Multiple Reaction Monitoring: Building and Optimizing MRM Transitions

Multiple reaction monitoring (MRM, also called selected reaction monitoring or SRM) is the acquisition mode that makes a triple quadrupole a quantitative instrument rather than a scanning one. Q1 is parked on a single precursor m/z, the collision cell fragments it, and Q3 is parked on a single product m/z. That precursor/product pair is a transition, and the instrument spends a defined slice of time — the dwell time — sitting on it.

Building an MRM method is therefore three decisions made in order: which precursor, which product ions, and how much collision energy to get from one to the other. What breaks real methods is a fourth thing that nobody decides explicitly — the time budget. Every transition you add takes time away from every other transition, and the arithmetic is unforgiving once a panel grows past a few dozen compounds. This guide walks the build end to end and then does that arithmetic properly, because it is the part that turns a method that validated on ten analytes into one that fails on eighty.

If you need the upstream context first — how the liquid chromatograph and the mass spectrometer are coupled at all, and what ionisation does before any of this — start with how LC and MS are coupled. If you are still deciding whether a triple quadrupole is the right instrument, the QTOF versus triple quadrupole comparison covers that trade-off, and the resolving power and mass accuracy each HRMS job actually requires works through the arithmetic on the high-resolution side of it. This page assumes the instrument is already in front of you.

Step 1: choosing the precursor ion

Infuse the neat standard, scan Q1, and look at what the source actually produces. The instinct is to take the most abundant peak. That instinct is wrong often enough to be worth naming the exceptions.

• Prefer the protonated or deprotonated molecule — [M+H]⁺ in positive mode, [M−H]⁻ in negative. These fragment predictably and reproducibly under collision-induced dissociation.
• Be wary of sodium and potassium adducts. [M+Na]⁺ is frequently the tallest peak in the Q1 scan, and it is frequently the worst choice: alkali-metal adducts are held together tightly and fragment poorly, so you burn collision energy and get a weak, unstable product ion. Worse, adduct abundance tracks the sodium content of your mobile phase, glassware and sample — a parameter you are not controlling. A transition whose precursor intensity depends on trace sodium will drift between batches.
• Watch for in-source fragmentation. Labile compounds (glucuronides, N-oxides, some esters) can lose their conjugate in the source before Q1 ever sees them. If the intact precursor is weak and a fragment mass is strong, you may be looking at an in-source product, not the molecule. Selecting it as a precursor is legitimate, but only if you understand that you have given up the specificity of the first mass filter — anything else in the matrix that produces that same in-source fragment will co-elute straight into your quantifier.
• Consider multiply charged species for peptides and large molecules. For tryptic peptides the doubly charged [M+2H]²⁺ is usually the right precursor: it sits in a cleaner region of the m/z range and gives good b/y-ion series on fragmentation.

Record the source conditions you used during infusion, because precursor formation is source-dependent — optimising those conditions is a separate ESI source-tuning procedure with its own order of operations. A precursor selected at 500 °C with one nebuliser setting may not be the dominant species under your final LC flow rate.

Step 2: selecting product ions — quantifier and qualifier

Ramp the collision energy on the chosen precursor and record the product-ion spectrum at several energies. You are selecting at least two transitions per analyte: a quantifier (the one you integrate) and one or more qualifiers (used to confirm identity via the ion ratio).

Selection rules that matter more than raw intensity:

• Avoid product ions close in mass to the precursor. A transition where Q3 sits only a few daltons below Q1 — a water loss at [M+H−18], an ammonia loss at [M+H−17] — is chemically generic. Water loss is the single most common neutral loss in the matrix as well as in your analyte, so this transition carries almost no specificity, which is the entire point of tandem MS. It is often the most intense product ion in the spectrum, and it is often the worst quantifier.
• Avoid the low-m/z region. Product ions below roughly m/z 100 sit in a crowded, chemically nonspecific part of the spectrum where matrix background is highest.
• Prefer a structurally diagnostic fragment — a cleavage that reflects the specific scaffold of your analyte rather than a functional group shared by half the compound class. In a multi-residue panel this is what stops two structural isomers from cross-talking.
• Check for cross-talk within your own panel. If analyte A and analyte B share a precursor mass, or if A’s product mass equals B’s, verify by injecting each standard alone and confirming no signal appears in the other’s channel. This is a real failure mode in large pesticide and drug-of-abuse panels and it is invisible in a mixed standard.

The quantifier should be the most intense transition that survives all of the above — not simply the most intense transition. Sensitivity you cannot trust is not sensitivity.

Step 3: optimising collision energy (and why there is no universal number)

Collision energy is optimised per transition by acquiring a breakdown curve: hold Q1 and Q3 fixed, step the collision energy across a range, and plot product-ion intensity against energy. The optimum is the apex. Do this for the quantifier and each qualifier independently — they frequently peak at different energies, and a qualifier acquired at the quantifier’s optimum will give you an ion ratio that is stable but unnecessarily noisy.

A collision energy value is not portable between instruments, and you should not treat a published one as a starting point without checking why. The number depends on:

• Collision cell design. A short RF-only hexapole, a long linear accelerating cell, and a curved collision cell impose different residence times and different numbers of collisions at the same nominal energy.
• Collision gas identity and pressure. Nitrogen and argon differ in mass and therefore in energy transfer per collision; cell pressure sets how many collisions occur.
• How the vendor defines the number. Some instruments express collision energy as a laboratory-frame potential difference in volts; others apply a charge-state or mass-dependent normalisation. The same physical fragmentation can be reported as two quite different numbers.

The practical consequence: a CE from a vendor application note or a published method is a region to search, not a value to copy. Re-run the breakdown curve on your own instrument. For a large panel, most vendor software will do this automatically via a compound-optimisation routine that infuses or flow-injects each standard — use it, but spot-check a handful of the resulting curves by eye, because automated routines will happily report an apex on a curve that is monotonic within the range tested (meaning the true optimum lies outside your range).

The same instrument-dependence applies to declustering / cone / fragmentor voltage, which is optimised on the precursor before you touch collision energy.

Step 4: the dwell time and cycle time budget — where methods actually break

This is the arithmetic that decides whether a panel works, and it is the step most often skipped because the instrument software will accept a method that cannot produce integrable peaks.

Two equations govern it. From SCIEX’s own guidance on choosing dwell and cycle time, the first is the definition of a cycle:

Cycle time = number of transitions × (dwell time + pause time between two MRM transitions)

The pause time (also called interscan or interchannel delay, typically around 5 ms by default) is the settling period while the quadrupole rods change voltage between transitions. It is dead time: no ions are counted. Critically, it is charged per transition and it does not shrink. That single property is what makes large panels fail.

The second equation comes from the chromatography, not the mass spectrometer:

Cycle time = peak width / number of points per peak

Combining them gives the number you actually need:

Dwell time = (peak width / points per peak) / number of concurrent transitions − pause time

How many points across a peak — and the convention trap

Two different conventions are in circulation and they do not give the same answer. SCIEX’s dwell/cycle-time guidance works from the peak width at the base, illustrating a target of 10 points per peak (a 15-second base-width peak therefore needs a 1.5-second cycle time). Other vendor guidance quotes a minimum of roughly 6–8 points across the peak at half height as adequate for integration. Because a peak is substantially narrower at half height than at the base, these are not interchangeable numbers — specifying “8 points” against a base width when the guidance meant half-height width will under-sample the peak by roughly a factor of two.

State which width you are using in your method documentation. The practical position: 10–12 points across the base width is a defensible target for a quantitative method, more if you are integrating asymmetric or partially resolved peaks, and fewer than about 8 across the base makes peak-area precision depend on where the sampling grid happens to fall relative to the apex — which is a source of run-to-run imprecision that looks like chromatography or injection error and is neither.

The budget, worked

Take a conventional LC peak of 15 s base width, a 10-point target (so a 1.5 s cycle time), and a 5 ms pause time. The available dwell time per transition as the panel grows:

Three things fall out of that table that are not obvious from the method editor:

1. There is a hard ceiling on concurrent transitions that has nothing to do with sensitivity. It is cycle time / pause time — here, 300 — and at that ceiling your dwell time is zero. The instrument is spending 100% of the run settling and 0% counting ions. You will hit this wall abruptly, and the software will not necessarily warn you.
2. Pause time becomes the dominant term long before the ceiling. At 10 concurrent transitions the pause costs you 3% of the cycle. At 100 it costs 33%. At 200 it costs 67%. Reducing pause time (where the instrument allows it) buys more at high multiplexing than any other single change — but reduce it too far and Q1/Q3 have not settled, which shows up as intensity loss and cross-talk rather than as an error.
3. Narrow peaks make everything worse, quadratically in effect. Move from conventional LC to UHPLC and a 15 s peak may become 6 s. The required cycle time drops to 600 ms, and at 50 concurrent transitions your dwell falls from 25 ms to 7 ms. Sharper peaks give better chromatographic resolution and worse mass-spectrometric statistics at the same time. This is the single most common reason a method that worked on an older LC degrades after a column or system upgrade.

Why short dwell times cost you sensitivity

Dwell time is an ion-counting interval. Ion counting is a Poisson process, so the relative standard deviation of the counts scales as 1/√N, where N is the number of ions counted — which is proportional to dwell time. Halving the dwell time does not halve the signal-to-noise ratio; it degrades it by roughly a factor of √2. That is a gentler penalty than most people assume, which is why heavy multiplexing is viable at all — but it compounds, and it hits your lowest-concentration analytes first. If your limit of quantitation was established at 100 ms dwell and the production method runs at 10 ms, that LOQ is no longer the LOQ.

Note also that some instruments have a genuine minimum usable dwell time and others are effectively limited only by the pause time; SCIEX documents that its linear-accelerator collision cell design permits dwell times of 5 ms or less, while cautioning that longer dwells remain preferable for low-abundance analytes. Check your own instrument’s specified floor rather than assuming a number — this is one of the parameters that varies most between platforms.

Step 5: scheduled MRM — buying the budget back

The table above says “concurrent transitions,” not “total transitions,” and that distinction is the entire escape route. In classical MRM every transition in the method is acquired throughout the whole run, so concurrent equals total. In scheduled MRM (vendor names vary: Scheduled MRM, dynamic MRM, timed SRM) each transition is acquired only inside a retention-time window around its expected elution. At any instant the instrument is monitoring only the transitions that could plausibly be eluting.

SCIEX states the mechanism plainly: scheduling decreases the number of concurrent MRMs monitored at any point in time, which allows both the cycle time and the dwell time to stay optimal at higher levels of multiplexing. A 400-transition panel with a well-distributed retention-time spread might never exceed 40 concurrent transitions — putting it back in the comfortable rows of the table.

The parameter you have to get right is the detection window width, and it is a direct trade-off:

• Too narrow and normal retention-time drift — a new column, a mobile-phase batch, an ambient temperature swing, gradual column ageing — walks a peak partly or wholly outside its window. You lose the front of the peak, the back of the peak, or the compound entirely. This failure is dangerous because a truncated peak still integrates and still reports a number, just a wrong one.
• Too wide and you have reintroduced the concurrency problem you were solving; the windows overlap, concurrent transition count spikes in the busy middle of the gradient, and dwell time collapses exactly where your chromatogram is most crowded.

Practical approach: size the window from measured retention-time variability, not from a default. Run your standards several times across at least two columns or two days, take the observed retention-time spread for each compound, and set the window to comfortably exceed it — then confirm the resulting maximum concurrency, which the vendor software will usually plot as a concurrency-versus-time trace. Read that plot. The peak of that trace, not the total transition count, is the number to put into the dwell-time equation. And verify what SCIEX flags explicitly: that the minimum dwell time the schedule produces at peak concurrency is not below your instrument’s floor.

If your retention times are not reproducible enough to schedule tightly, the fix is chromatographic, not mass-spectrometric — see column and stationary-phase selection and the troubleshooting material in the HPLC guide.

Step 6: qualifying the method against ICH M10

For regulated bioanalysis, the transitions you have just built have to survive formal validation. The governing document is ICH M10, Bioanalytical Method Validation and Study Sample Analysis, adopted at Step 4 in May 2022 and issued as FDA guidance in November 2022 (FDA copy of the M10 guideline). Several of its requirements bear directly on transition design rather than on sample handling:

• Carryover. Carryover in blank samples following the highest calibration standard should not exceed 20% of the analyte response at the LLOQ, and 5% of the internal standard response. M10 expects carryover to be assessed and minimised during method development, and assessed during validation by analysing blanks after a standard at the ULOQ. A transition on a sticky compound can pass sensitivity and fail here.
• Calibration curve. A calibration curve comprises blank samples, a zero sample, and at least six calibration standards including the LLOQ and the ULOQ. See building an analytical calibration curve for the fitting and back-calculation mechanics.
• Accuracy. Back-calculated concentrations should be within ±15% of nominal, except at the LLOQ where ±20% applies; this must be met by at least 75% of calibration standards, with a minimum of six.
• Selectivity. This is where the qualifier transition earns its place. Demonstrating that a signal in the quantifier channel is the analyte, and not an isobaric matrix component, rests on a consistent quantifier-to-qualifier ion ratio between standards and samples.

