Functional Genomics

Functional genomics is the systematic, genome-scale determination of gene function through deliberate perturbation — deletion, knockdown, knockout, or overexpression — paired with high-throughput phenotypic readout. It stands in contrast to single-gene, hypothesis-driven genetics, where a researcher studies one gene's mutant phenotype in isolation over years. Functional genomics instead perturbs every gene, or a large curated subset, in parallel, and lets the phenotypic data itself point to which genes matter for a given process. Three conceptual pieces make this possible. First, a perturbation method that can be targeted to a specific gene and applied at genome scale: gene-deletion cassettes in yeast, short hairpin or short interfering RNAs (shRNA/siRNA) for RNAi, or single-guide RNAs (sgRNAs) directing Cas9 (or catalytically dead Cas9, dCas9) for CRISPR-based knockout, interference (CRISPRi), or activation (CRISPRa). Second, a phenotypic readout sensitive enough to detect the consequence of perturbing one gene among thousands — commonly a fitness/growth assay, a fluorescent reporter sorted by flow cytometry, or a transcriptomic profile via RNA-seq and Transcriptomics. Third, a genotype-phenotype linking system, because in a pooled experiment millions of cells are perturbed simultaneously and mixed together; each cell's specific perturbation must be traceable back through a DNA barcode or the guide RNA sequence itself, recovered by deep sequencing. The results feed into two complementary products: loss-of-function or gain-of-function phenotype catalogs (which genes are essential, which confer drug resistance, which control a signaling pathway) and genetic interaction maps (how pairs of gene perturbations combine — synergistically, antagonistically, or independently — to reveal pathway architecture). Together these outputs transform an annotated genome into a functional wiring diagram of the cell.

Step 1 — Choosing the perturbation

Gene deletion/knockout removes a gene's coding sequence entirely, usually by homologous recombination replacing the open reading frame with a selectable marker (as in the yeast deletion project) or by CRISPR-Cas9-induced double-strand breaks repaired by error-prone non-homologous end joining, producing frameshift indels that destroy the reading frame. This is a permanent, typically null perturbation.

RNAi exploits the endogenous RNA-induced silencing complex (RISC). A synthetic siRNA or a vector-expressed shRNA is processed by Dicer into ~21-nucleotide duplexes; the guide strand loads into Argonaute-RISC, which then binds complementary mRNA and either cleaves it (near-perfect complementarity) or represses translation and promotes decay (partial complementarity, as with endogenous microRNAs). Knockdown is partial and dose-dependent, not a clean null, and RNAi acts in the cytoplasm post-transcriptionally.

CRISPR-Cas9 perturbations depend on the fusion partner attached to a catalytically modified Cas9:

• Cas9 (nuclease-active) + sgRNA → double-strand break → indels → knockout.
• dCas9 (catalytically dead) fused to a transcriptional repressor domain (e.g., KRAB) + sgRNA targeting the promoter → CRISPRi, blocking transcription without editing DNA.
• dCas9 fused to an activator domain (e.g., VP64, or the SAM/dCas9-SunTag systems) + sgRNA targeting the promoter → CRISPRa, boosting transcription.

CRISPRi/a are reversible, titratable, and avoid the confound of permanent DNA damage — an advantage especially for essential genes, where a clean knockout kills the cell before its later function can be studied.

Step 2 — Library design and delivery

For genome-wide screens, a pooled library of tens of thousands of sgRNA- or shRNA-expressing constructs (typically 4-10 guides per gene, to average out guide-efficiency variability) is cloned into a lentiviral vector. Cells are infected at a low multiplicity of infection (MOI ≈ 0.3) so that, on average, each cell integrates only one construct — this is essential so that the phenotype observed in a cell can be attributed to a single, known perturbation. Library coverage is kept high (each guide represented in several hundred cells) to ensure that stochastic cell loss does not erase real signal.

Step 3 — Selection and phenotypic readout

Two dominant readout strategies:

Growth/dropout (fitness) screens. Cells are grown under a selective condition (a drug, a nutrient limitation, or simply extended passage). Genomic DNA is harvested at an early time point (T0) and a late time point (Tfinal). Because each cell's integrated construct carries a unique guide or barcode sequence, PCR-amplifying and deep-sequencing this region at both time points gives the relative abundance of every perturbation in the population before and after selection. Guides that drop out over time mark genes required for fitness or survival under that condition; guides that become enriched mark genes whose loss confers a growth or resistance advantage.

Sort-based (FACS) screens. Cells are engineered with a fluorescent reporter for the phenotype of interest (e.g., a pathway-activity reporter, or an antibody stain for a surface marker). The population is sorted into fluorescence bins, and the guide/barcode distribution in each bin is sequenced and compared to the unsorted population, identifying perturbations that shift the phenotype up or down.

Step 4 — Statistical hit calling

Because millions of reads map to tens of thousands of guides, robust statistics are essential. Standard pipelines (e.g., MAGeCK) rank guides by a log-fold-change in abundance between conditions, combine multiple guides per gene using a rank-aggregation or negative-binomial model, and apply false-discovery-rate correction across all genes tested simultaneously. Non-targeting control guides and genes with well-established phenotypes (e.g., known essential ribosomal genes) serve as positive and negative controls to calibrate sensitivity and specificity.

