TPSA and Permeability: Reading the 140 and 90 Thresholds
What TPSA (topological polar surface area) is, how Ertl computes it from 2D, and reading the 140 oral and 90 CNS permeability thresholds.
You add a polar group to fix solubility, the activity holds, and then the candidate stops crossing into the brain. The number that would have warned you sits in the property panel: TPSA. Topological polar surface area is the descriptor that tracks how much of a molecule’s surface is polar, and two thresholds on it — 140 Å2 for oral absorption and roughly 90 Å2 for blood–brain-barrier penetration — do a lot of predictive work for very little computational cost. This explainer covers what TPSA is, how it’s computed, how to read those two thresholds, and where the descriptor stops being reliable.
What TPSA is and how it’s computed
TPSA is the surface area of a molecule contributed by its polar atoms — principally oxygen and nitrogen, plus the hydrogens attached to them. It is reported in Ångström squared (Å2). Polar surface correlates with hydrogen-bonding capacity, and hydrogen bonds are what a molecule must shed to slip through a lipid membrane, so polar surface area is a direct proxy for the energetic cost of permeation.
The “topological” part is what makes it fast. Rather than building a 3D conformer and measuring its solvent-accessible polar surface, Ertl’s method sums precomputed contributions for each polar fragment based on the 2D structure alone.
Because it’s computed from connectivity, TPSA is deterministic and reproducible: the same structure gives the same value every time, which is why it’s practical for screening thousands of compounds. Ertl’s original work showed it matches the slower 3D polar-surface calculation to within experimental noise while running two to three orders of magnitude faster.
The two thresholds: 140 Å2 oral, ~90 Å2 CNS
TPSA earns its place in a property panel because of two well-established cutoffs.
- TPSA ≤ 140 Å2 — the Veber threshold for good oral absorption. Above ~140 Å2, passive permeation drops off and oral bioavailability tends to suffer.
- TPSA < ~90 Å2 — the rough ceiling for blood–brain-barrier penetration. CNS-active molecules generally need a smaller polar surface; many CNS drugs sit well under 90 Å2.
- Aspirin (C9H8O4): TPSA ≈ 63.6 Å2. Under both thresholds — consistent with its strong oral absorption.
- Caffeine (C8H10N4O2): TPSA ≈ 58.4 Å2. Under ~90 Å2 — consistent with the fact that caffeine readily crosses into the brain.
The practical move when a candidate creeps over 140 Å2 is to ask which polar group is contributing the most surface and whether it can be masked, replaced with a bioisostere, or removed without losing the binding interaction.
Relationship to HBD and HBA
TPSA and the hydrogen-bond counts measure overlapping but distinct things. HBD and HBA are counts — how many donors, how many acceptors. TPSA is an area — how much polar surface those groups expose. A compound can have a modest acceptor count but a large TPSA if its acceptors are surface-exposed, or many acceptors buried in a way that doesn’t inflate the surface as much.
This is why Veber added TPSA and rotatable bonds (≤ 10) as absorption predictors on top of the Lipinski donor and acceptor counts: the area captures something the counts miss. Read TPSA alongside HBD/HBA rather than instead of them. For how the hydrogen-bond counts feed the drug-likeness badge, see the Lipinski Rule of Five explained; for the lipophilicity half of the absorption story, see how to interpret a logP value.
Where TPSA stops being reliable
TPSA is a strong predictor inside its design envelope and quiet outside it. Know the limits before you trust the number.
- It only counts N, O, and their hydrogens by default. Ertl’s standard parameterization omits sulfur and phosphorus polar contributions. For sulfur- or phosphorus-rich molecules, a TPSA value from the default method understates the true polar surface — check whether your tool uses the extended parameter set.
- It models passive permeation only. Like the Lipinski rules, TPSA says nothing about actively transported compounds, which can be absorbed despite a high polar surface.
- The thresholds are soft. 140 Å2 and ~90 Å2 are statistical boundaries, not cliffs. A molecule at 92 Å2 isn’t disqualified from the CNS; it’s a slightly worse bet than one at 70 Å2, all else equal.
- It ignores conformation. Because TPSA is topological, it can’t see intramolecular hydrogen bonds that fold polar surface inward and let a high-TPSA molecule permeate better than its number suggests — a known limitation for macrocycles.
Reading TPSA from the property panel
When you draw a structure in ChemStitch, the property panel computes TPSA through RDKit (the Ertl topological method) and reports it in Å2 alongside MW, logP (Wildman–Crippen), HBD, HBA, and rotatable bonds. The value is marked Computed (deterministic, green) rather than AI Suggested, so you know it came from the fixed Ertl calculation and will reproduce exactly. To read it: compare against 140 Å2 if you care about oral absorption, against ~90 Å2 if you care about CNS penetration. If you need to confirm a molecular weight in the same drug-likeness pass, the molecular weight calculator computes it directly from a structure or formula.
TPSA is one of the highest-value-per-cost numbers on the panel: a single 2D-derived figure that flags both oral and CNS permeability risk in one read. Use it as the fast filter it was designed to be, and escalate the edge cases.
References: Veber et al., “Molecular properties that influence the oral bioavailability of drug candidates,” J. Med. Chem. 2002 (DOI: 10.1021/jm020017n); RDKit TPSA descriptor documentation (rdkit.org).