“Matrix effect 20%.”
The same line can mean two different things. In one lab it means the analyte lost 20% of its signal in the sample. In another lab it means it kept 20% and lost 80%. Both labs did the same experiment and wrote it down correctly. They used different formulas.
That is a small example of a larger problem. Matrix effect is one of the first things to check when an LC-MS/MS result looks wrong, and it is measured and reported in ways that do not compare well between labs. I worked in bioanalytical LC-MS/MS at a bioequivalence lab and at a CRO, and the matrix effect experiment is the one I would most like analysts to understand, rather than just run.
What is matrix effect in LC-MS?
Matrix effect is a change in the analyte’s response caused by other compounds from the sample that co-elute with it. It happens mostly in the ion source. The co-eluting compounds compete for charge or change how the droplets evaporate, so the analyte gives less signal (suppression) or, less often, more signal (enhancement).
Electrospray is the most affected source. APCI is less affected because it has no liquid-phase suppression, but it is not immune (Cortese et al. 2020, pp. 12–14).
Two points make it hard. The same analyte responds differently in different matrices. And the same matrix affects different analytes differently. So a matrix effect number belongs to one analyte, in one matrix, with one sample preparation.
Where in the run does it happen? Post-column infusion
Before you measure how much, find out where.
Infuse a solution of the analyte through a tee after the column, so the MS sees a steady signal. Then inject an extracted blank sample. Wherever co-eluting matrix comes off the column, the steady baseline dips (suppression) or rises (enhancement). You get a map of the whole run.
The map usually shows two problem zones: the solvent front and the end of the gradient, where most of the interfering material elutes (Cortese, p. 14). If your analyte sits in either zone, moving it out is the cheapest fix you have. A divert valve that sends the front and the wash to waste also keeps the source cleaner (Cortese, p. 2).
Post-column infusion gives no number. It tells you where to look, and it is most useful early, when you are still choosing the sample preparation and the gradient.
How do you measure how much? The post-extraction spike
This is the experiment from Matuszewski and co-workers, and most guidelines are built on some version of it. You prepare three sets at the same concentration:
- A: analyte in clean solvent
- B: blank matrix, extracted, then spiked with analyte
- C: matrix spiked with analyte, then extracted
From these:
- Matrix effect, ME% = B/A × 100
- Recovery, RE% = C/B × 100
- Process efficiency, PE% = C/A × 100
In this form, 100% means no matrix effect, below 100% is suppression and above is enhancement (Cortese, p. 6). Divide the analyte’s value by the internal standard’s value and you get the IS-normalised matrix factor, which is the number that tells you whether your internal standard is doing its job.
Three formulas for the same experiment
This is the “20%” problem from the top of the page. The same B and A are written in at least three ways:
| Convention | Formula | No effect | 20% suppression reads as |
|---|---|---|---|
| Matuszewski | B/A × 100 | 100% | 80% |
| Buhrman | 100 − (B/A × 100) | 0% | 20% |
| EU food and PFAS guidance | (B/A − 1) × 100 | 0% | −20% |
The first two are from Cortese (p. 6). The third is the form in the EU reference laboratory’s PFAS guidance (EURL POPs, 2024, p. 12) and the EU pesticide guidance (SANTE/11312/2021, glossary), so for food labs in Europe it is the normal one.
None of them is wrong. The problem is a report that gives the number without the formula. Write the formula next to the number, every time.
How many lots, and what is the limit?
It depends on which document your lab works to, and they do not ask for the same thing.
- ICH M10 (bioanalysis, 2022) tests matrix effect as accuracy and precision: low and high QCs in at least 6 lots of matrix, accuracy within ±15% and precision 15% or less (pp. 11–12). It asks for no matrix factor, and recovery is a separate experiment.
- EMA 2011 (now replaced by M10) and CLSI C62 asked for an IS-normalised matrix factor, with a CV under 15% across lots, as summarised by Castillo-Ribelles et al. 2025, Table 1.
- EU pesticide guidance (SANTE/11312/2021) says a matrix effect above 20% has to be addressed, usually with matrix-matched calibration.
One consequence: under M10, a method can pass with matrix effect and recovery folded together, and you cannot tell from the result which of them is moving (Castillo-Ribelles, p. 5). Two labs can both be compliant and report numbers that cannot be compared.
Is recovery the bigger problem?
Sometimes, yes. Castillo-Ribelles and co-workers ran the full three-set experiment on a lipid assay in cerebrospinal fluid, in 3 lots. The peak-area CVs were 5.2–13.7% in clean solvent and 6.4–10.4% in post-extraction spiked matrix. When extraction was added, they jumped to 14.8–34.1% (p. 6).
So in that method, adding the matrix to the ion source changed little. The extraction step was where most of the variation came from.
That is one method and three lots, so it is not a general rule. But it is a good reason to run all three sets instead of only the post-extraction spike. If you measure only ionisation, you can spend a week on the source while the variation is in the extraction.
The same paper found a second trade-off: the lots with higher recovery also had more suppression, because an extraction that pulls out more analyte also pulls out more interferences (p. 8). Cleaner and more complete are not the same direction.
Does a stable-isotope internal standard fix it?
It compensates for the effect, and the effect is still there.
The reason a stable-isotope-labelled internal standard (SIL-IS) works is that it co-elutes with the analyte and sees the same matrix at the same moment in the source. ¹³C and ¹⁵N labels co-elute better than deuterium labels, because each deuterium can shift retention slightly. Guidelines and reviews prefer them for that reason (Cortese, p. 8; ICH M10, p. 10).
