hold for childhood ALL.
This review paper from Serbian researchers summarizes the latest evidence from high-throughput genomic technologies - including microarrays and next-generation sequencing - that are mapping out the genetic factors controlling how individual children respond to the six major classes of ALL drugs.
Pharmacogenomics studies how differences in a patient's DNA sequence - specifically in genes involved in drug metabolism and response - affect whether a drug works and how toxic it is. Every person inherits a slightly different version of thousands of genes, and some of these differences directly change how quickly the body breaks down a medication or how sensitive cancer cells are to it.
Pharmacotranscriptomics goes a step further by examining not just the DNA blueprint, but the actual RNA molecules that genes produce (the transcriptome). Even patients with identical DNA sequences can differ in how actively their genes are being read and translated, and these differences in gene activity can also predict drug response.
Researchers use two main strategies to find relevant genetic markers. The candidate gene approach tests specific genes already suspected of playing a role in drug metabolism - it is statistically powerful but limited to what scientists already know. Genome-wide association studies (GWAS) scan hundreds of thousands of DNA variants across the entire genome simultaneously, which can uncover entirely unexpected genes but requires very large patient groups to reach statistical significance.
Children with ALL actually have two distinct genomes relevant to treatment: their normal constitutional genome in all healthy cells (which controls drug metabolism and side effects), and the acquired tumor genome of the leukemia cells themselves (which controls drug resistance). Both must be considered when designing personalized therapy.
Glucocorticoid drugs such as prednisone and dexamethasone form the backbone of childhood ALL treatment. They work by activating the glucocorticoid receptor (GR) inside lymphoblast cancer cells, triggering a protein called Bim that causes the cell to self-destruct (a process called apoptosis). However, response to these drugs varies considerably between patients.
Multiple genes have been identified that influence steroid response. Variants in the NR3C1 gene (which encodes the glucocorticoid receptor itself) affect how strongly the receptor responds. A haplotype (combination of variants) called ACT in NR3C1 is strongly associated with glucocorticoid sensitivity. Similarly, a variant in the ABCB1 gene - which encodes a pump that ejects drugs from cells - has been linked to poor glucocorticoid response and increased relapse risk.
Large-scale GWAS studies scanning 440,000 DNA variants in over 2,500 children found four new variants associated with faster clearance of dexamethasone from the body, meaning some patients metabolize the drug so quickly it becomes less effective. Two of these variants were in the ABCB1 gene, confirming its importance across different research approaches.
One of the most serious side effects of steroid treatment is osteonecrosis - the death of bone tissue, particularly in the hips and knees. GWAS studies found that a variant (rs10989692) near the brain receptor gene GRIN3A, and variants in the ACP1 gene (involved in lipid metabolism), both increase the risk of this debilitating complication during dexamethasone therapy. Additionally, a non-coding RNA called GAS5 was found to block the glucocorticoid receptor, contributing to steroid resistance in some children.
Vincristine is a plant-derived drug used in nearly all childhood ALL treatment protocols. It kills cancer cells by disrupting the scaffolding (microtubules) that cells need to divide. However, this same mechanism can damage the nerve fibers of healthy cells, causing a painful condition called peripheral neuropathy - numbness, tingling, and weakness in the hands and feet - which sometimes forces doctors to reduce or stop the treatment.
The CYP3A5 gene encodes a liver enzyme that breaks down vincristine. Patients carrying two non-functional copies of the CYP3A5*3 allele have essentially no CYP3A5 enzyme, meaning vincristine stays in their bodies longer and causes more severe nerve damage. This is one of the clearest examples of a pharmacogenomic marker directly explaining a side effect.
A landmark GWAS discovered that a variant (rs924607) in the CEP72 gene - which encodes a protein essential for building microtubules - significantly increases the risk of vincristine-related nerve damage. The risk genotype creates a binding site that represses CEP72 gene activity, destabilizing the very microtubule network that vincristine targets, making nerve cells hypersensitive to the drug's effects.
More recently, whole-exome sequencing identified variants in SYNE2 and MRPL47 genes as risk factors for high-grade nerve toxicity. The MRPL47 gene is particularly interesting - it plays a role in the energy production of cells (mitochondrial function), and impaired energy production is known to make nerve cells more vulnerable to damage. A protective variant in the BAHD1 gene (involved in gene silencing) was also found to reduce neuropathy risk.
MicroRNAs - tiny regulatory molecules that control gene expression - also play a role. A combination of miR-125b with miR-99a or miR-100 creates a synergistic resistance to vincristine in certain ALL subtypes, meaning the cancer cells become harder to kill even as normal nerve cells remain vulnerable to toxicity.
