A 20 µL injection of mobile A on a Monday morning gave seven peaks between 3.2 and 11.8 minutes, the largest at 214 nm with an area of 42 mAU·s. The column was three weeks old. The mobile phase was the same lot that had run clean on Friday. The only thing that had changed was a bag of 0.22 µm PVDF syringe filters opened on Saturday. That is the signature of extractables from an HPLC filter showing up as ghost peaks, and it is almost always the last component anyone suspects because the filter is supposed to be the clean, inert part of the flow path.
I have chased this failure mode on three continents and in more method classes than I care to count. The pattern is consistent: the contamination is real, it is quantifiable, and it is traceable to a specific material, a specific lot, or a specific solvent and temperature combination. What follows is how I isolate it, which compounds bleed from which membrane at which temperature, and the blank and placebo injection design that turns a vague “the baseline is dirty” complaint into a signed corrective action.
Extractables are not a single thing. They are a family of low-molecular-weight species that migrate from the membrane, the housing, the support scrim, or the antistatic and release agents used during manufacturing. The species that reach your detector depend on the polymer, the solvent, and the temperature at which the sample and mobile phase contact the filter. Here is the map I use.
| Membrane material | Dominant extractable class | Detection wavelength / mode | Onset temperature | Worst-case solvent |
|---|---|---|---|---|
| PVDF (hydrophilic) | Fluorinated oligomers, residual NMP or DMAc, low-level surfactants | 210–230 nm, ESI- | ~40 °C (accelerates sharply above 50 °C) | Acetonitrile/water with 0.1% TFA |
| PTFE (hydrophobic) | Perfluoroalkoxy residues, fluorotelomer alcohols | 210–220 nm, ESI- | Stable to 80 °C in most solvents; bleeds in DMSO/DMF | DMSO, DMF, NMP |
| Nylon (polyamide) | Caprolactam monomer, cyclic oligomers, amide plasticizers | 200–215 nm, ESI+ | Onset near 35 °C; strong above 45 °C | Aqueous buffers, methanol |
| PES (polyethersulfone) | Sulfone oligomers, residual solvents | 220–260 nm, ESI- | ~45 °C | Acetonitrile, alkaline buffers |
| Cellulose acetate / RC | Plasticizers, residual acetic acid, carbohydrate fragments | 200–210 nm | Onset near 30 °C | Methanol/water |
| Glass fiber (prefilter) | Binder resins, sodium and calcium ions | Conductivity, ICP-MS | Not temperature-limited | Any aqueous |
Two things about that table matter more than the rest. First, the onset temperature is a real cliff, not a gradual slope. A nylon filter that is clean at 25 °C will start shedding caprolactam at a rate that produces a 5–15 mAU baseline lift at 210 nm once the autosampler tray or the column compartment sits at 40 °C. I have seen a method move from a clean blank to a five-peak ghost pattern purely because a lab raised the sample compartment temperature from 4 °C to 25 °C and the filter spent four hours in the tray before injection. Second, the wavelength you monitor determines whether you see the problem at all. Most of these extractables are weak chromophores. Run your blank at 254 nm and the ghost peaks are invisible. Run the same blank at 210 nm and the filter reveals itself.
The extractable load also scales with filter area and with the number of filters in the path. A 13 mm filter with roughly 1.0 cm² of membrane area releases less total mass than a 25 mm filter with about 3.8 cm², but per unit volume of sample the smaller filter can concentrate the extractables if you push a small volume through it. If your method uses a HPLC filter on a 100 µL sample, the extractable-to-analyte ratio is higher than on a 1 mL sample through the same device. That is one reason small-volume methods are more sensitive to filter contamination and why I default to the smallest diameter that holds the required sample volume.
Blank and placebo injection design that actually isolates the filter
The mistake I see most often is a single “blank” injection that is supposed to prove the system is clean. One injection cannot separate a filter contribution from a mobile phase contribution from a column contribution. You need a bracketed sequence that varies one component at a time. Here is the sequence I run when a ghost peak complaint lands on my desk.
