Genomes are often imagined as static strings of text, where mutations are merely single-letter typos. In reality, genomes undergo massive structural overhauls: entire segments of DNA spanning thousands or millions of base pairs can be duplicated or deleted. These structural alterations, known as Copy Number Variations (CNVs) or Copy Number Alterations (CNAs), alter gene dosage and serve as major drivers of genetic disorders, autism, and cancer.
Detecting these variations requires moving beyond simple sequencing to measure the relative abundance of genomic regions. Over the past few decades, the methodology has evolved from physical color competition on chromosomes to digital counting in sequencing data and statistical smoothing in single-cell transcriptomics. Understanding why these methods work requires looking at the core physical and statistical principles that underpin them.
The Mechanics of Competitive Hybridization
The earliest challenge in genome-wide CNV detection was establishing a reliable baseline. If you look at a patient's tumor DNA in isolation, it is difficult to determine whether a specific region is amplified or deleted because experimental conditions cause total DNA yields to vary.
Comparative Genomic Hybridization (CGH) solved this by turning the measurement into a competitive race. Instead of analyzing a sample alone, DNA from a test sample (e.g., a tumor) and DNA from a normal reference sample are isolated, labeled with two distinct fluorescent dyes (typically green for test and red for reference), and mixed in equal amounts. This mixture is then allowed to hybridize to a target substrate, such as a metaphase chromosome spread.
The two pools of DNA compete to bind to their complementary sequences on the substrate. If a chromosomal region is balanced between the test and reference genomes, the green and red dyes bind in equal proportions, emitting a yellow signal. If a region is amplified in the test sample, the green dye outcompetes the red, shifting the signal toward green. Conversely, a deletion in the test sample allows the red reference DNA to dominate.
As shown by Michael R. Speicher et al., this competitive framework serves as a global screening tool to map genomic chaos across the entire genome without requiring prior knowledge of specific target sequences (Analysis of Chromosomal Alterations in Non-Small Cell Lung Cancer by Multiplex-FISH, Comparative Genomic Hybridization, and Multicolor Bar Coding). This principle enabled researchers like Horst Zitzelsberger et al. to trace how chromosomal changes accumulate during tumor development and radiation-induced damage (Cytogenetic changes in radiation-induced tumors of the thyroid.; Chromosomal changes during development and progression of prostate adenocarcinomas). By measuring these competitive hybridization ratios, Yasuhiko Kimura et al. were able to pinpoint specific prognostic signatures, such as 20q gains and 18q losses, that correlate with tumor progression in gastric cancers (Genetic alterations in 102 primary gastric cancers by comparative genomic hybridization: gain of 20q and loss of 18q are associated with tumor progression).
Scaling Resolution with Microarrays
While metaphase CGH revolutionized cytogenetics, its resolution was physically limited by the structure of condensed chromosomes, making it impossible to detect alterations smaller than a few megabases. To catch microdeletions and microduplications, array CGH (aCGH) and chromosomal microarrays (CMA) replaced whole chromosomes with engineered microarrays containing thousands of defined, high-density DNA probes.
The underlying physics remained the same—competitive hybridization—but the target substrate was now a grid of precisely mapped genomic fragments. The authors of several clinical studies leveraged this increased resolution to link specific gene-dosage changes to disease states. For instance, Rabeah Al‐Temaimi et al. used aCGH to identify structural variations impacting gene dosage in colorectal cancer metastasis (Identification of 42 Genes Linked to Stage II Colorectal Cancer Metastatic Relapse), while Hongyan Nie et al. applied it to detect prenatal chromosomal deletions in congenital heart disease (Use of magnetic resonance imaging combined with gene analysis for the diagnosis of fetal congenital heart disease).
This high-resolution tracking is essential for observing phenomena like the global genome instability described by Satoshi Miuma et al., where caretaker gene deficiencies trigger widespread structural gains and losses (Fhit Deficiency-Induced Global Genome Instability Promotes Mutation and Clonal Expansion). It also allowed Xiaohong Han et al. to map prognostic copy number alterations in lung adenocarcinoma (Comprehensive Profiling of Gene Copy Number Alterations Predicts Patient Prognosis in Resected Stages I–III Lung Adenocarcinoma) and enabled Karen Woodward et al. to uncover atypical nested duplications associated with neurodevelopmental phenotypes (Atypical nested 22q11.2 duplications between LCR22B and LCR22D are associated with neurodevelopmental phenotypes including autism spectrum disorder with incomplete penetrance).
Deconvolving Mosaicism via LRR and BAF
Genotype arrays introduced a dual-signal framework that can even detect mosaicism—where a mutation exists in only a subset of a person's cells. This works by analyzing two orthogonal signals: the Log-R Ratio (LRR) and the B-allele frequency (BAF).
$$\text{LRR} = \log_2\left(\frac{\text{Observed Intensity}}{\text{Expected Intensity}}\right)$$
- Log-R Ratio (LRR): Measures the overall normalized intensity of the DNA signal at a given locus. If LRR is zero, the region is diploid. If it is positive, there is a gain; if negative, a loss.
- B-allele Frequency (BAF): Measures the relative proportion of parental alleles (A and B) at heterozygous sites. In a normal diploid cell, the BAF clusters cleanly at 0 (AA), 0.5 (AB), or 1 (BB).