Asparaginase is a unique cancer drug - it is an enzyme (originally derived from bacteria) that depletes the amino acid asparagine from the bloodstream. Normal cells can make their own asparagine, but leukemia cells cannot, so they starve and die when the supply is cut off. It is a cornerstone of childhood ALL therapy, but it causes severe side effects in many patients.
The most dangerous complications of asparaginase include life-threatening allergic reactions, pancreatitis (inflammation of the pancreas), blood clotting abnormalities, and liver damage. Understanding which children are genetically predisposed to these complications could allow doctors to switch to safer alternative formulations (such as PEGylated asparaginase or Erwinia-derived asparaginase) before a crisis occurs.
A major GWAS discovery found that the HLA-DRB1*07:01 allele - a variant in the immune recognition system - is strongly associated with asparaginase hypersensitivity. The HLA system controls how the immune system distinguishes between self and foreign proteins. This particular variant appears to cause the immune system to recognize asparaginase as a dangerous foreign invader and mount an allergic response. This finding was validated across multiple independent patient populations from Europe and beyond.
For pancreatitis specifically, GWAS studies identified variants in the PRSS1-PRSS2 locus (genes encoding digestive enzymes called trypsins) and in the NFATC2 gene (an immune regulator) as risk factors. The mechanism appears similar to alcohol-induced or gallstone-induced pancreatitis - inappropriate activation of digestive enzymes within the pancreas itself. Testing for these variants could identify high-risk patients before treatment begins.
A variant (rs738409) in the PNPLA3 gene, which controls fat metabolism in the liver, was found to increase the risk of hepatotoxicity (liver damage) during asparaginase therapy. This variant causes fat to accumulate in liver cells, making them more vulnerable to drug-induced injury - a finding confirmed in both human patients and mouse models.
Anthracyclines (including doxorubicin and daunorubicin) are powerful chemotherapy drugs that work by disrupting the enzyme that untangles DNA during cell division, causing cancer cells to die. They are essential components of childhood ALL treatment, but their use is limited by a serious side effect: cardiotoxicity, or damage to the heart muscle.
Anthracycline-induced heart damage can appear immediately after the first dose or emerge years or even decades later as cardiomyopathy (weakened heart muscle). Childhood cancer survivors who received anthracyclines as children face lifelong elevated risks of heart failure - making it critical to identify genetic risk factors so doses can be adjusted or protective medications prescribed.
A key GWAS discovery identified a variant (rs2229774) in the RARG gene - a gene encoding a retinoic acid receptor that is particularly active in heart tissue - as highly associated with anthracycline cardiotoxicity. This variant disrupts a pathway that normally protects heart cells from anthracycline damage, and its discovery suggested potential protective strategies involving retinoic acid signaling.
The SLC28A3 gene encodes a transporter that moves anthracycline drugs into cells. A specific variant (rs7853758) in this gene reduces anthracycline entry into heart cells and actually has a protective effect against cardiotoxicity - a striking example of how the same genetic variant that reduces drug uptake (potentially a problem for cancer cells) can simultaneously protect the heart.
The HAS3 gene encodes an enzyme that produces hyaluronan, a key component of the structural scaffolding that surrounds cells. Patients with a specific HAS3 genotype (AA at rs2232228) showed dose-dependent increases in cardiomyopathy risk as anthracycline doses increased, while patients with the GG genotype appeared protected regardless of dose. This suggests that tissue repair capacity after drug-induced heart injury varies genetically between patients.
Thiopurine drugs - particularly 6-mercaptopurine (6-MP) - are given during the maintenance phase of childhood ALL therapy, typically for 2-3 years. They work by incorporating toxic analogs into DNA, causing cancer cells to die. The key challenge is that the optimal dose varies enormously between patients - too little allows leukemia to return, while too much causes severe bone marrow suppression and life-threatening infections.
The enzyme TPMT (thiopurine S-methyltransferase) inactivates thiopurine drugs by methylation. Patients with two non-functional TPMT gene copies have very low enzyme activity and accumulate dangerous drug levels even at standard doses. The TPMT gene-drug pair is one of the first and most thoroughly validated examples in all of pharmacogenomics - TPMT testing before starting thiopurine therapy is now a clinical standard in many countries.
A major GWAS breakthrough discovered a second critical gene: NUDT15. Variants in NUDT15 (particularly rs116855232) were found to be especially important in East Asian patients, who have a lower tolerance for thiopurines than European patients despite having fewer non-functional TPMT alleles. Patients with two non-functional NUDT15 copies could only tolerate about 10% of the standard 6-MP dose. Based on this overwhelming evidence, NUDT15 testing is now also recommended before starting thiopurine therapy.
Somatic mutations (acquired mutations in cancer cells rather than inherited) in the NT5C2 gene were found to cause relapse in childhood ALL patients on thiopurine maintenance therapy. These mutations activate an enzyme that inactivates thiopurine metabolites, allowing leukemia cells to escape the drug's effects. These mutations were particularly associated with early relapse, underscoring their importance as prognostic biomarkers.
