Environmental performance of analytical methods: insights from life cycle assessment on four analytical chemistry workflows

Abstract

Analytical chemistry plays a central role in research and innovation, yet its environmental impacts remain insufficiently represented in sustainability assessments. Current score-based “greenness” indicators—defined as the compliance to the 12 green analytical chemistry principles—often overlook upstream and downstream processes, such as instrument manufacturing and data management. This study applies life cycle assessment (LCA) to characterize the environmental performance of four representative workflows in a cosmetic industry R&I setting: an untargeted UPLC-PDA/HRMS analysis of natural extracts and three phenoxyethanol quantification methods (GC–MS, HPLC–UV, UPLC-UV). Inventories were structured to distinguish sample preparation, instrumental analysis, and data processing from supporting flows like consumables and energy. The results reveal distinct environmental hotspots for each protocol. The UPLC-PDA/HRMS workflow exhibited a climate change impact of 4.30 kg CO₂eq, driven by the amortization of the devices and the sample transport. In contrast, the GC–MS workflow (3.82 kg CO₂eq) was heavily influenced by dichloromethane production, while the HPLC–UV protocol reached 6.84 kg CO₂eq due to consumable-related flows and longer run times. These findings demonstrate that environmental burdens are highly workflow-specific and that commonly cited factors, such as solvent use, are not always the primary drivers. The study concludes that the normalization factors from the product environmental footprint (PEF) method are inadequate for analytical laboratory contexts and suggests that quantitative LCA provides a necessary complement to score-based metrics like the analytical greenness metric (AGREE) for supporting robust eco-design and corporate sustainability initiatives.

Graphical Abstract

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1007/s00216-026-06594-z.

Introduction

Analytical chemistry is deeply embedded in cosmetic Research and Innovation (R&I), supporting raw materials development, formulation, and quality control throughout the product lifecycle to ensure performance and safety for consumers. Yet, despite this central role, analytical activities are rarely included in environmental assessments performed in R&I. Where environmental evaluation has been applied in laboratory contexts, it has largely followed the historical focus of green chemistry on visible material flows: single-use consumables, whose accumulation in waste streams contributes to plastic pollution [1], and solvents, whose volume and toxicity have long been central to green chemistry metrics [2]. While industrial-scale chemical processes have received significant attention driven by economic stakes and production volumes [3], and recent pioneering studies have begun to successfully apply LCA to specific analytical steps like sample preparation [4], the environmental footprint of analytical R&I operations remains systematically underestimated.

In the field of analytical chemistry, the evaluation of sustainability has largely been driven by the concept of “greenness.” Defined by the 12 Principles of Green Analytical Chemistry [5], this term is widely used in literature to describe a method’s adherence to eco-friendly practices. Several metrics, such as the Analytical GREEnness metric (AGREE) [6], have been developed to quantify this “greenness.” While numerous metric tools have emerged during the past two decades to guide the transition toward sustainable practices, recent reviews [7, 8] highlight that they fail to capture the full scope of analytical processes. They typically focus on solvent and reagent consumption, overlooking broader life cycle contributions such as instrument manufacturing, carbon intensity, and geographical origin of energy used, consumables, and data management. As a result, they provide only a partial representation of the actual system and lack the comparability required for evaluating absolute environmental footprints across the entire life cycle. In this study, we explicitly distinguish these compliance-based scores from the quantitative environmental impact assessment provided by life cycle assessment (LCA).

A common argument for the limited scope of current assessments is that analytical chemistry handles only small quantities of materials and therefore has a negligible environmental impact compared to industrial production. Yet, this assumption is misleading. Precisely because analytical workflows operate at low material throughput, they rely on high-tech, resource-intensive equipment while having limited leverage on global supply chains. This makes laboratories increasingly vulnerable to disruptions in the supply of energy, consumables, and critical materials. The recent shortages of microprocessors [9], which significantly affected research laboratories despite their marginal share of global demand, illustrate this exposure. Consequently, better quantifying these environmental impacts is not merely a sustainability exercise; it is a prerequisite for reducing dependencies and strengthening the resilience of laboratory operations.

