What a serum SERS spectrum actually measures

The paper is "Adsorption of serum components on Ag colloids: on the biochemical interpretation of surface-enhanced Raman spectra of human serum" by Roberto Gobbato, Stefano Fornasaro, Valter Sergo and Alois Bonifacio of the University of Trieste, published open access in Analytical and Bioanalytical Chemistry (online 17 November 2025). Its conclusion is blunt: under the most common setup (silver substrate, 785 nm excitation, serum at its natural pH of 7.4), most of the bands in a serum SERS spectrum come from uric acid, with hypoxanthine behind a few more.

The problem: same peaks, many stories

SERS (surface-enhanced Raman scattering) is attractive for diagnostics because it is fast, label-free and sensitive. The authors note that serum spectra from different groups share the same pattern, so the measurement itself looks repeatable. The trouble is interpretation. Surveying papers from 2015 to 2023, they found the most intense band, near 638 cm⁻¹, attributed mostly to tyrosine, but also to lactose, uric acid and, in a few papers, methionine.

Their sharpest example: two groups studying liver cancer (hepatocellular carcinoma) independently saw the same thing, a higher ratio between the bands at 638 and 724 cm⁻¹ in patients. One read it as more tyrosine and lactose and less coenzyme A. The other read it as a higher uric acid to hypoxanthine ratio. Same data, two different biochemistries. If you are hunting biomarkers, only one of those can lead anywhere.

How the wrong assignments spread

The authors traced where the assignments came from. Most papers didn't show a spectrum of the molecule they named. They cited another paper, which cited another. Following these citation chains, the team often found they ended at a normal Raman spectrum, not a SERS one, or at a paper with no supporting evidence at all.

That matters because a SERS spectrum is not just a louder Raman spectrum. The signal depends on how a molecule binds to the metal surface, so band positions and relative intensities can shift a lot. And the matching was usually done on a single peak. The tyrosine case shows why that fails: in the widely cited normal Raman spectrum, tyrosine does have a band at 641 cm⁻¹, close to serum's 638. But it also has a band at 828 cm⁻¹ five times more intense, plus one at 1177 cm⁻¹. Neither shows up in serum SERS spectra. The authors call single-band matching a form of "cherry picking."

How they tested it

Instead of arguing from citations, they ran experiments, using the protocol most common in the literature they surveyed:

Measure the candidates directly. They ran 53 biomolecules under the same SERS conditions at physiological concentrations. Tyrosine and lactose gave no SERS signal at all. Uric acid gave an intense spectrum matching most serum bands; hypoxanthine gave a weak band at 725 cm⁻¹.

Spike the serum. Adding a candidate to real serum and watching for growth is the more robust test, because it includes matrix effects. Only uric acid and hypoxanthine made serum bands grow, by roughly two-fold, which fits the fact that the spikes doubled their concentrations. (It is the same logic as a spike-recovery check; Report 204 covers what spiking does and doesn't prove.) Human serum albumin, the most abundant serum protein at around 600 μM, added nothing to the pattern and made spectra weaker.

Remove the suspect. Uricase, an enzyme that breaks down uric acid, made the serum bands disappear. In serum spiked with hypoxanthine, the hypoxanthine signal survived the enzyme while the other bands went, which rules out the enzyme simply smothering everything.

Check real people. Across serum from 81 blood donors, the bands that rose and fell together with the 638 band were uric acid bands; the 725 band tracked a separate set matching hypoxanthine. Fitting each donor's spectrum with just uric acid (free and albumin-bound) and hypoxanthine gave a median R² of 0.949. In a principal component analysis, the first component explained about 70% of the variation and the first three about 90%, with loadings matching those two metabolites.

Label the atoms. One serum band near 1659 cm⁻¹ matched none of the pure molecules. Using uric acid with heavier carbon and nitrogen isotopes, the band shifted to 1653 cm⁻¹, proving it belongs to uric acid bound to albumin, not to the protein itself.

Why one molecule can dominate thousands

Serum contains more than 4,000 metabolites, so "it's mostly uric acid" sounds wrong. The authors' explanation is the core of SERS: signal comes from what binds to the metal, not from what is most abundant. In their words, the affinity of an analyte for the metal "is far more important than its concentration." Their example is glucose, present at millimolar levels yet very hard to see with direct SERS, mainly because it binds poorly to silver and gold.

As a spectroscopist, this is the lesson I'd underline. Every technique has a gate that decides what reaches the detector. In SERS that gate is surface chemistry. Read the spectrum as "what stuck to the silver," not "what is in the blood." (Report 013 covers the broader ways a Raman spectrum gets over-read.)

What this does and doesn't say

It is condition-specific. The authors limit their conclusions to silver, 785 nm excitation and pH 7.4. Other wavelengths, metals or pH could bring other molecules into view.

It doesn't rule out disease-specific molecules. They say contributions from other biomolecules can't be excluded in specific pathologies, though the patient spectra in the literature share the same features.

It doesn't say serum SERS classifiers fail. The paper is about biochemical interpretation, not diagnostic accuracy. My reading, and it is mine, not the paper's: if a classifier separates patients from controls, the first question should be whether it is mostly tracking uric acid, hypoxanthine and their ratio, rather than the exotic biomarkers the discussion section names.

One human detail. The group first proposed the uric acid assignment in 2014. In an interview with The Analytical Scientist, the authors said they contacted authors and submitted formal comments with little effect: "This left us increasingly concerned – and somewhat powerless – as the misinterpretation persisted." So they wrote this paper instead, with the data and R code posted publicly. That is how a field corrects itself when comments don't work.