We processed the membrane module on the same day of retrieval. The case of the membrane module was opened using an electric saw. We inspected and photographed the macroscopic status of all membrane components to assess the extent and distribution of fouling. Then, membrane coupons of defined dimensions were dissected at about 10 cm from the inlet and 10 cm from the outlet side of the membrane. These sections did not contain glue lines. The surface area of coupons (stacked membrane and feed spacer) ranged from 3 to 7 cm².

To estimate the extent of organic fouling, the total organic carbon (TOC) of membrane coupons was measured. A membrane coupon was placed into a sterile tube containing 40 ml of filtered-sterilized (0.2 μm) milli-Q water. To detach the foulant from the membrane, the sample was sonicated at 4°C for 1 min at 50-amplitude using a sonicator probe (Q700 Qsonica sonicator, United States). The resulting extract was filtered (0.45 μm), and TOC was measured using a total organic carbon analyzer (TOC-VCPH, Shimadzu, Kyoto, Japan).

We used attenuated total reflection Fourier transformed infrared (ATR-FTIR) analysis to identify the functional groups of the organic matter deposited on the membrane surfaces. Membrane coupons were placed in Petri dishes and dried at 40°C for 24 h. Infrared spectra of virgin and fouled coupons were obtained with a Spectrum 100 Optica FT-IR spectrometer (PerkinElmer Inc., Waltham, Massachusetts, United States). Per measurement, 32 scans were recorded over a wavenumber range of 4,000–458 cm⁻¹ with a resolution of 4 cm⁻¹.

We measured the contents of adenosine triphosphate (ATP) as an indicator of the presence of active microbial cells on the membrane. The fouling material was extracted from the membrane by sonication as specified above. The ATP was analyzed by luminescence resulting from the enzymatic oxidation of D-luciferin by luciferase. Luciferase activity strictly depends on the presence of ATP. The ATP-driven luminescence of the extracted fouling samples was measured using a Celsis Advance II luminometer. The recorded relative light units (RLU) were converted to ATP concentrations using a calibration curve as described previously (Hammes et al., 2010).

The number of microbial cells was determined by flow cytometry from suspensions of detached fouling material from the coupons. Microbial cells in a 700 μl suspension were stained with 7 μl SYBR Green I (Invitrogen, 100 × diluted in 10 mM Tris buffer pH = 8.1), as recommended in (Nescerecka et al., 2016), and incubated 10 min in the dark at 35^°C. Then, stained cells were measured by flow cytometry (FCM) using a BD Accuri C6 instrument as described by (Prest et al., 2013). Briefly, the blue 488 nm laser (50 mW) was used to collect the green fluorescence at Fl1 = 533 30 nm, the red fluorescence Fl3 > 670 nm, as well as intrinsic cell parameters given by the side scattered and forward scattered light signals. The counting hardware of the instrument is calibrated to measure the number of particles in a 50 μl sample volume. Electronic gating was used to separate positive microbial stained signals from instrument, and water background noise. Sterilized filtered (0.2 μm) ultrapure water, as well as unstained samples were used as controls in all runs.

To identify, and estimate the contents of inorganic components in the fouling material, we used inductively coupled plasma mass spectroscopy (ICP-MS). The membrane coupons were immersed in 50 ml of 6% v/v HNO₃, and incubated 24 h at 40°C in a shaker. The samples were then filtered (0.45 μm) and diluted ten times. Elements were analyzed using an ICP-MS (Agilent 7500cx, Agilent Technologies, United States), and quantification was based upon calibration by a series of external standards. All elements were acquired in helium collision mode.