Step 5 — Interpreting genetic interactions

A single perturbation reveals what a gene does alone. Pairing perturbations — double knockouts, or combinatorial CRISPR libraries with two sgRNAs per cell — reveals epistasis: if the double mutant phenotype equals the sum/product expected from the two single mutants, the genes act independently; if the double phenotype is much worse than expected (synthetic lethality) or much better than expected (suppression), the genes lie in the same pathway or in redundant parallel pathways. This logic, systematized at genome scale, builds functional interaction networks analogous to (and cross-validated against) the yeast synthetic genetic array (SGA) maps.

The Yeast Systematic Deletion Project (Winzeler et al., 1999; consortium completed 2002). An international consortium constructed a deletion mutant for essentially every one of the ~6,000 open reading frames in Saccharomyces cerevisiae, replacing each gene with a KanMX resistance cassette flanked by two unique 20-bp "molecular barcode" sequences. Because every deletion strain carried its own barcode, thousands of strains could be pooled and grown together, and their relative abundance over time — measured first by microarray hybridization to a barcode chip, later by sequencing — gave each gene's fitness contribution in a single competitive-growth experiment. The result: approximately 1,105 of ~5,916 genes tested (about 19%) were essential for viability on rich medium, while the majority of genes were individually dispensable, revealing extensive genetic redundancy and buffering in the yeast genome. This was the proof of concept that a genome could be functionally profiled as a whole, not gene by gene, and it directly inspired the barcode/sequencing logic used in every subsequent pooled screen.

Genome-wide RNAi screens in Drosophila and human cells (early-to-mid 2000s). Following the discovery of RNAi mechanism by Fire and Mello (1998, Nobel Prize 2006), researchers built genome-scale shRNA and siRNA libraries and used them for loss-of-function screens in cultured cells — for example, screens for regulators of the innate immune NF-κB pathway and for genes required for viral infection. These screens proved RNAi's power for mammalian cells, where classical gene knockouts were slow and low-throughput, but also exposed a serious problem: off-target effects. Because siRNA silencing tolerates partial complementarity, a single shRNA could knock down dozens of unintended transcripts sharing a short "seed" sequence, generating false-positive hits. Cross-comparison of independent shRNAs targeting the same gene often showed poor phenotypic agreement — a rigor crisis that pushed the field toward requiring multiple independent reagents per gene and, eventually, toward CRISPR.

Genome-wide CRISPR knockout screens (Wang et al., Science 2014; Shalem et al., Science 2014). Both groups built lentiviral sgRNA libraries tiling essentially every annotated human gene (Wang et al.: ~7 guides/gene across ~18,000 genes; Shalem et al.: the GeCKO library, ~64,751 guides across ~18,080 genes), infected human cell lines (KBM7, a near-haploid leukemia line, and others), and performed dropout screens over weeks of culture. Guides targeting known essential genes (ribosomal proteins, spliceosome components, proteasome subunits) reproducibly and strongly dropped out, closely recapitulating the yeast essential-gene set and previously known essential gene lists — direct validation that CRISPR knockout screening recovers true biology with high sensitivity and far less off-target noise than RNAi, because Cas9-induced indels create a genuinely null allele at a single genomic locus rather than a partial, promiscuous knockdown. These papers established the modern standard workflow described above and triggered an explosion of CRISPR screens across cancer dependency mapping (the DepMap project), drug-resistance mechanisms, and immune-cell regulator discovery.

A cross-check worth noting: essential-gene fraction is remarkably consistent across methods and organisms when growth condition is matched — roughly 20% in yeast (rich medium), and comparable proportions (15-10%, depending on the cell line and stringency threshold) in human cancer cell lines from CRISPR dropout screens — providing independent, convergent evidence that gene essentiality is a robust, measurable biological property rather than a technology-specific artifact.

Example 1 — Calling a hit from a dropout screen. A CRISPR knockout screen targets a gene, TOR7 (hypothetical), with 5 independent sgRNAs. Cells are sampled at T0 (right after selection for infected cells) and after 21 days of proliferation (Tfinal), and guide abundance is measured by sequencing (reads per guide, normalized to total reads per million, RPM).

Four of five guides show strong, consistent depletion (log2FC roughly −3.2 to −4.0, i.e., an 8- to 16-fold drop in abundance); one guide (sg3) shows almost no effect. A robust rank-aggregation method (e.g., MAGeCK's RRA) would flag TOR7 as a significant dropout/essential-gene hit because the majority of independent guides agree, while treating sg3 as a low-efficiency guide (perhaps it fails to cut efficiently, or targets a non-functional exon). This illustrates why genome-wide screens use multiple guides per gene: a single-guide result is not trustworthy, but concordance across 4-5 independent reagents targeting the same gene is strong evidence of a true, on-target phenotype. If instead only 1 of 5 guides showed depletion, the gene would be discarded as a likely off-target artifact — the same logic, in modernized form, that RNAi screens learned the hard way.

