16. 【Lab】 One receptor sees many · Prof. Kwon's TCR-HLA and CD4/CD8, First Steps

Start with the arithmetic that does not work.

The naive T cell pool holds on the order of 10⁷ to 10⁸ distinct TCRs. The number of possible pMHC combinations is far larger. Nine-residue fragments alone give 20⁹ ≈ 5×10¹¹, and that is before multiplying by the number of plates.

We are short by more than ten thousand fold

There is only one way out. One TCR has to recognise several fragments. This is cross-reactivity, and it is not a defect but a requirement.

Let us count how many, in our own table.

5.9%. It looks low, and we are about to see why the table is biased downward.

Python Runs in your browser

# How many fragments does the same CDR3beta bind
import thla

f = thla.facts()
c = f["cross"]
print(f"distinct CDR3beta sequences   {c['n_cdr3b']:>8,}")
print(f"binding two or more fragments {c['multi']:>8,}   ({100*c['multi']/c['n_cdr3b']:.1f}%)")
print(f"the widest one                {c['max']:>8,} fragments")
print()
for k in sorted(c["hist"], key=int):
    n = c["hist"][k]
    print(f"   {k:>2} fragments  {n:>7,}  " + "█" * max(1, int(50 * n / c['n_cdr3b'])))

The ones at the top are mostly short or common shapes — hands that grab anything.

Python Runs in your browser

# Count it directly — do not take facts on trust
import thla

g = thla.cross_reactive(min_epitope=2, gene="TRB")
print(f"CDR3beta binding two or more fragments : {len(g):,}")
print()
print("the widest binders")
for cdr3, n in g.head(6).items():
    print(f"   {cdr3:<20}{n:>4} fragments   length {len(cdr3)}")

Move the threshold. 'How many counts as cross-reactive' changes the number completely.

Loading…

Short CDR3betas rise to the top here too — a short loop blocks less of the groove.

Python Runs in your browser

# Does it still bind if you change the plate rather than the fragment
import thla

d = thla.vdj(gene="TRB")
g = d.groupby("cdr3")["mhc_a"].nunique()
multi = int((g >= 2).sum())
print(f"CDR3beta appearing on two or more HLA molecules : {multi:,}")
print(f"that is {100*multi/len(g):.1f}% of all of them")
print()
print("the ones spanning the most plates")
for cdr3, n in g.sort_values(ascending=False).head(5).items():
    print(f"   {cdr3:<18}{n:>3} plates   length {len(cdr3)}")

Many people see the same fragment the same way

There is a skew in the other direction too: fragments with tens of thousands of CDR3β sequences against them.

These are called public responses. Different people recognise the same fragment with similar TCRs — the V gene and the CDR3 shape repeat from person to person.

Which matters because a public response is something you can predict. Tests that read a TCR sequence and say "this person has met CMV" come from here.

Lesson 7's bias again. The well-known fragments get better known.

Python Runs in your browser

# The fragment-plate pairs with the most TCRs piled on them
import thla

f = thla.facts()
print(f"{'fragment':<14}{'plate':<16}{'CDR3beta':>10}")
print("-" * 42)
for p in f["public"][:10]:
    print(f"{p['epitope']:<14}{p['allele']:<16}{p['n_cdr3b']:>10,}")
print()
print("SLLMWITQV is the NY-ESO-1 tumour antigen, NLVPMVATV is CMV pp65,")
print("GILGFVFTL is influenza M1. Eight of these ten rows are HLA-A*02:01.")

When cross-reactivity causes harm

Molecular mimicry

A T cell that recognised a pathogen fragment also binds a self fragment that looks like it. Long suspected as a route from infection to autoimmunity.

Drug hypersensitivity

Abacavir lodges inside the groove of HLA-B*57:01 and changes its shape. The self peptides that normally sit there become unfamiliar. HLA typing is now done before prescribing.

Toxicity of engineered TCRs

An affinity-enhanced TCR aimed at MAGE-A3 also bound a fragment of cardiac titin, and patients died (2013). Raise the affinity and you raise the cross-reactivity with it.

Read that last box against lesson 11. This is where the story of "a TCR was built to bind weakly, and that was not an oversight" finishes.

🤔 Check yourself

Why is TCR cross-reactivity a 'requirement' rather than a 'defect'?

🤔 Check yourself

Why can we not report our 5.9% as the size of cross-reactivity?

🤔 Check yourself

What risk grows when the affinity of a therapeutic TCR is raised artificially?