Tools for Following Cells and Genes

A neuron carries no label saying when it was born, which cell it descends from, or which genes it needs. Every claim in this course rests on a method that supplies one of those labels. So before the biology, the toolkit, sorted by the question each tool can answer.

Four questions come up again and again. What is this cell? Where does it send its axon? When did it stop dividing? And what does it need a given gene for? A method that answers one of them answers the others badly or not at all, which is why most of the mistakes in this field are mismatches between a tool and a claim.

What is this cell, and where does it connect?

Three staining strategies, three uses. A Nissl stain binds the RNA-rich cytoplasm around a nucleus, so it shows every cell body in a section and nothing else: good for counting cells and finding the boundaries of layers. A Golgi stain fills a small, unpredictable fraction of cells completely, including their dendrites, so you see the whole shape of a few neurons against a clear background. Immunostaining uses an antibody against one chosen protein, so it reports a molecular identity. The cost is specificity in the other direction: an antibody raised against a mouse protein may not recognize the chick version, and a monoclonal antibody that sees one site can fail when that site is masked. The logic of antibodies as detection reagents is covered in detecting a signal.

Connections need a different trick, and the trick is the neuron’s own transport machinery. Inject cholera toxin B subunit into a region and neurons there take it up and can carry it toward their terminals, so it can report where those cells project; in practice it is also widely used as a retrograde tracer, so the direction has to be established for each experiment. Inject Fluorogold into a target region instead, and terminals take it up and carry it back to the cell bodies, which lights up every neuron that projects there: a retrograde tracer. Either way, what you learn is a population’s wiring, not one cell’s synapses.

When was this cell born?

“How to label new-born neurons?” The answer uses DNA replication as a clock. Tritiated thymidine, and later the thymidine analog BrdU, are incorporated into DNA only during S phase. Give a pulse, wait, and look for label.

The reading needs care. Each further division splits the label between two daughters, so a nucleus that is still heavily labeled after a long chase divided only a few more times. A cell’s “birthday” from such an experiment means its last S phase, inferred from heavy residual label, and weakly labeled nuclei are the descendants of cells that kept cycling. The timestamp is approximate, and it says nothing about what the cell became.

Giving a thymidine analog to people only to study the brain is not done. The classic exception, Eriksson and colleagues in 1998, used postmortem tissue from patients who had already been treated with BrdU as part of cancer care. For humans generally there is a retrospective method: atmospheric nuclear weapons testing raised atmospheric carbon-14 sharply before the 1963 test ban treaty and it has fallen since, so the carbon-14 content of a cell’s genomic DNA reflects the year its DNA was last made. Spalding and colleagues described this approach in 2005, matching the carbon-14 in genomic DNA to the atmospheric record. Here the method is the point; what it has found about turnover in particular human brain regions is contested and belongs to the lesson on adult neurogenesis.

Whose descendant is this cell?

A birth date is one cohort of cells. A lineage is one family. Three ways to follow one.

Grafting with a nuclear marker is the classic approach. Quail nuclei carry a distinctive condensed-chromatin pattern, so quail cells grafted into a chick embryo stay recognizable in sections after they migrate and differentiate. Le Douarin built an experimental program on that difference, following grafted populations into ganglia, gut and head mesenchyme.

Retroviral labeling works inside one animal. A retrovirus integrates its genome into the DNA of a dividing cell, so every descendant of that cell inherits the insert and the marker gene it carries. Keep the dose low enough that labeled clones do not overlap, and a cluster of marked cells is one progenitor’s family. That assumption is a sampling argument, not an observation, and it is where this method can fail.

The genetic version is the one you will use. A Cre-dependent reporter is a fluorescent protein gene with a transcriptional stop signal flanked by loxP sites in front of it. When Cre recombinase appears in a cell, it deletes the stop, and the reporter is read in that cell and in every one of its descendants from then on. The marking is permanent and heritable through division, which is exactly what lineage tracing requires. Reporters and actuators themselves are taught in hacking the cell; here the point is the recombination.

