Pharmaceutical Water Generation Technologies: RO, EDI, UF and Distillation Compared

On a water system project, the difficult meeting is not the one where the supplier is chosen. It is the one where somebody has to write into the design report why the generation train is double-pass RO plus EDI rather than a still, or exactly the other way round. Whoever signs that paragraph knows that three years later an inspector will read it with the loop's microbiological trends, the sanitisation records and the TOC deviations in front of them.

Commercial literature is no help in writing it: it answers with superlatives, with rejection percentages quoted under standard test conditions, and with economic comparisons built on assumptions that are not yours. The useful question is not which technology is superior, because no answer holds independently of the site, the source water and the intended use of the water. The question is which criteria govern the choice, what data is needed to apply them, and how the reasoning is documented so that it still stands years later.

Scope: where generation starts and ends

This article covers the generation stage: from the water leaving pretreatment to the compendial water entering the storage vessel. Three things sit outside that boundary while heavily conditioning the choice: the water grade required at each point of use, which is set by intended use and not by the technology available (Purified Water or WFI: how to select the water quality); source water characterisation and pretreatment, which determine much of the downstream performance (Source water and pretreatment); and storage and distribution, where microbiological risk materialises regardless of how good the water is at the generator outlet.

The boundary is not academic: many arguments about the superiority of a technology are in fact arguments about an undersized pretreatment or a badly designed loop, blamed on the generator because it is the most visible object in the system.

The regulatory perimeter: what is allowed and what must be notified

USP defines Water for Injection as "water purified by distillation or a purification process that is equivalent or superior to distillation in the removal of chemicals and microorganisms". The definition is functional: it describes the required outcome, it does not mandate a technology.

In Europe, Ph. Eur. monograph 0169 Water for injections was revised with effect from 1 April 2017 to allow WFI to be produced by a purification process equivalent to distillation. EMA guideline EMA/CHMP/CVMP/QWP/496873/2018, in force since 1 February 2021, identifies such processes in section 4.2 as single- or double-pass RO coupled with EDI, ultrafiltration or nanofiltration. The same document requires prior notification to the supervisory GMP authority before introducing RO. This is not a formality to be handled once the plant is running: it is an obligation to place in the project schedule alongside qualification, because it conditions the date on which the water can be used in production.

EMA Q&A EMA/INS/GMP/443117/2017, in force since 1 August 2017, sets out inspection expectations for these technologies. It notes that RO typically operates at ambient temperature and therefore provides an ideal environment for biofilm formation; it expects extended testing, daily testing of all critical points in the initial phase, and data collected over approximately one year to capture seasonal variation; it mentions destructive membrane analysis to confirm the absence of biofilm, and thermal treatment above 75 °C with contact times left to the manufacturer's validation. It also contains a sentence that works as a reading rule for any trend: "Increasing of such limits is not good practice and may mask a failing system".

Annex 1 clause 6.10 requires WFI to be produced from water meeting specifications defined during qualification, by distillation or by a purification process equivalent to distillation, and to be stored and distributed so as to minimise microbial growth, citing constant circulation above 70 °C as an example; clause 6.15 requires continuous monitoring on WFI systems, such as TOC and conductivity. WHO TRS 1025, Annex 3 (2020) deals specifically with the production of WFI by means other than distillation and requires validated thermal and/or chemical sanitisation at specified intervals, leaving those intervals to the manufacturer.

On technical standards, ISO 22519:2023 (2nd edition, 4 April 2023) covers membrane-based WFI generation. Two cautions: the 1st edition, ISO 22519:2019, has been withdrawn since 4 April 2023, so specifications and URS documents citing it must be updated before going out to tender; and ISO 22519 sits under ISO/TC 282 Water reuse, not a pharmaceutical committee, which does not make it irrelevant but does place its weight correctly against a compendial text. The ISPE Good Practice Guide "Membrane-Based WFI Systems" (May 2022) is the dedicated industry guide and, like all ISPE guides, is not regulatory.

None of these texts tells you which technology to choose. They define the required outcome and the robustness of the expected evidence. The choice remains a design decision, to be justified through a structured risk assessment along the lines described in Quality Risk Management applied to pharmaceutical water systems.

The technologies, one by one

Single-pass reverse osmosis

Pressure-driven separation across a semipermeable membrane. It retains dissolved ions, organic molecules, endotoxins and microorganisms to an extent that depends on the membrane type, the operating conditions and the quality of the feed. Rejection figures quoted by manufacturers are nominal values referred to standard test conditions: they are not transferable to your plant without verification on your water.