Example 2 — Reading a genetic interaction (epistasis) result. In a combinatorial CRISPR screen, two genes are tested individually and in combination for their effect on cell growth, measured as relative growth rate compared to wild type (wild type = 1.0):

Under a multiplicative (independence) null model, the expected double-mutant fitness is 0.85 × 0.90 = 0.765. The observed value (0.05) is drastically lower than expected — this is a strong synthetic lethal interaction. Biologically, this pattern is the classic signature of two genes operating in parallel, redundant pathways that together perform an essential function: losing either pathway alone is tolerable because the other compensates, but losing both collapses the essential function entirely. This is precisely the logic exploited clinically in PARP inhibitor therapy for BRCA1/2-mutant cancers: BRCA-deficient cells rely on PARP-mediated single-strand break repair to survive (their homologous recombination pathway is already broken), so inhibiting PARP is synthetic lethal specifically in BRCA-mutant tumor cells while sparing normal BRCA-proficient cells — a genetic interaction discovered largely through this same synthetic-lethality screening logic, now foreshadowing the precision-targeting goals of Pharmacogenomics and Personalized Medicine.

:::warning[The single most dangerous confusion is treating a screen "hit" as a proven causal mechanism rather than a statistical association awaiting validation.] A gene scoring as a hit in a pooled screen has been shown only to correlate with the selected phenotype in that specific cell type and condition, filtered through the guide/barcode sequencing pipeline. Confirmation requires an orthogonal, arrayed (one-gene-at-a-time) validation experiment — an independent knockout or rescue construct tested alone, outside the pooled competitive context — before the hit is treated as biologically real. Pooled competitive growth can also introduce artifacts (e.g., a gene loss that changes a cell's growth rate for reasons unrelated to the selective pressure being studied), so screen hits are hypotheses generators, not final answers.

Functional genomics as a discipline does not trace to one individual the way Mendelian genetics traces to Mendel, but its technological backbone owes an enormous debt to Ronald W. Davis and the international Saccharomyces Genome Deletion Project consortium he helped organize and lead at Stanford in the late 1990s. Davis's laboratory pioneered the molecular-barcoding strategy — flanking each deletion cassette with unique 20-bp sequence tags — that made competitive pooled-fitness profiling possible for the first time at genome scale. This barcode logic, decades later, is structurally identical to the guide-RNA-as-barcode logic underlying every modern pooled CRISPR screen; Davis essentially invented the readout architecture that the entire field still uses, long before CRISPR existed. Equally central to the mechanistic side of functional genomics are Andrew Fire and Craig Mello, who discovered RNA interference in Caenorhabditis elegans in 1998 (shared Nobel Prize in Physiology or Medicine, 2006), demonstrating that double-stranded RNA — not single-stranded antisense or sense RNA alone — triggers potent, sequence-specific gene silencing. Their discovery supplied the molecular mechanism that RNAi screening technology exploited for a decade before CRISPR-based screening matured.

Cancer dependency mapping. The Broad Institute's DepMap project has run genome-wide CRISPR knockout screens across many hundreds of cancer cell lines, cataloging which genes each tumor type depends on for survival. This directly identifies candidate drug targets — a gene essential in a specific cancer subtype but dispensable in normal tissue is an ideal therapeutic target, extending the synthetic-lethality logic of PARP inhibitors to a systematic, genome-wide search.

Drug mechanism-of-action and resistance discovery. CRISPR knockout and CRISPRa screens performed under drug treatment identify genes whose loss confers resistance (revealing the drug's target or pathway) or hypersensitivity (revealing synthetic-lethal combination partners), accelerating both target validation and combination-therapy design in oncology and infectious disease.

Functional annotation of noncoding and regulatory elements. CRISPRi/a tiling screens across promoters, enhancers, and untranslated regions test the functional consequence of perturbing regulatory DNA rather than protein-coding sequence, complementing the genome-wide biochemical mapping performed by The ENCODE Project — ENCODE identifies candidate regulatory elements by biochemical marks (chromatin accessibility, histone modification, transcription factor binding), while functional genomics screens test whether perturbing those elements actually changes phenotype, providing causal validation that biochemical annotation alone cannot.

Agricultural and crop trait engineering. CRISPR screens in plant cell systems and model crops identify genes controlling yield, stress tolerance, and disease resistance, feeding directly into targeted genome-editing breeding programs rather than slow classical mutagenesis-and-selection breeding.

Immunotherapy target discovery. CRISPR screens in primary human T cells and tumor-immune co-culture systems have identified genes controlling T-cell exhaustion, persistence, and killing efficiency, informing the engineering of next-generation CAR-T cell therapies.

Pharmacogenomic variant interpretation. As whole-genome sequencing becomes routine in the clinic, functional genomic screens increasingly test the phenotypic consequence of specific patient-derived variants (via base editing or prime editing in pooled saturation-mutagenesis screens), directly supplying the causal evidence needed to move from "this variant is associated with drug response" to "this variant causes altered drug metabolism" — the mechanistic foundation that Pharmacogenomics and Personalized Medicine depends on.

Ask the tutor: How would you design a functional genomics screen to find genes that make a specific cancer cell line resistant to a chemotherapy drug?