How to make a mouse

To ask what a gene does, change it. In the classical route you work in mouse embryonic stem cells, cells from an early embryo that can still contribute to any tissue. Supply DNA matching the target gene on both sides of your intended change, and in rare cells homologous recombination swaps your version into the gene’s own place. Capecchi’s laboratory, together with Smithies’s and Evans’s, established this. Because the events are rare, you select twice: a positive marker inside the construct keeps cells that took the DNA up, and a negative marker outside the matching arms, classically a thymidine kinase gene that makes ganciclovir lethal, kills cells that inserted the construct at a random site. Destroying the gene is a knockout; writing a new sequence in, such as a fluorescent reporter, is a knock-in. The selected cells go into an early embryo, the pup is a chimera, and if those cells made its gametes the allele is heritable. CRISPR has shortened this path: a guide RNA brings the Cas9 nuclease to a matching sequence in a fertilized egg, so many alleles can now be built without stem cells or chimeras.

A straight knockout has two limits. If the gene is needed early, the embryo dies before the question you care about. And a whole-body deletion cannot tell you which tissue mattered. Cre and lox solve both. Flank the important part of the gene with loxP sites, a 34-base-pair sequence; that allele is “floxed” and works normally. Then supply Cre from a promoter active in one cell type, and the gene is deleted there alone. For timing, fuse Cre to a modified estrogen receptor (CreER) so it stays out of the nucleus until the animal is given tamoxifen, or use a tetracycline-controlled system driven by doxycycline. “How do you get expression in 2 different neuron types?” Add a second recombinase with its own sites, Flp, and run two independent switches; requiring both in the same cell is how intersectional specificity is achieved.

Read the notation and you can read the cross. A superscript gives what is written into a locus and a slash separates the two alleles: p75NTR-/- is a homozygous knockout; DR6+/- is a heterozygote; p75NTR-/-;TNFR1-/- is a double knockout; AvilCre/+ is one copy of a Cre allele controlled by Advillin, a sensory-neuron gene; TrkAfl/- carries one floxed TrkA allele over a null. So AvilCre/+;TrkAfl/- deletes TrkA in sensory neurons and keeps only the floxed copy available to lose, while AvilCre/+;TrkAfl/+ is its control.

One trap. A reporter cross tells you where Cre acted. It does not tell you where the floxed gene was lost. The two lox-flanked targets recombine independently, drivers can be leaky or mosaic, and protein made before the cut persists.

Design problem

Gene X is needed for the embryo to survive, so a straight knockout dies before birth. You want to know what X does in sensory neurons after birth. Design the mouse: list the alleles, the drug, the reporter, and three controls.

Necessary, sufficient, reporter, actuator

Two pairs of words will carry you through most papers in this course.

The first pair is about inference. “Sufficient” means you added the thing and the outcome followed: overexpress protein A and structure B expands. “Necessary” means you removed it and the outcome failed: knock out A and B disappears. Each can hold without the other. Sufficiency without necessity usually means a redundant relative covers the loss; necessity without sufficiency means A is one requirement among several.

The second pair is about tools. A reporter reads out a state: a fluorescent protein marking where a cell is, a tagged protein showing where it sits inside a cell, a calcium sensor brightening when calcium rises. An actuator imposes a state: channelrhodopsin, a cation channel that blue light opens, to drive firing; halorhodopsin, which moves chloride under yellow light, to suppress it. Light-gated channels are introduced in ion channels as receptors. Chemical actuators work too, and the professor’s phrase for the idea is exact: “use trp channels to activate any neuron you want.” Put TRPV1, a capsaicin-gated channel, into chosen neurons of a mouse whose own TRPV1 is knocked out, and capsaicin activates those neurons alone. Reporters can only ever tell you that activity correlates with something. Only an actuator can ask whether the activity is enough.