Life Cycle Assessment (LCA), standardized under ISO 14040/44 [10], offers a comprehensive and quantitative framework to evaluate environmental impacts across the full analytical workflow. However, applying LCA to laboratory activities presents specific challenges, particularly regarding interpretation. The use of global normalization factors, such as those from the Product Environmental Footprint (PEF) [11], remains poorly adapted to the analytical context. Designed for product-level assessment, these factors distort the relative weighing of impact categories and hinder context-relevant interpretation for R&I activities.

In this study, two LCA case studies are presented to illustrate both the potential and the methodological challenges of applying this framework to analytical research. The first assesses a UPLC-PDA/HRMS workflow for the qualitative analysis of natural extracts, representative of a common untargeted need in R&I laboratories [12]. The second compares three targeted quantification methods for phenoxyethanol—one of the main preservative ingredients used [13] and whose level in cosmetic products is regulated (Annex V of the European Cosmetic Directive [14])—using GC–MS, HPLC–UV, and UPLC-UV. Although both studies follow the same standardized approach, they yield markedly different environmental profiles. These results emphasize that analytical systems cannot be characterized by universal “hotspots”; instead, the relevance of each impact category depends strongly on the analytical context, workflow configuration, and resource intensity. These findings underscore the need for developing an analytical-specific normalization approach to enable context-relevant prioritization of environmental burdens and support eco-design strategies that enhance both sustainability and resource autonomy in R&I laboratories.

Materials and methods

Goal and scope definition

The aim of this study is to assess and compare the environmental impacts of analytical workflows representative of R&I activities in the cosmetics industry, using the Life Cycle Assessment (LCA) framework standardized under ISO 14040 and ISO 14044. Two case studies were performed to illustrate both the potential and the methodological limitations of applying LCA to analytical chemistry within the geographical context of France.

The functional units (FU) were defined to reflect specific analytical objectives:

• Functional Unit 1 (qualitative screening): To perform one qualitative analysis of a natural extract obtained from a 100 mg sample using a UPLC-PDA/HRMS workflow. This definition reflects a standard untargeted analysis including sample preparation, chromatographic separation, detection, and comprehensive data processing.

• Functional Unit 2 (targeted quantification): To quantify phenoxyethanol in a series of 10 cosmetic formulation samples under standard laboratory operating conditions. This unit ensures comparability across three methods (GC–MS, HPLC–UV, and UPLC-UV) designed to achieve equivalent analytical performance (i.e., comparable limits of quantification, accuracy, and precision for the intended application). The FU was defined at the analytical sequence level to capture the reality of laboratory organization, where sequence design (calibration, blanks, QCs) is driven by method requirements rather than single-sample analysis.

Chemicals and reagents

LC–MS grade solvents (water, ethanol, acetonitrile, methanol) and HPLC grade dichloromethane stabilized with amylene were purchased from Carlo Erba (Val de Reuil, France) or VWR International (Fontenay-sous-Bois, France). Analytical standards, including 2-phenoxyethanol (min. purity 99%) and ethylparaben, were obtained from Fluka/Sigma-Aldrich (Saint-Quentin-Fallavier, France). Reagents for mobile phases and buffers, including formic acid, acetic acid (Normapur ACS), sodium acetate (ACS Reagent), and phosphoric acid, were supplied by VWR International. The calibration solution for HRMS (Pierce™ FlexMix™) was obtained from Thermo Fisher Scientific.

Description of analytical workflows

Case Study 1: untargeted analysis of natural extracts (UPLC-PDA/HRMS)

• Sample preparation: The natural extract (100 mg) is solubilized at 10 mg/L in a water/ethanol mixture (50/50, v/v) and filtered through Acrodisc® GHP 0.2-µm syringe filters.

• Instrumental analysis: Analyses are performed on an Ultimate 3000 UPLC system coupled to a photodiode array (PDA) detector and an Orbitrap Exploris 240 HRMS (Thermo Fisher Scientific). Separation is achieved on an ACQUITY BEH Shield RP18 column (Waters, 100 × 2.1 mm, 1.8 µm) at 35 °C with a 25-min gradient of water/acetonitrile (both with 0.1% formic acid) at 0.6 mL/min. The HRMS operates in full scan and data-dependent MS² mode (ddMS²) with polarity switching (±).

• Data processing: An analytical sequence typically includes 20 injections. Data are processed using Xcalibur (Freestyle module v4.4) and Compound Discoverer (v3.3) on an HP Z4 G4 workstation, requiring approximately 2 days of processing per sequence.

Case Study 2: quantification of phenoxyethanol (comparative study)

Three quantification methods were compared. Their specific operating conditions are summarized in Table 1.

Table 1.

Operating conditions for the three phenoxyethanol quantification workflows

Parameter	GC–MS	HPLC–UV	UPLC-UV
System	Agilent GC 7890B/MSD 5977A	Shimadzu Nexera-i 2040	Waters ACQUITY iClass
Column	CP WAX 57 CB(25 m × 0.25 mm × 0.20 µm)	Merck Purospher Star RP-18-e (3 µm)	Restek Force Biphenyl (1.8 µm)
Mobile phase/gas	Helium	ACN/acetate buffer pH 4.5 (25/75)	ACN/0.1% H₃PO₄ (gradient ~ 48/52)
Flow rate	1 mL/min	1 mL/min	0.6 mL/min
Injection volume	1 µL	10 µL	2 µL
Run time	31 min	15 min	10 min
Injections/sequence*	47	46	48
Data processing	Masshunter v10.0 (40 min)	Empower Pro v3.6 (2 h)	Empower Pro v3.6 (2 h)

*Total injections required to analyze 10 samples, including calibration, blanks, and QC

• GC–MS: Samples (150–220 mg) and standards are dispersed in a dichloromethane/methanol mixture (50/50, v/v) containing ethylparaben as an internal standard.

• HPLC–UV: Samples (100–1000 mg) are dispersed in an acetonitrile/acetate buffer (pH 4.5) mixture (25/75, v/v). Preparation includes vortexing and sonication.

• UPLC-UV: Samples (100–1000 mg) are dispersed in a methanol/water mixture (50/50, v/v).

All sample solutions are filtered (0.45 µm PTFE, Pall) and transferred into 2-mL glass vials prior to injection. In the baseline scenario, all single-use consumables including 2-mL borosilicate vials, glass inserts, PTFE/silicone septa, and 0.45-µm PTFE syringe filters are modeled as single-use to be consistent with the standard operating procedures of the cosmetic R&I laboratories where the primary data were collected.

System boundaries and Life Cycle Inventory (LCI)

A “cradle-to-grave” perspective was applied. The system boundaries encompass the production, transport, and end-of-life of chemicals and consumables, instrument manufacturing and amortization, electricity use during operation, and data processing.

Transport assumptions were defined separately for samples and for reagents and consumables. Sample transport was modeled differently across the two case studies, depending on actual logistics.

For the UPLC-PDA/HRMS workflow (Case Study 1 ), sample transport was included in the inventory, as samples are physically transferred between sites by road.

For the phenoxyethanol quantification workflows (Case Study 2 ), sample transport was excluded from the system boundaries: analyses are conducted on a single industrial site, and sample transfer is limited to short-distance handling between operators of different services, which is considered negligible.

Transport of reagents and consumables was modeled implicitly through the “market for” activities of the Ecoinvent v3.9 database, which embeds average European transport distances and modal splits (lorry, train).

Primary data were collected directly from L’Oréal analytical laboratories (Aulnay-sous-Bois, Chevilly-Larue, and Saint Ouen, France). Each measured flow was then associated with the corresponding background process from the Ecoinvent v3.9 database to model upstream production and end-of-life stages.

Life Cycle Impact Assessment (LCIA)

Environmental impacts were calculated using the Product Environmental Footprint (PEF, EF 3.1) method. This method covers 16 midpoint impact categories (Table 2) and provides normalization factors on the European scale.

Table 2.

List of EF 3.1 impacts and their units

Impact category	Unit
Climate change	kg CO₂ eq
Ozone depletion	kg CFC-11 eq
Ionizing radiation	kBq U-235 eq
Photochemical ozone formation	kg NMVOC eq
Particulate matter	disease incidence
Human toxicity, non-cancer	CTUh
Human toxicity, cancer	CTUh
Acidification	mol H⁺ eq
Eutrophication, freshwater	kg P eq
Eutrophication, marine	kg N eq
Eutrophication, terrestrial	mol N eq
Ecotoxicity, freshwater	CTUe
Land use	m2a
Water use	m3 world eq
Resource use, fossils	MJ
Resource use, minerals, and metals	kg Sb eq

This specific methodological framework was selected to ensure consistency with L’Oréal’s Sustainable Product Optimization Tool (SPOT) [15]. SPOT is the company’s internal life-cycle-based assessment platform, used to quantify and optimize the environmental performance of commercial products. Aligning this study with the PEF framework ensures methodological continuity between corporate-level product evaluations and the research-scale assessments performed in the laboratory.

The LCA models were built and analyzed using OpenLCA software (GreenDelta, Germany).

Sensitivity analyses were conducted to evaluate the influence of electricity consumption, equipment lifespan, and data processing duration.

All datasets are available from the corresponding author upon reasonable request.

Results

Contribution analysis

The contribution analysis reveals distinct environmental hotspots across the analytical workflows, associated primarily with differences in solvents, instrument types, and data-processing requirements. All contribution values were rescaled so that, for each impact category, the sum of all processes equals 100%. This allows direct comparison of the relative contributions within each category.

Contribution analysis of UPLC-PDA/HRMS natural extracts untargeted analysis

The contribution analysis of the UPLC-PDA/HRMS workflow shows a consistent pattern across all PEF indicators. Consumables constitute the main contributors across most categories, followed by devices, while electricity and chemical products generate secondary contributions.

Potential climate change impacts amount to 4.3 kg CO₂ eq for the whole sequence, and the main contributors are different than in other impact categories, in decreasing order: sample transport (33%), devices (20% total), chemical products (15%), consumables (8%), and electricity (5%). This distribution over all impacts reflects three main drivers: (i) transport processes linked to sample logistics, (ii) amortized equipment manufacturing, and (iii) the upstream burdens associated with the manufacturing of single-use consumables such as polypropylene filters and borosilicate vials. Unlike targeted methods, the extensive data processing required for high-resolution mass spectrometry contributes to electricity use; however, under the French electricity mix, this contribution remains secondary in terms of climate change impacts. This pattern remains consistent with workflows operated in the French electricity context and would differ in regions with more carbon-intensive electricity mixes. For climate change, the contribution profile differs from other indicators. Devices and transport dominate the relative contributions, while consumables play a more limited role.

For toxicity-related indicators (freshwater ecotoxicity, human toxicity, cancer, and non-cancer), consumables and devices dominate the contributions. The manufacturing of borosilicate glass and plastic consumables involves energy and resource-intensive upstream processes, which explains their systematic contribution across these indicators. Devices constitute the second main contributor for toxicity and resource-related categories, reflecting the material composition of analytical equipment. Metallic alloys, printed circuit boards, electronics, and semiconductor components involve extraction and refining processes with high characterization factors, which explain the contribution of device amortization.

Contribution analysis of three protocols for the phenoxyethanol quantification analysis

Potential climate change impacts amount to 3.82 kg CO₂eq for the GC–MS protocol, 6.84 kg CO₂eq for the HPLC–UV protocol, and 3.77 kg CO₂eq for the UPLC-UV protocol. In the GC–MS workflow, impacts are primarily driven by dichloromethane production, which represents the main contribution (64%), followed by consumables (21%) (including polypropylene vials, borosilicate glass, and PTFE septa), electricity use (8%), and devices (7%). The high contribution of dichloromethane reflects both its direct production impacts and the energy-intensive manufacturing processes for this chlorinated solvent. In the HPLC–UV workflow, the dominant contributions originate from consumables (notably polypropylene and borosilicate), ahead of solvent production and electricity use linked to the longer analytical sequence. In the UPLC-UV workflow, contributions are distributed between consumables, solvent production, and electricity, resulting in an intermediate profile consistent with the shorter sequence duration and the lower solvent volumes compared with HPLC–UV.

For toxicity-related indicators, freshwater ecotoxicity remains highest in the GC–MS protocol (75 CTUe), primarily due to dichloromethane production (46%), with additional contributions from electricity use (19%) and device use (15%). Human toxicity indicators show distinct profiles across workflows, with devices representing a major contributor in the UPLC-UV workflow (particularly for cancer-related impacts at 58%), while HPLC–UV and GC–MS exhibit more distributed contribution patterns across consumables, chemicals, and devices.

Devices’ contributions depend both on instrument material composition and amortization parameters, including equipment mass and the number and duration of analytical sequences over the instrument lifetime. As a result, a lighter system operated over fewer analytical sequences can yield higher per-analysis contributions than a heavier system whose manufacturing impacts are amortized over a larger number of analyses. Because the two case studies rely on different functional units, their results cannot be directly compared in absolute terms, and they do not imply in any way that workflows relying on distinct analytical purposes or detection systems (GC–MS and UPLC-UV in our case) would imply comparable environmental performance. Nevertheless, it shows that assumptions commonly made in analytical chemistry (such as considering solvents or sample-preparation steps as the systematic main contributors) are not supported by quantitative analysis. Across the four workflows, the variability in contribution structures confirms that environmental hotspots are method-specific and cannot be inferred without quantitative life-cycle modeling.

Normalization results

Normalization was performed using the Product Environmental Footprint (PEF) factors published by the European Commission (version 3.1). These factors express each midpoint result relative to the annual environmental footprint of an average European citizen. In accordance with PEF guidance, normalized values cannot be compared across categories; they indicate which environmental pressures are most influential within a given workflow when expressed in “European-equivalent” units. In other words, normalization changes the reference scale, not the underlying pattern of contributions.

Normalization using EF 3.1 highlights a consistent set of dominant impact categories across the four analytical workflows investigated. Regardless of the analytical technique, ozone depletion, resource use, fossils, and ecotoxicity, freshwater systematically emerges as the most influential category after normalization. Differences between workflows are primarily expressed as variations in magnitude rather than as changes in the hierarchy of impact categories.

This apparent convergence of normalized profiles indicates that, within the scope of the studied configurations, analytical workflows rely on similar background systems.

Discussions

Analytical environmental assessment: limits of greenness indicators

Numerous metrics have been proposed to assess the environmental performance of analytical methods, including National Environmental Methods Index (NEMI), the Analytical Eco-Scale, Green Analytical Procedure Index (GAPI), Analytical Greenness Metric (AGREE) [6], and more recent developments such as AGREE for sample preparation (AGREEprep), ComplexGAPI for software assessment, Blue Applicability Grade Index (BAGI) [16], and Red–Green–Blue (RGB) fast [17]. As reviewed by Durand et al. [7], new greenness indicators continue to emerge almost every year, reflecting the growing attention for environmental aspects in analytical method development.

If these tools are often characterized by straightforward application, simple implementation, and visual representation, their methodological limitations however restrict their applicability beyond qualitative screening. Although some criteria rely on measured parameters (e.g., solvent use, waste generation), greenness tools cover only partial stages of analytical workflows and exclude upstream and downstream processes, such as instrument manufacturing or reagent production. The absence of standardized weighing procedures also limits reproducibility and comparability across studies. Moreover, electricity use is typically treated as a generic input, without geographical differentiation. This aspect is particularly important for multinational organizations such as L’Oréal, where laboratories operate under distinct regional electricity mixes.

Another limitation concerns the definition of scope. Analytical workflows differ in operating conditions and sequence structures, and current metrics do not specify whether they assess a single run, a complete sequence, or a full method implementation. Without explicit functional definitions, methods as distinct as GC–MS (targeted quantification) and UPLC-PDA/HRMS (untargeted profiling) could appear similar under a given greenness score, even though their life-cycle impacts differ by more than one order of magnitude.

To this end, the AGREE indicator, which transforms the 12 principles of Green Analytical Chemistry (GAC) into a unified 0–1 scale, was computed for each workflow. The detailed scoring and criteria for the AGREE indicator for each workflow are provided in Tables S1 and S2 in the Electronic Supplementary Material (ESM). Previous publications have evaluated other analytical workflows using AGREE, deeming their methods “green” [18–21]. In contrast, few studies have challenged both LCA and AGREE for a given method in the past. In this study, AGREE metrics were assessed for natural extracts untargeted analysis (UPLC-PDA/HRMS) and phenoxyethanol quantification workflows in comparison to LCA. All methods registered low levels of greenness, with AGREE scores inferior to 0.41. The UPLC-PDA/HRMS workflow, despite registering a climate impact of 4.3 kg CO2eq by LCA, unexpectedly achieved the greatest score of 0.41, implying superior relative greenness among all methods. In contrast, phenoxyethanol quantification yielded lower scores (0.3 for GC–MS, 0.25 for HPLC–UV, and 0.35 for UPLC-UV).

To better align the LCA and AGREE results, we must examine the specific categories driving these outcomes. In AGREE, the low scores for the phenoxyethanol methods (GC–MS, HPLC, UPLC) are primarily driven by heavy penalties in Principle 7 (analytical waste generation) and Principle 11 (use of toxic reagents). These penalties correlate with the chemical and solvent categories in our LCA, but they fail to capture the nuances revealed by LCA. These values align with the LCA-derived potential carbon footprints (3.82, 6.84, and 3.77 kg CO₂eq, respectively), and notably, for the HPLC–UV method, the lowest AGREE score coincided with the highest climate impact. AGREE metrics were compared to the climate impact of each workflow, because toxicity-related indicators in LCA mainly reflect contributions from upstream manufacturing. AGREE focuses on direct toxic operator exposure rather than indirect contributions, such as solvent production or device manufacturing. It is fundamental to note that AGREE and LCA operate on different system boundaries by design, as AGREE mainly evaluates the use-phase and laboratory-level hazards, whereas LCA provides a cradle-to-grave assessment. A direct correlation should therefore not be expected. The comparison illustrates how use-phase metrics can overlook major upstream burdens such as IT infrastructure or device manufacturing, which is precisely the contribution that LCA brings to method evaluation. A detailed mapping of AGREE penalties against LCA hotspots for each of the four workflows is provided in Table 3, making the divergence in scope between the two methodologies fully transparent. The table illustrates the categories of impact that are systematically invisible to score-based greenness metrics by construction, and that LCA reveals. Climate change values are reported alongside the AGREE score to illustrate the absence of correlation between the two indicators. AGREE penalties refer to the criteria of the AGREE software (Pena-Pereira et al., 2020) where each workflow received a sub-score below the maximum; only the top three contributing penalties are listed for clarity, and the full per-criterion scoring is provided in Table S1 (UPLC-PDA/HRMS) and Table S2 (GC–MS, HPLC–UV, UPLC-UV). Climate change values are expressed in kg CO₂eq per functional unit (one qualitative analysis for UPLC-PDA/HRMS; quantification of phenoxyethanol in 10 cosmetic samples for GC–MS, HPLC–UV, and UPLC-UV) and were calculated using the EF 3.1 method. LCA climate change hotspots are reported qualitatively to maintain consistency with the contribution analysis presented in Figs. 1, 2, 3, and 4. Toxicity hotspots refer to the relative contributions to freshwater ecotoxicity and human toxicity (cancer/non-cancer) categories of the EF 3.1 method.

Table 3.

Mapping of the AGREE compliance penalties (use-phase) against the LCA hotspots (cradle-to-grave) for each of four analytical workflows

Analytical workflow	AGREE score	LCA impact: Climate change (kg CO₂eq)	Top AGREE penalties (use-phase)	LCA climate change hotspots (cradle-to-grave)		LCA toxicity hotspots (freshwater and human)	Categories invisible to AGREE by construction
UPLC-PDA/HRMS (untargeted)	0.41	4.30	- Principle 5: semi-automatic, not miniaturized - Principle 8: low throughput (0.12 samples/h) - Principle 12: reagents toxic to aquatic life and highly flammable	- Sample transport - Devices (amortization) - Chemical products - Consumables - Electricity	- Consumables manufacturing (PP filters, borosilicate vials) - Devices manufacturing (electronics, metals) - Workstation/displays for data processing (human toxicity, cancer)		- Inter-site sample transport - Upstream device manufacturing (electronics, semiconductors) - Electricity for two-day data-processing workload
GC–MS (targeted)	0.30	3.82	- Principle 7: high analytical waste volume (782 mL) - Principle 11: high volume of toxic reagents (782 mL) - Principle 12: highly flammable and toxic reagents	- Dichloromethane production - Consumables (PP vials, borosilicate, PTFE septa) - Electricity - Devices	- Freshwater ecotoxicity: dichloromethane 46%, electricity 19%, devices 15% - Distributed pattern across chemicals, consumables and devices for human toxicity		- Cradle-to-gate footprint of chlorinated-solvent synthesis - Manufacturing footprint of single-use solid consumables
HPLC–UV (targeted)	0.25	6.84	- Principle 1: large number of pre-treatment steps - Principle 7: highest waste volume (1.39 L) - Principle 11: highest toxic-reagent volume (1.39 L)	- Consumables (PP filters, borosilicate vials): leading contributor - Solvent production - Electricity (longer analytical sequence)	- Distributed pattern across consumables, chemicals and devices		- Manufacturing footprint of single-use solid consumables (suggested but not operationalised by AGREE) - Electricity for 2 h data processing
UPLC-UV (targeted)	0.35	3.77	- Principle 7: high waste volume (763 mL) - Principle 11: high toxic-reagent volume (763 mL) - Principle 12: highly flammable, corrosive reagents	- Distributed between consumables, solvent production and electricity (intermediate profile)	- Devices manufacturing: 58% of human toxicity (cancer) - Consumables manufacturing as secondary contributor		- Device-amortization footprint - Manufacturing footprint of single-use solid consumables

Fig. 1.

Contribution analysis for UPLC-PDA/HRMS workflow

Fig. 2.

Contribution analysis for GC–MS workflow

Fig. 3.

Contribution analysis for HPLC–UV workflow

Fig. 4.

Contribution analysis for UPLC-UV workflow

Moreover, limitations were encountered during the AGREE computations. For analytical waste, the amount can be assessed only in liters or grams. While the AGREE methodology suggests including solid single-use devices in the waste metric, doing so accurately in routine practice is highly challenging. Consequently, solid consumables are often neglected by users in published applications, even though our LCA shows they drive the climate impact of HPLC–UV and UPLC-UV. Finally, AGREE omitted data processing for UPLC-PDA/HRMS, although it was identified by LCA as a significant contributor to human toxicity (cancer), notably due to the use of workstations and displays. It is well established that LCA accounts for data processing whereas AGREE does not [7]. Furthermore, while score-based metrics like AGREE can be completed in a few minutes, a full LCA requires significant expertise and time (typically days to weeks per workflow, depending on data availability and inventory complexity), which limits its use for rapid routine screening but justifies its application for strategic eco-design decisions. Hence, while the AGREE metric remains useful for a first evaluation or comparison between methods, only a holistic approach like LCA delivers a comprehensive picture.

Method-specific variability and limits to generalization

The results of this study show that commonly held assumptions about environmental impacts in analytical chemistry do not hold when examined across multiple workflows. In literature and in routine discussions, environmental impacts are often described in broad, undifferentiated terms, which can mask the diversity of underlying mechanisms. Such conceptual ambiguity tends to encourage generic statements that are not grounded in impact-category modeling or in quantified inventories.