Read the guideline itself before writing a validation plan against it — the thresholds above are the ones most relevant to transition selection, not a complete list of M10’s requirements, and incurred-sample reanalysis, matrix-effect assessment and stability testing all carry their own criteria. For laboratories operating under electronic-records requirements, the acquisition method and its parameters are themselves regulated records; see 21 CFR Part 11.

Outside regulated bioanalysis — environmental, food-safety and forensic work — the analogous criteria live in the governing method rather than in M10, and ion-ratio tolerances in particular are prescribed differently by different method authorities. Do not carry an M10 threshold into a method that specifies its own.

Troubleshooting: symptom to cause

Frequently asked questions

What is the difference between MRM and SRM?

None, functionally. Selected reaction monitoring (SRM) is the more formally correct term for monitoring one precursor-to-product reaction; multiple reaction monitoring (MRM) describes doing several of them in one method. In practice the two are used interchangeably, with MRM dominant in small-molecule and vendor literature and SRM more common in proteomics. Both describe the same acquisition mode on the same instrument.

How many transitions can a triple quadrupole realistically monitor?

The question has no fixed answer because the limit is on concurrent transitions, not total. With scheduled MRM and a well-spread gradient, panels of several hundred to over a thousand total transitions are routine, because concurrency stays low. Without scheduling, the practical limit is set by cycle time / (dwell + pause), and with a 1.5-second cycle, a 5 ms pause and a usable dwell you will run out of budget somewhere in the low hundreds — well before you run out of method-editor rows.

How many data points do I need across an MRM peak?

Aim for 10–12 across the base width for quantitative work. Vendor guidance also appears in a half-height convention quoting 6–8 points; the two are not equivalent, so confirm which width a given recommendation refers to before applying it. Below roughly 8 points across the base, peak-area precision starts to depend on where the sampling grid falls relative to the apex.

Why do I need two transitions per compound?

One transition gives you a number; two give you a reason to believe it. The quantifier-to-qualifier ion ratio is the primary evidence that the peak you integrated is your analyte and not an isobaric interference — and demonstrating selectivity is a validation requirement, not an optional extra. Some confirmatory frameworks require more than two.

Can I use collision energies from a published method or application note?

As a starting range, yes; as a final value, no. Collision energy depends on collision cell geometry, collision gas identity and pressure, and the vendor’s own convention for expressing the number — so the same physical fragmentation appears as different values on different platforms. Re-acquire the breakdown curve on your instrument.

What is pause time and should I reduce it?

Pause time (interscan or interchannel delay) is the settling period while the quadrupoles switch voltages between transitions; no ions are counted during it. It is charged once per transition per cycle, so at high concurrency it can consume most of your cycle. Reducing it is the highest-leverage change available for large panels — but if the rods have not settled you get intensity loss and cross-talk, which present as chemistry problems rather than as instrument errors. Reduce it deliberately, and verify sensitivity and cross-talk after you do.

My method worked at 20 analytes and fails at 80. What changed?

Almost certainly the time budget. At fixed cycle time, quadrupling the concurrent transitions divides the time slot per transition by four, and the fixed pause time takes a proportionally larger bite of what remains. Recalculate dwell at your actual peak concurrency, and if it has fallen below your instrument’s usable floor, the answer is scheduled MRM rather than a longer cycle time — lengthening the cycle would cost you points across the peak instead.

Does scheduled MRM affect quantitative accuracy?

Not inherently — provided the detection window fully contains the peak. When it does not, the consequence is severe and quiet: a truncated peak integrates to a smaller area and reports a lower concentration without raising an error. Window sizing should be driven by measured retention-time variability across columns and days, not by a software default.

Related CASRAI resources

• SPE cartridge selection and method optimization — the sample-preparation step upstream of every transition here; wash strength decides how much co-eluting matrix suppresses the ions you are monitoring.
• Electrospray ionization: source parameters and how to tune them — the ion source upstream of Q1, and the vendor cross-walk for cone, fragmentor and declustering potential.
• LC-MS explained: how liquid chromatography and mass spectrometry are coupled — ionisation, mass analysers and matrix effects upstream of transition design.
• QTOF vs. triple quadrupole — when targeted MRM quantitation is the right architecture and when it is not.
• Limit of detection vs. limit of quantitation — establishing LOD/LOQ, which dwell time directly affects.
• Building an analytical calibration curve — linear range, weighting and back-calculation.
• GC-MS vs LC-MS — choosing the separation front end.
• Laboratory equipment & instrumentation — the full cluster hub.
• Mass spectrometry proteomics: choosing DDA or DIA, sample prep and run QC — the discovery counterpart, where acquisition is not pre-declared and the QC panel is different.