We conducted SEM to examine the ultrastructure of fouling formed on the membrane surface. SEM-EDS analyses were used to determine and map the elemental composition of the fouling materials. To preserve the fouling materials for SEM imaging and EDS analyses, small pieces of the fouled membrane were freeze dried (K775X Turbo Freeze Drier, Quorum Technologies, United Kingdom). To prepare cross-sections of freeze dried samples, small pieces of the membrane were immersed in liquid nitrogen and then were fractured. Fractured membranes were mounted on a 45° pre-tilted aluminum stub. To dissipate charging during SEM imaging and EDS analysis, samples were coated with 5 nm iridium using Q150T S Sputter coater (Quorum Technologies, United Kingdom). A Teneo VS SEM (Thermo Fisher Scientific, Netherlands) equipped with a secondary and backscattered detectors (ETD and T1 detectors), and an energy dispersive spectroscopy (EDS) detector was used for imaging and EDS analysis of inorganic elements. SEM images were acquired at an accelerating voltage of 1 kV and 9–10 mm working distance in optiplan mode. EDS analysis and mapping were done at an accelerating voltage of 14 kV and 10 mm working distance in analytical mode.

An inspection of the spiral wound RO membrane module revealed no macroscopic damage to the outer module components, and there was no telescoping of the membrane. After the destructive opening of the membrane module, the unwound membrane sheets were visually inspected: Visible membrane defects or glue-line damage were absent. An orange-colored fouling material 1) was homogeneously distributed over the membrane sheets, and 2) covered all individual membrane sheets of the membrane module (Figure 3).

Results of TOC, ATP, and cell counts accumulated on the feed spacer and membrane evidenced the presence of organic and biological fouling (Figure 3). The TOC contents on the inlet and outlet samples differed significantly (p = 0.032), with the inlet samples containing higher amounts of organic material. However, as a whole, the TOC ranged from 75 to 98 μg/cm², and inlet and outlet TOC values were in the same order of magnitude (Figure 4A).

The ATP contents, and the number of microbial cells in the inlet and outlet random samples did not differ significantly (p > 0.05). The ATP values varied between 6,500 and 21,000 pg/cm² (Figure 4B). The number of microbial cells varied between 2.0 and 3.5 × 10⁸ cells/cm, with the highest value of 7 × 10⁸ cells/cm detected in the inlet zone (Figure 4C).

Compared to the infrared spectra of a virgin membrane, the fouled membrane module showed distinct differences (Figure 5), indicating that organic fouling occurred. Infrared spectra of the fouled membrane showed the presence of characteristic carbohydrates peaks (1,000–1,100 cm⁻¹). Furthermore, amide I and amide II peaks attributed to proteins were also detected (1,600–1700 cm⁻¹). The higher absorption peak at 3,300 cm⁻¹ in the fouled membrane confirms the presence of carbohydrates as this band is due to the OH stretch. Moreover, a peak in the inorganic region at 458 cm⁻¹ was likely from the presence of iron (Figure 5).

The general structure of the membrane and the fouling cake layer was inspected using SEM. We examined coupon samples and diverse fields of view, and consistently, the fouling cake layer covered the membrane homogeneously (Figure 6). The thickness of the fouling cake layer ranged from about 3 to 20 μm. Apparently compression of the polysulfone layer occurred during operation. The polysulfone was barely visible at some points along the membrane (Figure 6A, arrows, see also Supplementary Figure S1). In contrast, an inspection of a virgin membrane showed a well-defined polysulfone layer (Figure 7A), and a pristine polyamide (Figure 7B). The polyamide was completely covered by the cake layer in the fouled membrane (Figure 6B).

Moreover, we noticed two polysulfone structures on the polysulfone layer of the fouled membrane (Supplementary Figure S2). One was laminar-like (Figure 6C) while the other was spongy-like (Figure 6D). Analysis of the virgin membrane showed that these two distinct polysulfone structures were arranged one on top of the other along the membrane (Figures 7C,D). However, in the fouled membrane, we observed zones with either laminar-like (Figure 6C) or the spongy-like (Figure 6D) polysulfone structures. This indicates that the compression of the polysulfone layer in the fouled membrane was not homogeneous. The points of compression seemed to correspond to undulations observed on the membrane (arrows in Figure 6A), and these undulations may correspond to the geometry of the 28 mil (∼710 μm thick) spacer.