If a deletion occurs across all cells, the 0.5 (AB) cluster disappears entirely because the cell is left hemizygous (only A or B remains). However, if a mosaic duplication occurs in only a fraction of the cells, the AB cluster splits into intermediate frequencies (such as 0.66 and 0.33) proportional to the percentage of cells carrying the variant. As demonstrated by Maxwell A. Sherman et al., analyzing these joint deviations allows computational callers to identify massive, post-fertilization mosaic rearrangements that contribute to autism risk but are invisible to standard analysis (Large mosaic copy number variations confer autism risk).
Decoding Copy Numbers from Sequencing Depth
With the advent of next-generation sequencing (NGS), physical hybridization competition was largely replaced by digital counting. The core intuition of read depth analysis is straightforward: if genomic DNA is fragmented and sequenced randomly, the number of reads that map to a specific chromosomal window is directly proportional to the number of copies of that window present in the cell.
As explained by Chaima Aouiche et al., significant deviations from expected baseline coverage levels allow algorithms to mathematically flag duplicated or deleted segments across the genome (Copy number variation related disease genes). If a region has twice the expected number of reads compared to a normalized baseline, it represents a genomic amplification.
Normal Genome: [--- Window 1 ---] [--- Window 2 ---]
Reads Mapped: |||||||||||||||||| |||||||||||||||||| -> Balanced Coverage
Tumor Genome: [--- Window 1 ---] [--- Window 2 (Duplicated) ---]
Reads Mapped: |||||||||||||||||| |||||||||||||||||||||||||||||||||| -> Spike in Depth
This computational approach allows researchers to extract structural insights directly from sequencing data. For example, Georgia Kontogianni et al. applied specialized copy number detection algorithms to exome sequencing data from melanoma patients to map structural alterations alongside point mutations (Dissecting the Mutational Landscape of Cutaneous Melanoma: An Omic Analysis Based on Patients from Greece). Similarly, Xin Gao et al. combined whole-genome sequencing with tools like GISTIC 2.0 to isolate driver amplifications and deletions in spinal schwannomas (Whole Genome Sequencing Identifies Key Genes in Spinal Schwannoma). On a broader scale, Waleed Iqbal et al. leveraged pan-cancer copy number analysis to demonstrate that specific genes, such as RRM2B, are systematically amplified across multiple distinct tumor types, altering cellular survival mechanisms (RRM2B Is Frequently Amplified Across Multiple Tumor Types: Implications for DNA Repair, Cellular Survival, and Cancer Therapy).
Inferring Genomic Alterations from Single-Cell RNA
The latest frontier in CNV analysis addresses a major limitation of bulk sequencing: tissue heterogeneity. Bulk sequencing blends tumor cells with normal stromal and immune cells, masking cell-specific variations. Single-cell RNA sequencing (scRNA-seq) captures the transcriptomes of individual cells, but it measures RNA, not DNA. How can we infer DNA copy number variations when we only have transcriptomic data?
At first glance, RNA expression seems too noisy because individual genes turn on and off dynamically based on immediate cellular needs. However, large-scale copy number alterations exert a persistent "dosage effect" on gene expression. If a massive segment of a chromosome is amplified, almost all genes within that region will show a slight, systematic increase in baseline transcription. If a segment is deleted, expression drops.
To exploit this, algorithms calculate a moving average of gene expression across adjacent chromosomal windows for each individual cell. By smoothing the expression data across hundreds of contiguous genes, the localized, dynamic transcriptional noise of individual genes cancels out. What remains is the structural baseline:
$$\text{Smoothed Expression} = \frac{1}{N}\sum_{i=1}^{N} \text{Normalized Expression of Gene}_i \text{ on Chromosome } X$$
If this moving average significantly deviates from known normal reference cells (such as infiltrating T-cells in the same sample), the algorithm flags that individual cell as possessing a CNV.
This technique has become a cornerstone for identifying malignant cells within complex microenvironments. For instance, Wen Wei et al. used single-cell CNV analysis to isolate malignant cell populations characterized by genomic instability in hepatocellular carcinoma, separating them from benign stromal cells (Exploring the level of metabolic reprogramming and the role of prognostic factor SF3A3 in hepatocellular carcinoma through integrated single-cell landscape analysis).
Similarly, Cuiying Chu et al. leveraged this approach to solve a long-standing debate in glioblastoma research by distinguishing tumor-derived pericytes from normal host pericytes (Immunoregulatory Roles of Tumor‐Originated Pericytes Identified by Single‐Cell Analysis in Glioblastoma). Because both cell types express identical surface markers, they look identical under standard single-cell clustering. However, by smoothing the transcriptomic data across chromosomal windows, the authors revealed that a specific subset of pericytes carried the exact large-scale genomic amplifications and deletions characteristic of the glioblastoma tumor cells, proving their malignant origin.
Key Foundational Work
Cytogenetic changes in radiation-induced tumors of the thyroid. (1999) — contributed 1 concept elements to this direction.
Analysis of Chromosomal Alterations in Non-Small Cell Lung Cancer by Multiplex-FISH, Comparative Genomic Hybridization, and Multicolor Bar Coding (2000) — contributed 1 concept elements to this direction.