Methotrexate (MTX) is one of the most widely used drugs in childhood ALL and is given at all stages of treatment - from initial therapy through maintenance and even directly into the spinal fluid to prevent brain relapse. It works by blocking the folate pathway, which cancer cells need to make DNA building blocks. Without folate, cancer cells cannot replicate.
The speed at which the body clears methotrexate from the blood directly affects both its effectiveness and its toxicity to normal tissues. A landmark GWAS study found that variants in the SLCO1B1 gene - encoding a transporter in the liver that removes drugs from the bloodstream - account for approximately 10% of the variation in MTX clearance between patients. This finding was replicated in an independent cohort of 1,300 children with ALL, making it one of the most robustly validated pharmacogenomic findings in pediatric oncology.
The expression levels of several key enzymes also predict MTX response. DHFR (dihydrofolate reductase) is the primary target of methotrexate, and patients whose cancer cells express high levels of DHFR can overcome the drug's blockade and have poorer survival. Conversely, FPGS (folylpolyglutamate synthase) activates MTX inside cells by adding molecular chains that trap the drug inside - patients with high FPGS activity accumulate more active drug and have better survival.
The subtype of ALL matters enormously for MTX response. B-cell ALL patients are generally more sensitive to MTX than T-cell ALL patients, because their cancer cells express more FPGS and less DHFR. Children with hyperdiploid ALL (cancer cells with extra chromosomes) are particularly sensitive to low-dose MTX because they carry extra copies of chromosome 21, where the SLC19A1 drug transporter gene is located, increasing drug uptake. These subtype-specific differences require tailored dosing strategies rather than uniform protocols.
Microarrays were the dominant technology for large-scale pharmacogenomic studies over the past two decades. These devices allow researchers to simultaneously measure hundreds of thousands of known DNA variants (single nucleotide polymorphisms, or SNPs) across the entire genome in a single experiment. While powerful, microarrays can only detect variants that were anticipated and included on the chip - they cannot find entirely new types of genetic alterations.
Next-generation sequencing (NGS) technologies - including whole-genome sequencing (WGS), whole-exome sequencing (WES), and targeted gene panel sequencing - read the actual DNA sequence letter by letter and can discover entirely new variants that were not previously known or anticipated. The cost of sequencing has dropped dramatically, making these approaches increasingly practical for large clinical studies.
For pharmacotranscriptomics specifically, RNA sequencing measures the activity level of every gene simultaneously by counting the messenger RNA (mRNA) molecules in a cell. Unlike microarrays (which measure pre-selected genes), RNA-seq provides an unbiased picture of the entire transcriptome. However, RNA-seq studies in pediatric ALL pharmacogenomics remain relatively rare compared to DNA-based approaches.
A critical challenge with all high-throughput studies is the statistical burden of testing millions of variants simultaneously. With so many tests, some will appear significant by chance alone (false positives). The standard solution is the very stringent Bonferroni correction, which requires enormously large patient cohorts to achieve adequate statistical power - a particular challenge in rare pediatric cancers where patient numbers are inherently limited.
Despite the enormous volume of pharmacogenomic data generated over two decades, translation into clinical practice remains limited. Many promising genetic markers have been discovered but not yet validated in independent patient populations or integrated into treatment protocols. The gap between laboratory discovery and actual changes in how children are treated remains the central challenge in this field.
The authors propose a concrete roadmap: use data mining to identify the most robustly validated genetic markers, create custom genomic testing panels combining DNA variants and gene expression signatures, then feed all this molecular data - alongside clinical information - into machine learning algorithms to build predictive models of drug toxicity and treatment response.
Such artificial intelligence models would be trained on large groups of carefully characterized ALL patients. Once validated, they could prospectively identify which children are at high risk of severe drug toxicity before treatment begins - allowing doctors to make preemptive dose adjustments or drug substitutions in randomized clinical trials, the gold standard for changing medical practice.
Population diversity is an important consideration: pharmacogenomic variant frequencies differ substantially between ethnic groups. Databases like FINDbase are being built to capture population-specific data, ensuring that personalized medicine tools work equally well for children of all ethnic backgrounds. Studies have documented significant pharmacogenomic differences across European, East Asian, and other populations for key ALL drugs.
The field is advancing rapidly: over 3,500 gene-drug associations have been validated with strong evidence, and more than 200 drugs now carry FDA labels recommending or mandating pharmacogenomic testing. For childhood ALL specifically, TPMT and NUDT15 testing before thiopurine therapy is already standard practice in leading centers - a proof of concept that genomics-guided treatment is achievable and clinically meaningful.