The future of HPLC autosamplers: trends for the next decade

Over the next decade HPLC autosamplers are likely to get smaller, easier to drive from code, and more often built into robots, reactors and portable analysers instead of sitting in a stack of lab modules. The trends with the strongest evidence today are capillary and portable LC, open interface standards (SiLA 2 and OPC UA LADS), online sampling for process analytical technology, and built-in health checks. Fully autonomous, AI-scheduled labs exist in peer-reviewed papers, but they are not yet routine in QC labs.

Key points

• Capillary LC ships in portable form: one system weighs 8.0 kg, pumps 0.4-10 µL/min at up to 689 bar and injects 4-40 nL.
• SiLA 2 and OPC UA LADS are published with open-source code, but the autosampler datasheets we checked still list serial or Ethernet with vendor software.
• Self-driving labs keep the LC; a robot loads the vials and an algorithm picks the next injection.
• In an Agilent energy comparison, sample cooling was the largest part of the extra energy a modular LC used over a compact one; moving from a 4.6 mm to a 0.3 mm column cuts solvent use more than 200-fold.
• Predictive maintenance and AI method development exist in pieces, not yet as routine, validated practice.

How we sorted the evidence

We put each claim into one of three groups. Real now: a product on sale (checked on its datasheet), a published standard or a regulatory text. Emerging: peer-reviewed demonstrations or early products. Speculative: our extrapolation, with no product or paper behind it yet.

Will HPLC get smaller? Portable and capillary LC

A 2020 review of portable HPLC concluded that "while instrumentation is becoming more capable it has yet to find adoption outside of research" (Rahimi et al., 2020). That has started to change at the capillary end.

The Axcend Focus LC datasheet lists 22.8 × 35.5 × 20.8 cm, 8.0 kg, a binary pump for 0.4-10 µL/min at up to 689 bar (10,000 psi), internal loop injection of 4, 10 or 40 nL, 15 mL solvent and waste vials, and a battery that typically runs more than 10 h without column heating. Automated sampling comes from a separate 40-vial or 96-well module that sits on top of or beneath the LC.

A May 2025 LCGC article describes a mobile LC-MS "lab-in-a-van" for PFAS screening that covered more than 3000 km and 10 sites in a 10-day roadshow, analysing more than 200 environmental samples on site (Mikhail et al., 2025). The authors see "a need for similarly compact and mobile (and if possible automated and integrated) sample preparation technology."

For autosamplers this means two things. Capillary columns need nanolitre injections (see the worked example below). And samplers for conventional columns are shrinking too: the SIELC Alltesta Mini holds 48 vials in 160 × 145 × 195 mm (W × H × L), weighs 3 kg and is rated to 5000 psi (345 bar). Our guide to compact HPLC autosamplers compares what you can buy.

Will instruments speak a common language? SiLA 2 and OPC UA LADS

Two open standards aim to replace one driver per instrument:

• SiLA 2 runs over HTTP/2 with Protocol Buffers, following the gRPC wire format. An instrument exposes "Features" made of Commands and Properties, and clients can discover servers automatically. Version 1.1 added a server-initiated connection method, and the consortium links open-source implementations in C#, Java and Python (SiLA Consortium).
• OPC UA LADS (OPC 30500) "defines a device agnostic information model for analytical and laboratory instruments", with a Hardware View and a Functional View. Version 1.0.0 is dated 30 November 2023 and 1.0.1 (no technical changes) 18 January 2024 (OPC Foundation). It comes from a joint working group with associations including SPECTARIS, VDMA and JAIMA, and example servers and a Python client are on GitHub.

OPC UA is already inside at least one commercial online LC: Agilent says its InfinityLab Online LC Solutions use it to exchange data with sampling devices from multiple vendors (Agilent 5994-6086EN).

On the sampler side, the datasheets we checked for this article listed neither standard. The Alltesta Mini uses an "RS232 Virtual COM port via USB"; the Focus LC uses TCP/IP Ethernet with Axcend's software, Agilent OpenLab or DataApex Clarity. Our own prototype, the Elutrolab AS-48, also takes the serial route, with a documented plain-text command set over a USB virtual COM port; SiLA 2 or LADS support for it is not decided.

Our view (speculative): the practical pattern for the next decade is a thin, documented device protocol with a SiLA 2 or LADS wrapper on top, rather than standards built into every firmware. More in SiLA 2, OPC UA LADS and serial integration.

Where does the autosampler fit in a self-driving lab?

A 2024 Chemical Reviews article surveys self-driving labs, including their hardware, software and integration with lab infrastructure (Tom et al., 2024). Three published systems show where HPLC sits:

1. Flow optimisation. Bédard et al. built a plug-and-play flow synthesis system whose software processes HPLC, MS and vibrational spectroscopy data and runs automated optimisations (Science, 2018).
2. Robot plus HPLC under remote control. ChemOS ran an HPLC auto-calibration on a North Robotics N9 platform with an Agilent 1260 series HPLC: 100 autonomous experiments per session over about seven hours, hardware in Vancouver, software in Cambridge, USA, requests handled first-in, first-out (Roch et al., 2020).
3. Mobile robots sharing instruments. Dai et al. linked an automated synthesis platform, an LC-MS and a benchtop NMR with mobile robots, which "allows robots to share existing laboratory equipment with human researchers without monopolizing it" (Nature, 2024).