Step 1: System blank
Inject mobile phase that has never touched a filter. If peaks appear here, the problem is the mobile phase, the water system, or the column, and the filter is a red herring. This injection is your floor. Record the largest peak area at your lowest detection wavelength.
Step 2: Filtered mobile phase blank
Take the same mobile phase and pass it through a fresh filter from the suspect lot. Inject. The difference between step 1 and step 2 is the filter’s contribution to the mobile phase, which is usually small because the filter is in contact with the solvent for seconds, not hours.
Step 3: Filtered sample placebo
Prepare the placebo matrix (the sample without analyte) and filter it exactly the way the method filters real samples. Inject. This is where most filter contamination shows up, because the filter is now in contact with the sample matrix, often at an elevated temperature, for the full preparation time.
Step 4: Unfiltered placebo, then filtered placebo, same vial
Split the placebo. Inject the unfiltered half, then the filtered half. Any peak that appears only in the filtered half is a filter extractable. This is the single most useful comparison in the whole sequence because it removes the mobile phase and the column from the equation entirely.
Step 5: Temperature stress
Hold a filtered placebo at 40 °C for two hours and re-inject. If the ghost peaks grow, you have a thermal extractable problem and the fix is a different membrane or a lower preparation temperature, not a longer flush.
Run this sequence at 210 nm and at your method wavelength. The 210 nm trace is your early warning; the method wavelength trace is what your reviewer will see. I also run the filtered placebo on a PTFE membrane filter as a control, because PTFE is the most inert common membrane in most reversed-phase solvents and gives me a clean reference point to subtract.
One more practical note. Use the same autosampler vial and insert for every injection in the sequence. If you switch vial types mid-sequence, you introduce a second variable. I keep a dedicated set of vial insert units for contamination work so the vial is never the suspect. The same logic applies to your COD vial if you run any wet-chemistry confirmation alongside the chromatographic work; a dirty digestion vessel will mimic a filter problem and send you down the wrong path for a week.
Reading the chromatogram: what a filter ghost peak looks like versus what it is not
Filter extractables have a recognizable fingerprint. They tend to appear as a cluster of small, closely spaced peaks in the first third of the run, often with one dominant member and several shoulders. They grow with filter lot, with preparation temperature, and with contact time, and they shrink when you switch to a more inert membrane. They do not shift retention time with mobile phase pH the way an analyte does, and they do not co-elute with your internal standard unless the standard is itself a small polar molecule.
Three lookalikes will waste your time if you do not rule them out early. Mobile phase additive peaks (TFA at 210 nm, ammonium acetate at low wavelength) are present in the system blank and do not change when you change the filter. Column bleed is a rising baseline, not discrete peaks, and it appears in the system blank after the column has been in service. Autosampler carryover appears only after a high-concentration injection and disappears after a blank, which is not the filter pattern. If your ghost peaks survive a system blank and track the filter lot, you have your answer.
Field case one: the nylon filter that only bled on Tuesdays
A contract lab running a stability method on a small-molecule API kept getting a reproducible ghost peak at 4.7 minutes, area 8–12 mAU·s at 215 nm, only on runs prepared after 10 a.m. Morning runs were clean. The method used a 0.45 µm nylon syringe filter and a 25 °C autosampler tray.
The measurement that cracked it was a temperature log of the sample preparation room. The HVAC setpoint was 21 °C, but the bench nearest the window reached 31 °C by mid-morning in summer. A filtered placebo held at 31 °C for 90 minutes produced the 4.7-minute peak at 11 mAU·s; the same placebo held at 21 °C produced nothing above 1 mAU·s. The compound was caprolactam from the nylon membrane, confirmed by LC-MS (m/z 114.1 [M+H]+).