To understand better the fouling of the membrane, we first examined the elemental composition of a virgin membrane using energy-dispersive X-ray spectroscopy analysis (EDS). The surface, made of polyamide, distinctively contained about 3–4% of nitrogen (Table 1). An estimate of the nitrogen in the virgin membrane is important since this element can be an indicator of organic (bio) fouling when examining a fouled membrane. Moreover, results of the elemental map suggest the presence of two distinct polysulfone layers. One with a sulfur content of about 2–3% (low sulfur layer), and one with a sulfur content of about 10–16% (high sulfur layer). The physical appearance of these polysulfphone layers was also distinct (Figure 7D). The high sulfur layer had a laminar-like appearance, while the low sulfur layer was spongy-like (see Inspection of the Membrane and the Fouling Cake Layer at High Magnification).

On the other hand, the cake layer from the inlet and outlet contained nitrogen contents at about 11–15% (Table 1, Supplementary Tables S3–S4), which contrast with the nitrogen content of about 4% of the virgin membrane surface (Table 1), and supports that organic (bio) fouling was present on the fouled membrane.

In addition, the EDS results showed that Fe was an important fouling component (about 6–7%, Table 1). A quantitative assessment of the inorganic contents of the fouling material by ICP-MS showed that the Fe content on the membrane and spacer was about 150 μg/cm² (Supplementary Table S5).

When conducting an autopsy of a desalination RO membrane, we need to assess whether the presence of sodium in the fouling material originates from remnants of seawater as heavy scaling by Na minerals would be rare. The assessed Na contents found in this membrane may originate from remnants of seawater since its ratio with respect to the presence of other elements closely resemble those found in seawater (Table 2).

Although comparisons are not straight forward, if one considers the reported TOC and ATP values—which are in all cases measured by the same approaches, combustion catalytic oxidation and luminescence, respectively—the extent of organic fouling and biofouling in this membrane was important when compared to previous autopsies data (Table 3).

Considering a value of 8.9 × 10⁻¹⁷ g ATP/cell (Hammes et al., 2010), we can calculate the number of active cells based on the measured maximum ATP value of 21,000 pg/cm² for the studied membrane. The number of active cells would be about 2.4 × 10⁸/cm², which three times lower than the measured maximum total cell counts of 7 × 10⁸/cm². This suggests that about most of the cells on the membrane were probably not active. Although the use of propidium iodide as a stain in combination with SYBR Green to determine the number of active cells has limitations, we propose using it in future autopsies to gain rough estimates on the proportion of total and intact microbial cell numbers. This information is relevant in cases where the effectiveness of specific pretreatments or membrane cleaning procedures need investigation.

Assuming that all TOC would be due to microbial cells, we can estimate the number of microbial cells based on the measured average TOC content of 80 μg/cm², and considering a carbon content of 39 fg C/cells (Vrede et al., 2002). The estimated cell number would be about 2 × 10⁹/cm², which would be about six to ten times higher than the cell counts we measured by flow cytometry. This suggests an important contribution of organic fouling.

Since the membrane itself has organic components, ATR-FTIR analysis of fouled membranes can be challenging, particularly when the fouling is mild. Here, fouling was important as reflected by the performance declined and the ATR-FTIR spectra of fouled and virgin membranes contrasted (Figure 5). The ATR-FTIR spectra support that the fouling of the membrane was both organic/biological and inorganic. In the case of mild fouling, an option for the analysis of the fouling material would be to dry the membrane sample, scrap the fouling material, and then analyze it. This would eliminate the interference of the membrane signals.

Taken together, our results and two previous studies (Rehman et al., 2019; Fortunato et al., 2020) indicate that biofouling can occur on RO membranes desalinating Red Sea water.