Performance depends on source water composition (hardness, silica, chlorides, iron, organic load, fouling tendency), temperature, pressure, the recovery set point, pretreatment and the flushing regime; it is undermined by residual oxidants on incompatible membranes, precipitation of sparingly soluble salts, organic and colloidal fouling, and operation at ambient temperature under reduced flow. The most delicate GMP point is not chemical rejection, which conductivity and TOC observe well: it is that the integrity of the barrier cannot be verified by a simple, conclusive test in the way a sterilising filter can. Control is indirect and trend-based, which is precisely why the EMA Q&A insists on an extended data set and a year of observation.

Double-pass reverse osmosis

The permeate of the first stage feeds a second stage. It serves two purposes worth keeping distinct in the justification: reducing the residual load and, above all, increasing the margin against feed variability. The cost is not only capital: it is a second section to qualify, sanitise, monitor and maintain, with additional wetted surface and new potential stagnation points. The incremental energy and water consumption is real but is not a universal number: it must be calculated on your own mass balance, not assumed from a generic comparison.

EDI

Continuous deionisation combining ion-exchange membranes, resins and an electric field, with continuous electrochemical regeneration. The structural advantage over batch-regenerated resin beds is twofold: no chemical regenerants, and none of the microbiological risk windows and effluent handling associated with regeneration cycles.

It sits almost always downstream of RO because it requires a feed that is already low in ionic load, and it is sensitive to residual hardness, silica, dissolved CO2 and residual oxidants: those parameters therefore become design criteria for pretreatment and for the RO section, not only for the EDI. Thermal sanitisability depends on module construction and must be stated as an explicit URS requirement: taking it for granted and discovering its absence during qualification of the sanitisation cycles is one of the most frequent mistakes.

Ultrafiltration

A physical barrier by molecular size. It retains microorganisms, cell wall fragments, endotoxins and colloids; it does not remove dissolved ions, so it does not replace the demineralisation section but completes it. In membrane-based WFI trains it is typically the final stage, placed as close as possible to the point where the water leaves the generator, because it guards the very attribute that distinguishes WFI from PW. Unlike RO, a UF module allows forms of integrity verification: method and frequency must be agreed with the manufacturer and built into the qualification and maintenance plans. The conceptual limit is that UF acts on what reaches it: it does not protect against contamination generated downstream and does not justify a badly designed loop.

Nanofiltration is cited by the EMA guideline among the processes that can be coupled with RO; it has an intermediate cut-off and remains a niche choice, to be justified on a specific source water characteristic rather than adopted by analogy.

Distillation

A phase change: water evaporates and recondenses, leaving salts, particulates, endotoxins and microorganisms behind. Two properties make it the historical reference, and the reason the other routes are described as equivalent to distillation: the separation mechanism does not depend on the integrity of a membrane, and the process operates at high temperature, a condition hostile to microbial proliferation and consistent with hot distribution.

It is not free of criticalities. Distillate quality depends on how effectively entrained droplets are separated, therefore on the feed and on correct operation; the feed must in any case be demineralised upstream, because scaling and corrosion of the heat exchangers are the typical failure modes; multiple-effect and vapour compression configurations have very different utility profiles, some requiring clean steam, others mainly electrical power. Maintenance is less frequent but heavier, and availability depends on utilities that may not be under the department's control.

You do not choose a technology, you choose a train

None of these technologies produces compendial water from potable water on its own. The real choice is the sequence, and every sequence moves the risk to a different point in the system.

Source water weighs more than the technology

Two sites buying the same skid get different results if the feed water differs. Before comparing technologies you need a documented picture of the feed over a period covering seasonal variation: Annex 1 clause 6.8 explicitly requires qualification and validation of water systems to take seasonal variation into account, and the EMA Q&A asks for roughly one year of data for the same reason.

What moves the decision is the ionic load and its composition, silica, organic load, iron and manganese, colloids and fouling tendency, residual oxidants in the mains supply, feed temperature and its annual swing, seasonal contaminants linked to weather events, and the reliability of the supply itself. Water with strong seasonal variability shifts the decision criterion from peak performance to robustness: this is the case where an additional stage is justified not by "better quality" but by stability.

Microbiological and endotoxin control

This is where the practical difference between technology families concentrates, and it must be stated precisely: not "membranes are less safe", but "membranes operate under conditions in which microbiological risk must be actively controlled, and that control must be designed, validated and demonstrated".