The interesting result from Castillo-Ribelles is what “works” means. Their internal standard was deuterated and it did not behave like the analyte. It was suppressed more, and it recovered about half as well (p. 7). By the usual test it was a poor match.
But it varied in the same pattern as the analyte from lot to lot. After normalisation, the lot-to-lot CV of recovery fell from 20.8–42.6% to 1.5–8.6% (p. 8).
So the question to ask of an internal standard is whether it moves with the analyte across lots. It does not have to match it. This is also why ICH M10 asks you to monitor internal standard response during study runs: an internal standard that stops moving with the analyte is a failure the calibration curve will not show you.
Food labs see the same thing. Banno et al. 2024 found that matrix effect on 25 pesticides varied between samples of the same vegetable, and recommended blank matrix of the same cultivar from the same field for matrix-matched calibration (p. 11). The blank you calibrate in is itself variable.
Minimize it or compensate for it?
Cortese and co-workers reduce the decision to two questions (Figure 1, p. 3).
Is sensitivity critical? If yes, minimize the effect:
- Dilute, if your LOQ allows it.
- Move the peak away from the solvent front and the end of the gradient.
- Clean up more: from protein precipitation to supported liquid extraction or SPE, or to phospholipid-removal plates. Protein precipitation is fast and leaves the phospholipids behind.
If sensitivity is not critical, compensate. Then the second question is whether you have blank matrix.
- With blank matrix: matrix-matched calibration, a SIL-IS, or both.
- Without it: a SIL-IS, a surrogate matrix, a surrogate analyte, background subtraction, or standard addition.
One limit on standard addition, from the Eurachem guide: it corrects interference that changes the slope, and does nothing for a fixed background signal that changes the intercept (Eurachem, The Fitness for Purpose of Analytical Methods, 2025, p. 23). If something co-elutes and adds its own signal, standard addition will not rescue the method.
What I would write in the method report
If a reviewer, an auditor or another lab will read your matrix effect result, give them enough to understand it:
- the formula you used
- how many lots, and what they were (plasma, haemolysed, lipaemic; or which crop)
- the concentrations tested
- absolute matrix factor and IS-normalised matrix factor, with the CV across lots
- recovery as a separate number, if you ran set C
- the injection order
The last one is easy to forget. Castillo-Ribelles saw a 15–30% drop in signal when the same batch was injected a second time (p. 8). The experiment that measures matrix effect is itself running on an instrument that drifts.
Sources
Every claim above comes from these documents. Page numbers are given in the text.
- Cortese M et al. Compensate for or minimize matrix effects? Strategies for overcoming matrix effects in LC-MS: a tutorial review. Molecules 2020; 25:3047. doi:10.3390/molecules25133047. A review, so several of its numbers are quoted from other papers; I have used it for definitions, the formulas and the decision frame.
- Castillo-Ribelles L et al. Systematic assessment of matrix effect, recovery, and process efficiency using three complementary approaches. ACS Omega 2025; 10:52449. doi:10.1021/acsomega.5c05399. One method, 3 lots of cerebrospinal fluid.
- Banno A et al. Variability in the matrix effect on stable isotope-labeled internal standards in LC-MS/MS analysis of 25 pesticides in vegetables. J Pestic Sci 2024; 49:65. doi:10.1584/jpestics.D23-060.
- ICH M10, Bioanalytical method validation and study sample analysis, Step 4, 2022.
- Eurachem, The Fitness for Purpose of Analytical Methods, 3rd edition, 2025.
- EURL POPs, Guidance document on the analysis of PFAS in food and feed, 2024.
- European Commission, SANTE/11312/2021, Analytical quality control and method validation procedures for pesticide residues analysis in food and feed.
The LC-MS troubleshooting cheatsheet has the short version of where to look when a signal drops. And if matrix effect is where your method keeps failing, tell me the analyte and the matrix in one line on LinkedIn. I will tell you where I would look first.
Common questions
- What is matrix effect in LC-MS?
- A change in the analyte's response caused by co-eluting compounds from the sample, usually in the ion source. It can be suppression or enhancement. Electrospray is the most affected source; APCI is less affected but not immune.
- How do you calculate matrix effect?
- Compare the analyte spiked into blank matrix after extraction (B) with the same amount in clean solvent (A). In the Matuszewski form, ME% = B/A × 100, so 100% means no effect and 80% means 20% suppression. Other conventions report the same experiment as 100 − B/A × 100, or as (B/A − 1) × 100, so always write the formula next to the number.
- What is the difference between matrix effect, recovery and process efficiency?
- Matrix effect compares post-extraction spiked matrix with clean solvent. Recovery compares matrix spiked before extraction with matrix spiked after. Process efficiency compares the pre-extraction spike with clean solvent, so it is the product of the other two.
- How many lots of matrix do I need to test?
- ICH M10, the bioanalytical guideline, asks for at least 6 lots, tested with low and high QCs, with accuracy within ±15% and precision of 15% or less. Food and environmental guidance sets its own rules; the EU pesticide guidance asks for matrix effects above 20% to be addressed.
- Does a stable-isotope-labelled internal standard remove matrix effect?
- It compensates for it, and the effect is still there. It works if the internal standard's response varies from lot to lot in the same way as the analyte's. ¹³C or ¹⁵N labels co-elute with the analyte better than deuterium labels, which is why guidelines and reviews prefer them.
- Should I minimize matrix effect or compensate for it?
- If sensitivity is critical, minimize it: dilute if you can, move the peak away from the solvent front and the end of the gradient, and clean up the sample. If it is not critical, compensate with matrix-matched calibration or a stable-isotope internal standard, or with standard addition or a surrogate matrix when there is no blank matrix.