Another important contributor to the fouling of this membrane was Fe. The EDS analysis over various areas of the inlet and outlet sections of the membrane showed that the distribution of Fe was homogeneous (Supplementary Tables S3, S4). The estimated Fe content on the membrane and spacer was 150 μg/cm². This is a high value compared to that reported for a RO membrane treating groundwater, where the Fe was 1–40 μg/cm² using filtered and unfiltered groundwater, respectively (Hijnen et al., 2011). And, Fe contents of about 40 μg/cm² were reported for RO membranes of a desalination plant on the Rea Sea (Fortunato et al., 2020). Iron is a common foulant, and 40% of 99 RO membrane autopsies showed the presence of iron (Chesters et al., 2013). This is because ferric chloride (FeCl₃) is one of the most commonly used coagulants in the pretreatment of seawater to remove organic colloids (Al-Sheikh, 1997; Edzwald and Haarhoff, 2011). A recent laboratory study showed that using liquid ferrate, ten times less iron might be required to achieve similar pretreatment efficiencies (Alshahri et al., 2019). However, convincing practitioners to replace FeCl₃ with liquid ferrate may need a demonstration of its benefits at pilot scale.

The addition of FeCl₃ to waters with natural bicarbonate alkalinity results in the formation of iron hydroxide precipitates. When proper retention of these precipitates prior to the RO system fails, they cause serious inorganic fouling as observed in this membrane. Since the addition of FeCl₃ and electrolytes aims at removing suspended solids and colloids, a rational addition of these chemicals should be tightly linked to the online monitoring of turbidity. Residual Fe and electrolytes should be avoided because they could promote fouling—Fe is an essential metal for microbial growth and electrolytes can bind to the RO membrane inducing in situ coagulation. Another problem with the presence of iron in the RO membranes is that iron could catalyze the oxidation of the polyamide layer by aqueous chlorine (Gabelich et al., 2002).

Therefore, and based on the diagnosis of this autopsy, we recommend thoroughly 1) revise the extent of iron coagulant dosage, 2) revise the schedule for maintenance of the pretreatment units (MMF) and replacement of the cartridge filters, and 3) revise the (online) controlling process for coagulant addition. Due to the severity of Fe fouling (Table 3), cleaning at this fouling stage might be unpractical, but a cleaning test would be needed for confirmation.

High magnification imaging of cross-sectioned membranes can be a powerful forensic tool. This approach allows accurate measurements of the fouling layer. In addition, the elemental composition of the fouling layer can be readily contrasted to that of the inherent organic structure of the membrane (Supplementary Tables S3, S4). This provides added confidence in results particularly when the fouling material is organic. Moreover, cross-sectioned images expose the potential compression of membrane layers. When fouling increases, membranes are operated at increased pressures that result in severe membrane stress. In the studied membrane, we observed zones with high compression of the polysulfone layer (Figure 6A vs. Figure 7A), suggesting that the stress due to operating pressure was heterogeneous. Physical membrane compaction can reflect incomplete cleaning and irreversible fouling (Hoek et al., 2008). Under these conditions, high hydraulic pressures are needed to maintain a desired product water flux (Hoek et al., 2008). The physical compaction in the studied membrane is irreversible as observed by SEM imaging. We did not measure the extent of compaction, but it was reported that compaction could reduce the polysulfone surface porosity by up to ∼95% (Davenport et al., 2020). Furthermore, 20–30% of flux is lost due only to polysulfone compaction within 12–20 h after exposure even at low pressures (17 bar) (Hoek et al., 2008). Due to the detrimental effect of membrane compaction on process performance, current studies on spacer configuration aim at minimizing not only biofouling but membrane deformation (Lee et al., 2020).

New analytical methods to determine the extent of organic fouling and biofouling have advantages over conventional methods. For example, loss-on-ignition is not an accurate measure of the organic matter content in foulant samples with a high proportion of inorganics. This is because of significant reductions of FeCl₃ (50%), NH₄Cl, and (NH₄)₂SO₄ (both 100%) when the sample is heated at 550°C (da Costa and Schneider, 2019). Simultaneously determining ATP and cell counts seems appropriate for the diagnosis of biofouling (Vrouwenvelder et al., 2008). In particular, the use of flow cytometry to determine cell counts and the extent of biofouling has the advantage of being a high throughput and specific (DNA fluorescence staining) method. This contrast with cell counts via microscopy, which is time-consuming and can be prone to bias. However, biofouling observations via light and fluorescence microscopy provide information about the microbial cell arrangements (i.e., clusters) and coverage. Recently, there is increased interest in assessing the microbial communities of reverse osmosis membranes by high-throughput sequencing (Nagaraj et al., 2017; Rehman et al., 2019). For example, a metagenome study showed that interestingly Planctomycetes was the most abundant bacterial phylotype on an RO membrane from a desalination plant on the Red Sea (Rehman et al., 2019). However, how the microbial makeup on full-scale membranes relates to plant operation and how this information can be exploited to better manage (bio)fouling is far from elucidated. For the moment, advanced methods to assess organic fouling and biofouling are mostly used in laboratory investigations using flow cells. However, we anticipate their application as standard methods for autopsies of full-scale RO membrane modules. Here, a word of cautious is pertinent since, in some autopsies, advanced/sophisticated analysis have been used yielding results that are difficult to translate into practice. Unless an autopsy is part of fundamental research efforts, the selection of an array of methods should aim at providing a complete picture of the membrane status to provide rational corrective measures.

The parameters to consider when conducting an autopsy should ideally target all main fouling types. For example, quantification and spatial distribution of inorganic fouling can be assessed by ICP-MS and SEM-EDS, respectively. For the latter, the examination of cross-sectioned samples is recommended since these inform on the thickness of fouling. For organic fouling, measuring TOC seems appropriate, and can be combined with elemental SEM-EDS scans on cross-sectioned samples. From the latter, organic fouling signals are evident since the nitrogen content increases about three times compared to clean membranes (Table 1, see also Supplementary Table S2 vs. Supplementary Tables S3, S4), and the contents of carbon and oxygen become comparable (i.e., the carbon to oxygen ration approaches one). It should be considered that when fouling appears visually heterogeneous, SEM-EDS observations can be very demanding to acquire a representative overview of fouling. On the other hand, biofouling can be assessed by counting the microbial cells from the membranes using flow cytometry. Although not conducted in this study, this method also allows distinguishing the number of intact (i.e., live) and damaged (i.e., dead) microbial cells. In addition, measuring ATP indicates the presence of active microorganisms. Thus, flow cytometry and ATP estimations are important parameters to include combined when conducting autopsies.

The cake layer of this membrane was relatively compact and thick. However, we consider that it would be advisable to include a brief and gentle rinse of membrane coupons prior to autopsy with filtered sterile ultrapure water to remove any remnant process water, which could influence results particularly in the case of low fouled membranes. The brief and gentle rinse might be dipping the coupon a couple of seconds in the ultrapure water.

The need for autopsies would be limited when fouling was better understood allowing improved management of fouling issues in full-scale plants. Since fouling is still an important issue in full-scale installations, membrane autopsies are still required. Currently, most autopsies published refer to laboratory-scale studies with biodegradable nutrient dosage to accelerate fouling. One main limitation of laboratory-scale studies is the difficulty translating the gained information to real systems. For example, chemical cleaning efficiencies in full-scale reverse osmosis membranes are lower than when investigated under laboratory conditions (Jafari et al., 2020). This highlights the need for full-scale research and reporting of data. Unfortunately, autopsies from full-scale membranes are limited. Certainly, practitioners conduct membrane autopsies. However, the information is not in the public literature. Ideally, besides being published, results of full-scale membrane autopsies along with information of key process parameters, water quality and dosed chemicals, could be archived in a public database. One can envision such a database to serve as a diagnostic and corrective tool in a similar manner as medical databases are used by medical practitioners to diagnose and medicate patients. This idea highlights the need for a consensus procedure for conducting autopsies and reporting data. Finally, we propose that a scale to determine the degree of fouling could be constructed based on the above-mentioned database. In such a scale, degrees of fouling from mild to severe could be associated to defined indicators.