For membrane systems the EMA position is explicit: ambient temperature equals an ideal environment for biofilm. Concrete design consequences follow, to be settled before purchase and not at start-up: flow maintained even during periods of no draw-off, no stagnant sections, hygienic design of sections and instruments, the ability to sanitise every section without bypassing it, and sampling points located where biofilm forms rather than where sampling is convenient.

For distillation the microbiological risk within generation is structurally low, but it does not disappear from the system: it migrates upstream, into the section feeding the still, and downstream, into the vessel and the loop. Replacing a judgement on the system with a judgement on the generator alone is one of the most frequent errors in comparative assessments. The same applies to endotoxins: phase change and size-exclusion barriers are both effective mechanisms, but the outcome depends on controlling the upstream load and preventing downstream recontamination. And an adverse trend is never managed by raising the limits: the EMA wording on this point is unambiguous and is used as a criterion during inspection.

Sanitisability: the criterion you pay for over the whole plant life

Sanitisability is not an accessory: it determines operating cost and risk profile across the entire life cycle. Annex 1 clause 6.12 requires sterilisation, disinfection or regeneration according to a predetermined schedule and as a remedial action after out-of-limit results.

The questions to put in the URS before receiving quotations are few: is every section sanitisable thermally, chemically or both, by what procedure and with what parts excluded? Does sanitisation require the plant to be shut down, and for how long? What temperature and chemical compatibility limits does the manufacturer state for membranes, modules, gaskets and instruments? The EMA Q&A mentions thermal treatment above 75 °C while leaving contact times to validation: the cycles must therefore be qualified on your own plant, following the logic described in Qualification of sanitisation cycles. A system that tolerates more aggressive and more frequent sanitisation leaves you room to recover when trends deteriorate; a system that does not leaves you one option: stop and replace.

Energy, water recovery, redundancy

More invented numbers circulate on these three criteria than on any other. The rule is simple: do not use generic brochure values. Ask each bidder for specific energy consumption and recovery calculated on your feed specification and your draw-off profile, with the assumptions stated, and turn those values into a condition verifiable at FAT and in PQ. A number that cannot be verified at acceptance is not a design datum: it is a promise.

On recovery what counts is the whole chain, not just the permeate-to-feed ratio: where the concentrate goes, whether it can be reused in non-GMP services, and what local discharge constraints apply. On energy an honest comparison includes every utility involved, including clean steam where the configuration requires it and any heat recovery, not just the electrical load of the generator.

Redundancy is decided from the consequence of unavailability, not from habit: how many hours of downtime a failure on the most critical section generates, how long restoration and requalification take, and what happens to production meanwhile. Only then does it make sense to discuss duplicating the whole train, a single section, or nothing.

Impact on qualification and on the regulatory path

• Notification: introducing RO requires prior notification to the supervisory GMP authority under EMA guideline 496873/2018, with its own lead time, before start-up.
• Extent of the data set: for membrane systems the EMA Q&A expects extended testing, daily testing of critical points in the initial phase, and about one year of data. That is time, laboratory capacity and cost, and it belongs in the project plan.
• Control strategy: Annex 1 clause 6.13 requires alert levels based on initial qualification data and periodically reviewed. Online instrumentation is not an optional line item in the comparison between options.
• Methodological reference: Annex 15 has been in operation since 1 October 2015; the concept paper on its revision (9 February 2026, consultation closed 9 April 2026) is a consultation document, not a requirement. ASTM E2500-25 provides the science- and risk-based approach to specification, design and verification.

TCO: building it without inventing numbers

A credible TCO comparison is not a number taken from a benchmark: it is an itemised model in which every value has a traceable source, declared as a binding quotation, a site historical figure or an explicit assumption. The items most often forgotten are the ones that do not appear in quotations: annual utilities, including clean steam and water lost to drain; consumables and spares with their expected life, probes and recalibration included; operating labour for running, sanitising, sampling and testing, bearing in mind that on a membrane system the initial analytical load is appreciably higher; initial qualification, requalification, change management and periodic review; the expected cost of unavailability and of investigations; major replacements and automation obsolescence. The comparison must use the same horizon and the same production profile for all options: different horizons are the most common cause of reversed conclusions.

Decision matrix

The matrix is deliberately delivered with the weighting column empty and with no pre-filled scores. Weights depend on your context: intended use of the water, product criticality, source water variability, local utility cost, available skills, and the company's position on risk. Assigning them for you would mean deciding for you.

Worked example: Site Delta

Site Delta is a teaching example, not a real plant. It produces non-sterile forms and one small-volume sterile line, and must replace a generator at the end of its life. The choice is between a fully membrane-based train for both PW and WFI, and a membrane train for PW with a dedicated still for WFI.

Applying the matrix, the discriminating criteria are not the expected ones. Achievable chemical quality does not separate the options: both meet the compendial attributes on the basis of the feed characterisation. Three elements separate them. Historical mains water data show a marked seasonal swing in organic load, which weighs on the robustness criterion. The qualification plan for the fully membrane-based option carries a high initial analytical load and prior notification to the GMP authority, with schedule consequences the project had not considered. Finally, the site has reliable clean steam and staff already trained on thermal equipment, but no experience of membrane systems for WFI.

Site Delta's conclusion is not that distillation is better. It is that, with its own weights, "skills and maintainability" and "qualification effort" outweigh the operational advantage expected from the membrane option. A site with the same water, without clean steam and with a team already experienced in membranes would legitimately reach the opposite conclusion, and its documentation would be equally defensible. That is exactly the point: the justification does not lie in the technology, it lies in the traceability of the reasoning.

Frequent mistakes and red flags

• Choosing the technology before characterising the source water over a full seasonal cycle.
• Comparing quotations built on different feed assumptions, flow rates and time horizons.
• Treating prior notification for RO as something to handle after start-up.
• Assuming thermal sanitisability of a section without having required it in the URS and verified it in the offer.
• Assuming that a final UF compensates for a badly designed loop.
• Copying another site's alert limits instead of deriving them from your own initial qualification.
• Citing ISO 22519 in a specification without stating the edition, thereby invoking the withdrawn one.

Red flags during selection: rejection figures quoted without reference conditions; offers that do not distinguish performance guaranteed at FAT from typical performance; energy consumption quoted without calculation assumptions; the answer "the system does not need frequent sanitisation" with no supporting data; the absence of an explicit biofilm control strategy in a membrane train; and the proposal to raise in-house limits in response to an adverse trend.

Checklist before freezing the choice

• Source water characterisation documented over a seasonal cycle.
• Sanitisation requirements written into the URS before the request for quotation is issued.
• Decision matrix completed with weights assigned and minuted by the project team.
• TCO comparison on identical horizon and assumptions, with a source for every value.
• Qualification plan sized on the chosen technology, with PQ duration and analytical load stated.
• Regulatory obligations identified, including prior notification where RO is involved.
• Skills gap analysis and training plan approved together with the technical choice.

The full picture of the cluster sits in the Pharmaceutical Water & WFI Systems hub; the specific comparison between membrane-based WFI generation and distillation is developed in Membrane-based WFI generation versus distillation.

If you work on topics like this, The Pragmatic GMP collects technical and regulatory analysis along the same editorial line, with no promotional content.

Key takeaways

• Neither the compendial texts nor the guidelines point to a preferred technology: they define the required outcome and the robustness of the evidence. The choice remains a design decision to be justified.
• Source water and pretreatment explain much of the performance difference attributed to the generator.
• In membrane trains the dominant risk is biofilm, with explicit inspection expectations on the extent and duration of testing; with distillation the risk migrates upstream and downstream of the generator.
• Sanitisability, qualification effort and available skills weigh as much as chemical performance.
• A TCO comparison is only worth anything if built on identical horizon, assumptions and production profile, with a traceable source for every value.

References

• Ph. Eur. 0169 Water for injections, revision effective 1 April 2017; 0008 Water, purified; general chapters 2.2.38 and 2.2.44.
• USP Purified Water and Water for Injection; chapters <643>, <645> and <1231>.
• EMA/CHMP/CVMP/QWP/496873/2018, in force since 1 February 2021 — ema.europa.eu; EMA/INS/GMP/443117/2017 Q&A, since 1 August 2017.
• EudraLex Volume 4, Annex 1 (in operation since 25 August 2023) and Annex 15 (since 1 October 2015) — health.ec.europa.eu; WHO TRS 1025 Annex 3 (2020) and TRS 1033 Annex 3 (2021).
• ISO 22519:2023, 2nd edition — iso.org. The 1st edition of 2019 is withdrawn.
• ISPE GPG Membrane-Based WFI Systems (2022); ISPE Baseline Guide Vol. 4 Water and Steam Systems, 3rd ed. (2019); ASTM E2500-25; ICH Q9(R1).
• FDA Guide to Inspections of High Purity Water Systems (1993), reference material for investigators, not binding — fda.gov.