None of these replaces the autosampler. They change its job: a robot loads the vials, software adds injections as results arrive, and the sampler must report ready, busy and error states in a form a program can read. See where sampling and injection fit in a self-driving lab.

How green can an HPLC get? Solvent and energy

Solvent: the column diameter rule

A textbook rule of thumb: at the same linear velocity and particle size, flow rate and injection volume scale with the square of the column internal diameter.

F2 = F1 × (d2 / d1)²
V2 = V1 × (d2 / d1)²

Worked example for a 1.0 mL/min method with a 10 µL injection on a 4.6 mm column, run 8 h a day:

The 0.3 mm row falls inside the Focus LC pump range and next to its 40 nL internal loop, which shows why capillary systems inject from nanolitre loops rather than a conventional microlitre sampler.

Energy: sample cooling dominates the extra load

Agilent measured four InfinityLab LC systems through idle, ready and run states over a modelled lab day (Rieck, 2020). At 40 samples a day, a gradient 1220 Infinity II used about 190 kJ per sample and a modular 1260 Infinity II with sample thermostat about 380 kJ: "Cooling the samples, however, accounted for the greatest part of increased energy consumption." A 1290 Infinity II running 16 h a day reached 88 kJ per sample, a third of the isocratic 1220 (267 kJ), so throughput matters as much as hardware. The lesson for autosamplers: cool only samples that need it, and compare energy per sample, not per day.

Labels and scores

My Green Lab's ACT Ecolabel uses a 100-point weighted score covering energy use, materials, chemical hazards and end-of-life options, with claims validated by a third party; the 2020 Agilent note reports ACT labels for its 1220, 1260 and 1290 Infinity II families. For methods, the open-source AGREE calculator turns the 12 principles of green analytical chemistry into a 0-1 score, and one principle is that automated and miniaturised methods should be selected.

Will more HPLC move into the process? Online and at-line PAT

The FDA's PAT guidance (2004) defines at-line ("the sample is removed, isolated from, and analyzed in close proximity to the process stream") and on-line ("the sample is diverted from the manufacturing process, and may be returned to the process stream") measurement. ICH Q13, adopted on 16 November 2022, names "on-line HPLC to monitor conversion of a chemical reaction" as an example of PAT in continuous manufacturing.

Agilent's Online Sample Manager has an external sampling interface for reactors, direct injection or vial sampling with automated dilution, sampling volumes of 0.1-100 µL and a retain-sample function for later offline analysis (Agilent 5994-6086EN). Axcend offers an InFocus module for on-line and at-line monitoring with its capillary LC.

The skills that matter here differ from tray work: drawing from a flowing line, quenching or diluting before injection, keeping a retain sample, and running unattended for days. See automated reaction monitoring with online HPLC.

Autosamplers as modules inside other instruments

Spark Holland supplies autosamplers, pumps and column ovens for integration into partners' products, and KNAUER sells pumps, detectors, liquid handlers and valves under the customer's label. SIELC lists a customisable Alltesta Mini with 96- or 384-well plates and extra valve options.

Embedding brings its own requirements: low-voltage DC power, one control board, a documented protocol, mounting points and service parts the integrator can stock. Our OEM integration checklist goes through them.

Self-checks, sensors and predictive maintenance

Thermo Fisher says "certain Thermo Scientific HPLC and UHPLC systems allow users to set automated System Health Checks", which can be scheduled, and its Vanquish Solvent Monitor warns if there is not enough eluent or waste capacity to finish the sequence (Thermo Fisher).

In an autosampler, the valuable sensors catch silent failures: a valve that did not reach its position, a needle crash, a pressure change on the sample line, a leak. In the AS-48 prototype, for example, an absolute angle sensor confirms every valve switch, and a power-up self-check covers homing, valve position, door and pressure sensors.

Predictive maintenance is thinner on evidence. Forecasting rotor seal or needle seat wear from logged pressure and cycle counts is plausible, but we found no peer-reviewed validation of such models for autosamplers. Ask vendors what data their predictions rest on.

Can AI develop methods and schedule injections?

Closed-loop method development works in published research. Boelrijk et al. connected a Bayesian optimisation algorithm directly to an LC and let it design gradients unsupervised; for two dye mixtures (more than 18 and more than 80 components), both single- and multi-objective strategies "found satisfactory optima in under 35 measurements" (Analytica Chimica Acta, 2023). The authors note that scalability may be limited by the number of parameters optimised at once.

Scheduling is simpler than the word AI suggests: ChemOS queued requests first-in, first-out, and the Dai et al. robots used a heuristic decision-maker. The near-term change for autosamplers is sequences that can be edited while running, so an algorithm can add a re-injection when a result arrives. Unattended development and validation of regulated methods by AI remains speculative.

What to ask for if you buy or build an autosampler now

1. A documented control protocol that reports status and errors, not only a CDS driver. Our guide to controlling serial lab instruments from Python shows the basics.
2. A footprint and power supply that fit where the samples are: a fume hood, a reactor skid or a robot deck.
3. An injection range matched to the column: nanolitres for capillary, single microlitres for 2.1 mm, tens of microlitres for 4.6 mm.
4. Sample cooling only if your samples need it.
5. Self-checks that catch valve, needle and leak faults before a sequence wastes samples.
6. Field-replaceable wear parts and published service intervals.

Most of the parts already exist. The next decade is about building them into instruments that are small, scriptable and honest about their own state.