Week 3 — Sep 28: NMR Spectroscopy

(Lecture 3.) Risk management gave us the question — how much evidence is enough — and the machinery for deciding. This week begins the middle movement of the course: the techniques, taken as ways of producing evidence and defending it. NMR comes first because it is the technique that leans least on a reference standard: it answers what is this and how much is here from first principles.

A note on the schedule. This block was two weeks of NMR with a specialist instructor, who has retired. It is now one week — identity, quantitation, structure, and interpretation — with the freed time redistributed to the data-to-decision movement and the modalities block later in the term. NMR interpretation remains the single best training ground for the judgment the rest of the course depends on, so it keeps a full week.

The one idea

NMR does not need to be told what it is looking at. Every other technique in this course compares an unknown to a standard; NMR reads the structure directly, and the signal is inherently proportional to the number of nuclei.

That property — response proportional to molar amount, with no compound-specific calibration — is why NMR is a primary ratio method: it can put a purity value on a reference standard that every other method then depends on.

The principle, for interpretation

We are not deriving the quantum mechanics. Four observables carry almost all the information a pharmaceutical analyst uses:

2D experiments (COSY, HSQC, HMBC, NOESY) resolve overlap and establish connectivity through bonds and space. For a pharmaceutical analyst the common uses are: confirming a synthetic intermediate, assigning an unknown degradant, and distinguishing regio- and stereoisomers that mass spectrometry cannot tell apart.

What NMR is for in a regulated setting

Making a qNMR result trustworthy

qNMR is only “reference-standard-free” if the experiment is set up so that every nucleus is counted equally:

• Relaxation delay long enough for full recovery — a recycle delay of 5–7 × the longest T1 of the peaks being integrated. Too short, and slowly-relaxing protons are under-counted.
• 90° pulse calibrated on the actual sample; acquisition time long enough to capture the full FID.
• Internal standard — a certified, high-purity compound with well-separated singlets (benzoic acid, maleic acid, dimethyl sulfone, 1,4-BTMSB in organics) and traceable purity.
• Signal-to-noise typically ≥ 150:1 on the analyte signal; adequate digital resolution; consistent, documented integration limits (including how far into the peak wings, and baseline/phase correction).
• Validated as any assay is (Q2): specificity, linearity, precision, accuracy — but note the accuracy claim rests on physics plus the internal standard, not on a matrix-matched calibration curve.

Compendial basis: USP General Chapter ⟨761⟩ (NMR spectroscopy) and Ph. Eur. 2.2.33, plus the qNMR-specific chapters and the metrology literature on qNMR as a primary method.

Worked case — the 2008 heparin contamination

Heparin is a heterogeneous polysaccharide extracted from pig intestine; its potency is a biological assay and its identity was, at the time, a set of general tests. In 2007–2008, batches sourced through the Chinese supply chain were adulterated with oversulfated chondroitin sulfate (OSCS) — a cheap semi-synthetic mimic that passed the existing pharmacopeial identity and potency tests. It was associated with hundreds of severe anaphylactoid reactions and a number of deaths.

The method that resolved it was 1H NMR: OSCS produces a distinct signal (a well-resolved methyl resonance near 2.15 ppm, offset from heparin’s ~2.04 ppm N-acetyl signal) that is unambiguous and quantitative. NMR (with capillary electrophoresis as a second method) was written into the USP and Ph. Eur. heparin monographs within months.

The lesson for the course: the contaminant was invisible because no method in the specification was designed to see it — a detection failure, exactly the kind risk management is meant to surface. And the fix was a technique that reads structure directly, so a novel adulterant still shows up as a peak that doesn’t belong.

Interpretation as craft

This is why NMR keeps a full week. A chromatogram can be integrated by software; an NMR spectrum has to be assigned, and assignment is a disciplined argument:

1. Predict — from the proposed structure, what should the spectrum look like? How many signals, at what shifts, with what multiplicity and integration?
2. Compare — where does the real spectrum match, and where doesn’t it?
3. Resolve the discrepancies — an extra signal, a missing coupling, an integration that’s off. Each one is either a flaw in the structure or a feature of the sample (impurity, solvent, exchange, rotamer).
4. Close the argument — every signal accounted for, or the structure is wrong.

That loop — predict, compare, revise — is Week 1’s definition of science at the bench, and it is the same loop the machine-learning week will try to partially automate.

Where the analyst sits

Automated structure-verification software will tell you a spectrum is “consistent” with a structure. It will not tell you that the one unassigned multiplet at 5.3 ppm is a rearrangement product that co-formulates as a covalent protein adduct. Reading a spectrum as a story about a sample — knowing which discrepancies are noise and which are the whole point — is the STEAM “A”, and NMR is where it is trained most directly. The refrain holds: science → evidence → reduced uncertainty → control → regulatory confidence → patient trust, and here the evidence is a structure you can defend line by line.

For discussion

• qNMR is called “reference-standard-free,” yet it needs a certified internal standard. In what sense is it standard-free, and why does that matter for valuing a primary reference standard?
• You run qNMR with a recycle delay of 1 s because instrument time is tight. The longest T1 in your analyte is 4 s. Which way is your assay biased, and by roughly how much?
• OSCS passed every test in the heparin monograph. Design the risk assessment that should have caught the gap before patients did — what would have flagged “we cannot see a novel adulterant”?
• A degradant shows the same accurate mass as the parent on LC–HRMS. What NMR experiments do you run, and in what order, to decide whether it is a positional isomer, a stereoisomer, or a rotamer?
• Solid-state NMR can quantify 2% of a second polymorph in a tablet. When is that worth the instrument time versus XRPD or Raman?
• If structure-verification software agrees with your proposed structure, what would still make you not sign off?

Source note. Anchored in USP ⟨761⟩ and Ph. Eur. 2.2.33 for NMR, the qNMR metrology literature (e.g. Malz & Jancke; the work of NMIJ, BAM, and LGC on qNMR as a primary ratio method), and the heparin/OSCS papers in Nature Biotechnology (2008) and the subsequent USP/Ph. Eur. monograph revisions. Structure-elucidation workflow follows the impurity-identification practice in Q3 and the Alsante degradant-profiling workflow in the Q1 page. (Instructor: confirm current ⟨761⟩ status and whether the department still has working high-field time for the interpretation exercise; decide how much 2D to cover given it is now one week, not two.)

• ←2 · Sep 21
• 4 · Oct 5→