The fix was twofold. The lab switched to a hydrophilic PVDF filter for this method, and the preparation SOP now specifies that filtered samples are capped and placed in a 4 °C autosampler tray within 15 minutes of filtration. The ghost peak disappeared and stayed gone across three lot changes. The lesson is that nylon is fine for many methods but its thermal ceiling is lower than most people assume, and a warm bench is enough to cross it.
Field case two: the PVDF lot with a fluorinated oligomer problem
An environmental lab running EPA method work under SW-846 saw a broad, late-eluting hump at 9–11 minutes on a C18 column, plus a cluster of small peaks at 2–3 minutes, only on samples filtered with a new PVDF lot. The system blank was clean. The unfiltered placebo was clean. The filtered placebo showed the same cluster at 210 nm and nothing at 254 nm.
I ran the filtered placebo by LC-MS in negative mode and found a series of fluorinated oligomers with mass defects consistent with perfluoroalkyl chains. The lot had a residual manufacturing solvent that the vendor’s certificate of analysis did not report. The fix was to reject the lot, switch to a PTFE membrane for the aqueous samples, and add a lot-qualification step: every incoming filter lot gets a filtered-placebo injection at 210 nm against a retained reference lot, and the acceptance criterion is no peak greater than 2 mAU·s in the 2–12 minute window. That criterion has caught two more suspect lots since.
If you want a reference point for what a well-characterized consumable looks like on paper, the technical documentation from Agilent and Thermo Fisher shows the extractable data packages that a serious vendor should be able to supply. Restek publishes useful background on ghost peaks and baseline artifacts that is worth reading before you blame the filter.
Decision checklist before you blame the filter
Work through this in order. Stop at the first yes.
- Does the peak appear in the system blank at 210 nm? If yes, the filter is not the source. Check water, mobile phase, and column.
- Does the peak appear only in the filtered placebo and not the unfiltered placebo? If yes, the filter is the source. Go to the membrane table above.
- Does the peak grow when the filtered placebo is held at 40 °C for two hours? If yes, you have a thermal extractable. Lower the preparation temperature or change membrane.
- Does the peak track the filter lot and disappear with a retained reference lot? If yes, reject the lot and add lot qualification to your incoming inspection.
- Is the peak present at your method wavelength but absent at 210 nm? If yes, it is unlikely to be a filter extractable; look at the sample matrix or the column.
- Does the peak shift retention time with mobile phase pH? If yes, it is an analyte or a matrix component, not a filter extractable.
What to specify so this does not happen again
The procurement fix is a specification, not a hope. For methods that run below 230 nm or above 30 °C, I specify a membrane with a documented extractable profile: hydrophilic PVDF for aqueous and mixed aqueous/organic work up to 40 °C, PTFE for aggressive solvents and higher temperatures, and nylon only where the method has been qualified at the actual preparation temperature. I ask for the vendor’s extractable data at 210 nm in the actual mobile phase, not in water, because water extractables tell you almost nothing about what happens in 0.1% TFA in acetonitrile. I ask for the lot-to-lot acceptance criterion and the retention sample protocol. And I write the lot-qualification injection into the incoming inspection SOP with a numeric threshold, usually no peak greater than 2 mAU·s in the ghost-peak window at 210 nm against the reference lot.
On the operational side, three habits kill most of this failure mode. Filter at the lowest temperature the method allows, and cap and chill the filtered sample within 15 minutes. Never let a filtered sample sit in a warm autosampler tray for an hour while you prepare the next one. And keep a retained reference lot of filter in the freezer so you always have a clean comparator when a new lot arrives.
The decision rule is simple: if a ghost peak survives a system blank and tracks a filter lot, treat the filter as the source until a filtered-versus-unfiltered placebo pair proves otherwise, and set your acceptance threshold at 210 nm rather than at the method wavelength, because the method wavelength will hide the very extractables that are causing the problem.
References & further reading
Related resources from our lab consumables network:
- Hplc Filter at hplcfilters.com
- Ptfe Membrane Filter at ptfemembranefilter.com
- Cod Vial at codvial.com
- Vial Insert at hplcinsert.com
External references and standards: