Source Water and Pretreatment for Pharmaceutical Water Systems: What Must Be Characterised and Controlled

A Purified Water system passes PQ in March and starts producing microbiological excursions in September. The investigation focuses on the loop, the point of use, the sanitisation cycles. The data nobody had collected was the behaviour of the source water: feed temperature several degrees higher between spring and late summer, organic load in the catchment rising over the same period, residual disinfectant in the mains changed by the utility. The loop had not changed. The incoming water had, and pretreatment had been sized on a single certificate of analysis attached to the quotation.

This is the recurring pattern in pharmaceutical water investigations: the symptom appears downstream, the cause sits upstream. Pretreatment gets the least attention in the URS, the least budget at purchase and the least instrumentation in operation, and it is the section that decides whether downstream generation will run under stable conditions or in permanent firefighting mode. This article covers two concrete things: which source water attributes must be characterised and why, and how to build a characterisation plan that holds up in qualification.

Why pretreatment generates the largest share of downstream problems

It is the only section whose inlet is outside the site's control. Everything downstream receives a fluid already processed by a qualified unit; pretreatment receives potable water whose composition depends on a third-party utility, a well or a surface catchment, and which varies for meteorological and maintenance reasons nobody reports to the department. Variability is not a defect: it is the boundary condition.

It is designed to remove the residual disinfectant. Potable water arrives protected by free chlorine or chloramines; downstream polyamide membranes and resins do not tolerate oxidants. Dechlorination is therefore mandatory, and it creates, in the heart of the plant, a stretch of undisinfected water at ambient temperature, rich in nutrients and with enormous specific surface areas. Q&A EMA/INS/GMP/443117/2017 states this for reverse osmosis: operation at ambient temperature is an ideal environment for biofilm formation. The same applies, earlier in the train, to the beds that precede it.

It is historically under-instrumented. A loop has continuous conductivity and TOC; a softener often has only a volumetric meter and a manual hardness test. Its drifts therefore become visible only once they have already produced an effect on generation: falling rejection, rising differential pressure, rising microbial counts at the point of use.

Where pretreatment starts and where it ends

The boundary must be fixed in writing in the URS and on the P&ID, so that every interface has an owner. Upstream: the potable water delivery point. If a storage tank, a break tank or a booster set exists, those items belong to the pharmaceutical water system from a microbiological risk standpoint, even when they sit with facility management. A buried tank that cannot be inspected or sanitised upstream of a PW plant is a gap in scope, not a plumbing detail. Downstream: the inlet of the first generation unit. The specifications at that point are simultaneously the acceptance criteria for pretreatment and the feed requirements of the generation supplier, and they must be the same document. Minor interfaces must not be left orphaned either: chemical dosing and its storage, stand-by recirculation, by-passes, sampling connections, temporary site connections.

The boundary also determines the qualification perimeter. A unit declared a "non-GMP utility" whose outlet feeds PW generation directly is a difficult position to defend: Annex 1 clauses 6.1-6.5 require risk-commensurate controls, and the risk criterion applies along the whole chain leading to product contact.

What the regulatory framework says, and what it leaves to the site

The starting point is compendial. The USP Purified Water monograph states that the water is obtained by a suitable process and is prepared from water complying with the U.S. EPA National Primary Drinking Water Regulations, or with the drinking water regulations of the European Union or of Japan, or with the WHO Guidelines for Drinking Water Quality. That is not a recommendation: it is part of the definition. If the feed is not potable against one of those references, the water produced does not meet the monograph, whatever the outlet conductivity and TOC.

21 CFR 211.48(a) is the only provision in Part 211 that names water explicitly: it requires potable water to be supplied under continuous positive pressure in a plumbing system free of defects that could contribute contamination to any drug product. Two operational requirements, not analytical ones, and no numerical criterion beyond "potable".

Annex 1 of EudraLex Volume 4, in operation since 25 August 2023, adds the two decisive clauses. Clause 6.7 requires the water treatment and distribution plant to be designed, constructed, installed, commissioned, qualified, monitored and maintained so as to prevent microbiological contamination and to minimise particulates, microbial proliferation and endotoxins or pyrogens. Clause 6.8 requires systems to be qualified and validated for physical, chemical and microbiological control taking account of seasonal variation. That phrase is the regulatory mandate for the characterisation plan.

The EMA Q&A, in force since 1 August 2017, adds the time expectation: extensive testing and daily testing of all critical points in the initial phase, with data over approximately one year to capture seasonal variation. WHO TRS 1033 Annex 3 (2021) anchors alert and action levels to the system's historical data; guideline EMA/CHMP/CVMP/QWP/496873/2018, in force since 1 February 2021, defines the minimum grade of water as a function of intended use but sets no numerical limits and no frequencies. On the non-regulatory technical side there remain ISPE Baseline Guide Vol. 4 (3rd ed., 2019), the informational USP chapter <1231>, which explicitly covers source water, and, as reference material for investigators rather than binding guidance, the FDA "Guide to Inspections of High Purity Water Systems" (1993) and Inspection Technical Guide No. 36 on reverse osmosis (1980).

The point to grasp: none of these sources publishes an inlet limit for hardness, chlorine, silica, SDI or TOC, nor a regeneration or sanitisation frequency. Those values are design specifications the site sets, justifies and documents. Looking for the number in the regulation is looking in the wrong place; copying it from a supplier's manual means adopting a specification without owning it.

Building a characterisation plan that holds up in qualification

A characterisation plan is a protocol, not a folder of certificates: purpose, perimeter, methods, duration, data analysis, signed conclusion. Whoever writes it must answer three questions.

Which attributes, and for which downstream decision

Every attribute enters the plan because it governs an identifiable technical decision, not because the laboratory can measure it.

The attributes the utility certifies by law are a subset of this list, not the whole of it. The potability certificate answers "can this water be drunk?"; the plan answers "can this water, in all its conditions, be treated by the plant I am buying?".

For how long and at what frequency

The defensible minimum duration follows from Annex 1 clause 6.8 and from the explicit expectation in the EMA Q&A: a full annual cycle, because only a year contains every seasonal condition. Frequency must be graded: denser for attributes that vary quickly (temperature, residual disinfectant, turbidity, microbial load), sparser for the geologically stable attributes of a deep aquifer. Any departure from this pattern must be justified by reasoning, not by laboratory availability.

The realistic case is that the project cannot wait twelve months. The defensible answer is not to shrink the plan but to declare it in phases: intensive initial characterisation, sizing on a worst case estimated from available historical data (multi-year series from the utility, neighbouring sites, local hydrogeological literature, all declared as such), completion of the annual cycle during PQ and the first operating phase, with a formal review criterion should the data contradict the design assumptions. That is documented risk management; reducing the plan to a certificate is an undeclared assumption.

Where samples are taken and who owns the data

Sampling points must include at least the delivery point, the outlet of every intermediate storage and the generation inlet: sampling only at delivery hides exactly what matters, namely the contribution of the tank and of the internal distribution network. Methods must be declared together with quantification limits, because a "not detected" without an LOQ is not a datum. Correlation with external events is then required (heavy rainfall, mains work, source changes, site shutdowns), together with a formal channel to the utility: notification of a switch from free chlorine to chloramine is a change control for the site, even if the utility does not treat it as one. The link to the water system risk assessment is what turns this collection into traceable decisions.

Unit operations: what they solve and what they introduce

There is no universally correct pretreatment sequence: the correct one follows from the data for that specific source water and from the feed requirements of that specific generation technology. A unit present because "that is how it is done" rather than because a datum requires it is additional wetted surface and additional microbiological risk, with no return.

Free chlorine and chloramines are not the same problem. Free chlorine is relatively easy to remove and to measure. Chloramines have different removal kinetics on activated carbon, require a different contact time and release ammonia, which passes the RO and reappears downstream as ionic load and as a nitrogen nutrient. A plant sized for free chlorine and fed with chloraminated water does not have a reduced margin: it is outside its design specification. This is why the plan must distinguish the two forms, and why the information channel with the utility is a control element.

Softening and antiscalant are a trade-off, not an obvious alternative. Softening removes the problem at source but introduces a bed, a regeneration, salt storage and a microbiologically sensitive mass. Antiscalant eliminates the bed but adds an organic compound upstream of the RO and shifts the risk from microbiology to chemistry. The choice depends on the hardness and alkalinity data, on design recovery, on the presence of hot surfaces downstream and on the site's real ability to manage chemical dosing in a qualified way.

The microbiological paradox of pretreatment

Pretreatment is the only section whose function requires locally degrading microbiological control: removing the disinfectant is necessary to protect the membranes, and removing it creates the ideal condition for biofilm precisely where specific surfaces are largest and the temperature is ambient. Every granular bed is an unintended immobilised-biomass reactor.

It follows that pretreatment must be treated as a microbiologically active system, not as a mechanical accessory. Periodic release of biofilm fragments produces peaks that appear downstream as isolated, random events; their structure becomes readable only by comparing counts upstream and downstream of each unit. Mechanisms and countermeasures are developed in the article on microbiological and endotoxin control; here the design principle is enough: every component must be sanitisable by a method compatible with its materials, and that compatibility must be verified at specification stage, not at the first out-of-limit event.

Sanitisation of pretreatment

Annex 1 clause 6.12 requires sterilisation, disinfection or regeneration to be carried out according to a predetermined schedule and as a remedial action following out-of-limit or out-of-specification results: a plan established in advance and a structured response to deviation. WHO TRS 1025 Annex 3 (2020), for the production of WFI by means other than distillation, requires validated thermal and/or chemical sanitisation at specified intervals, the intervals being set by the manufacturer. The EMA Q&A mentions thermal treatment above 75 °C while explicitly leaving contact times to the manufacturer's validation.

No source fixes a frequency. Frequency is built: start from the qualification data and the monitoring of the intensive initial phase, identify the characteristic regrowth time of each component, choose an interval with margin against that time, verify it in the field and review it against trending data. The method is constrained by materials: many components, resins and certain carbon configurations in particular, do not tolerate the thermal cycle applicable downstream and require chemical agents with a rinsing plan and verification of the absence of residues. Cycle construction and validation are covered in the article on the qualification of sanitisation cycles.

Monitoring: from compliance checking to trending

Monitoring of pretreatment has two functions that are often confused. The first is protective and must be fast: measuring residual disinfectant immediately upstream of the RO exists to stop the plant before a membrane is damaged, and it must be designed with the reliability, interlock logic and redundancy of a safety function, not of an indicator. The second is diagnostic and continuous: differential pressures, hardness at the softener outlet, temperature, intermediate microbial counts and regenerant consumption, placed in time series, show the drift before the failure.

Annex 1 clause 6.13 requires regular and ongoing chemical and microbiological monitoring, with alert levels based on initial qualification data and periodically reviewed on the basis of requalification, routine monitoring and investigations; clause 6.14 requires alert excursions to be documented, reviewed and investigated, distinguishing an isolated event from an adverse trend or system deterioration. Applied to pretreatment, this means levels are not inherited from another site and not copied from a specification sheet: they are derived from one's own data. The EMA Q&A is blunt on this point: "Increasing of such limits is not good practice and may mask a failing system". Raising an alert level because it is exceeded too often is the wrong answer to the right question.

How the plan holds up in qualification

Annex 1 clause 6.10 establishes that WFI is produced from water complying with specifications defined during qualification. That is the link that closes the argument: source water characterisation does not end with the purchase order, it is the source of the acceptance criteria verified in IQ, OQ and PQ and re-verified at every requalification.

The defensible documentary chain has five links, and each must cite the previous one: characterisation plan → report with the worst case identified and justified → feed specifications in the URS and in the contract → acceptance criteria in the qualification protocols → alert and action limits for routine monitoring. If an inspector asks where the hardness figure used to size the softener came from, the answer must be a signed report with the data, not a line in a supplier's spreadsheet.

A PQ run entirely within a single season demonstrates operation under one boundary condition only. What clauses 6.8 and 6.13 require is a PQ covering the conditions known to be critical for that source water, or a PQ completed by a declared extended monitoring phase, with review criteria, until the annual cycle is covered. The overall picture of the cluster is in the pharmaceutical water and WFI systems hub.

Worked example: Site Delta

Site Delta is a fictitious site used as an example. It manufactures non-sterile oral solid dosage forms, feeds a PW system from the municipal mains through a buried tank inherited from an earlier extension, and has multimedia filtration, softening and activated carbon as pretreatment, followed by single-pass RO and EDI. It holds six years of utility certificates and no data of its own.

When it starts the RO replacement project it opens a characterisation plan with sampling at the delivery point, at the buried tank outlet and at the generation inlet. Three findings emerge that no certificate contained. The microbial count at the tank outlet is systematically higher than at delivery, with a gap that widens in the warm months: the tank is not a neutral part of the network. The utility alternates two sources according to the season, with different ionic profiles, and had never communicated this because both are potable. The residual disinfectant at delivery is in combined form for part of the year, which changes the dechlorination requirements.

Three traceable decisions follow. The sizing worst case becomes the combination of the higher ionic load source with summer temperature, not the average of the certificates. The buried tank enters the water system perimeter, with an inspection and sanitisation plan and a replacement assessment. A communication agreement is established with the utility on source and disinfectant changes, managed as an internal change control.

Decision matrix: scaling control strategy

The weight column is deliberately empty: weights are assigned by the site, according to its own context and the profile of its own source water.

Frequent mistakes and red flags

• A single certificate as the design basis: it describes one instant and contains neither the variability nor the worst case.
• No sampling on the internal network: it excludes from characterisation the components the site is directly responsible for.
• Potability mistaken for process suitability: the first is a compendial prerequisite, the second a design specification.
• A sequence copied from another site: it introduces unnecessary units or omits necessary ones.
• Pretreatment outside the qualification perimeter and the sanitisation plan: it is the section with the highest microbiological load in the plant.
• Alert levels raised after recurring exceedances: identified by the EMA Q&A as a practice that may mask a failing system.
• No formal channel with the potable water supplier: a source or disinfectant change is a modification to the site's process, decided by someone else.
• Residual oxidant measurement treated as a mere indicator: without interlock logic, calibration and periodic verification it protects nothing.

If this kind of operational analysis is useful to you, the The Pragmatic GMP newsletter collects practical work on qualification, utilities and GMP systems. One email, no promotional content.

Key takeaways

• Compliance of the feed with the applicable drinking water regulations (EPA NPDWR, EU, Japan, WHO) is part of the USP compendial definition of Purified Water, not a recommendation.
• 21 CFR 211.48(a) requires potable water under continuous positive pressure in a plumbing system free of defects: a system requirement, not an analytical one.
• Annex 1 clause 6.7 requires preventing microbiological contamination and minimising particulates, proliferation and endotoxins; clause 6.8 requires qualification taking account of seasonal variation; clause 6.10 requires WFI feed water to comply with specifications defined during qualification.
• Q&A EMA/INS/GMP/443117/2017 sets the expectation of data over approximately one year and notes that RO at ambient temperature is an ideal environment for biofilm.
• No regulatory source publishes inlet limits for hardness, chlorine, silica or TOC: these are specifications the site sets, justifies with its own data and verifies in qualification.
• Free chlorine and chloramines require different dechlorination strategies; the ammonia released by chloramines passes through the RO. By removing the disinfectant, pretreatment creates by functional necessity an environment favourable to biofilm: sanitisability and material compatibility are specification requirements.
• The chain plan → report → URS → acceptance criteria → monitoring limits is what makes the choice defensible during an inspection.

Regulatory and technical references

• USP, Purified Water and Water for Injection monographs; informational chapter <1231> (official since 1 December 2021).
• 21 CFR 211.48 Plumbing — eCFR Part 211.
• EudraLex Volume 4, Annex 1 (C(2022) 5938 final), in operation since 25 August 2023, clauses 6.7, 6.8, 6.10, 6.12, 6.13, 6.14 — EudraLex Volume 4.
• EMA Q&A EMA/INS/GMP/443117/2017 (since 1 August 2017) and EMA/CHMP/CVMP/QWP/496873/2018 (since 1 February 2021) — EMA; WHO TRS 1033 Annex 3 (2021) — WHO; WHO TRS 1025 Annex 3 (2020).
• PIC/S PI 009-4 Aide-Memoire "Inspection of Utilities", in force since 1 January 2021 — PIC/S.
• FDA "Guide to Inspections of High Purity Water Systems" (1993) and ITG No. 36 "Reverse Osmosis" (1980): reference material for investigators, not binding guidance — FDA.
• ISPE Baseline Guide Vol. 4 "Water and Steam Systems", 3rd ed. (2019); ISPE GPG "Membrane-Based WFI Systems" (2022); PDA TR No. 69 (2015): industry documents, not regulations. ICH Q9(R1), Step 4 on 18 January 2023 — ICH.