Applying these assumptions to the workflows analyzed here would have produced incomplete interpretations. In the phenoxyethanol case study, three workflows designed to achieve the same analytical objective display markedly different contribution profiles because of differences in instrument configuration, sequence length, consumable requirements, and the relative weighing of preparation, analysis, and data-processing steps. Depending on the workflow, the largest contributions may arise from solvent production, electricity consumption, or equipment manufacturing, and these patterns cannot be anticipated without quantitative assessment. This demonstrates that environmental hotspots are strongly workflow-specific.

These findings also highlight that LCA does not constitute the final objective, but rather a decision-support tool. In the natural-extracts workflow, the hotspot analysis identified the data-processing stage as one of the main contributors to the overall impacts, indirectly confirming an operational bottleneck often encountered in the era of untargeted data exploitation. The observed result thus necessitated a re-evaluation of the chemoinformatics processing strategy, leading to the integration of automated retention index checking with existing constructor workflows. Through this revised strategy, a twofold reduction in the data treatment phase is projected, which will concurrently alleviate operator workload, mitigate the risk of errors, and decrease the environmental footprint. In the comparative assessment of the phenoxyethanol methods, the LCA led to a re-examination of how analytical sequences are constructed, including the number of injections, the balance between blanks, standards and samples, and analytical quality assurance required for the intended purpose. These outcomes illustrate the capacity of LCA to foster methodological optimization rather than simply quantify impacts.

Conducting a full workflow-specific LCA for every analytical method would provide the most robust evidence base, but such an approach is difficult to scale. Laboratory environments involve a large number of methods, frequent adaptations, and small quantities, which increases sensitivity to inventory errors. Developing simplified, operationally compatible LCA approaches (while ensuring consistency with core life-cycle principles) will therefore be essential to support wider adoption. Strengthening operators’ understanding of inventory structure, boundary definition, and foreground–background separation will also be required to ensure reproducibility and comparability.

Finally, workflow-level LCAs offer a basis for integrating analytical activities into corporate environmental accounting. While single workflows may contribute modest amounts, their repetition across platforms, projects, and R&D workflows can justify their inclusion in organizational inventories, particularly when seeking to identify levers for operational optimization rather than major emission sources.

Analytical chemistry-based normalization

To be consistent with the Life Cycle Inventory Assessment (LCIA) method selected in this study, we applied the per-capita PEF normalization factors, which are derived from the total annual environmental footprint of the European Union divided by its population. However, its representativeness for analytical chemistry is limited. Laboratory workflows operate at scales several orders of magnitude below those underlying European per-capita flows, in terms of both material throughput and electricity use. The construction of PEF normalization factors also constrains interpretation. They rely on aggregated regional inventories with documented gaps for several impact pathways, including toxicity-related flows.

In the present study, although the analysis is restricted to four analytical workflows, normalization consistently highlights similar impact hierarchies across all cases (Fig. 5). It should be clarified whether this apparent convergence reflects common structural features of analytical workflows or rather a mismatch between the selected normalization scale and laboratory-level activities.

Fig. 5.

Normalization for GC–MS, HPLC–UV, and UPLC-UV for phenoxyethanol quantification workflows and for the whole UPLC-HRMS sequence

A more meaningful reference for this domain could be a laboratory-specific normalization factor, for example, the annual environmental footprint of a typical European analytical laboratory. Such a reference would preserve the comparative value of normalization while providing a scale aligned with laboratory operations, offering results that are immediately interpretable by operators. Developing sector-specific or industrial normalization benchmarks would therefore constitute a significant improvement for the environmental assessment of analytical workflows.

However, the per-capita PEF normalization provides a shared reference. A more consistent interpretation may be achieved by integrating analytical workflows at the product level they support. In this case, the environmental impacts of analytical activities would contribute as a support function to the product system, which would then constitute the appropriate object for PEF-based interpretation (as products are directly representative of consumption patterns and European lifestyles, unlike laboratory activities).

Reducing the environmental footprint of analytical workflows: levers, sensitivity, and methodological boundaries

Sensitivity to the electricity mix

The baseline assessment was conducted under the actual operating context of the laboratories where the data were collected, that is, the French electricity mix. Every national mix is atypical in its own way: the French mix is nuclear-dominated, the German coal-heavy, the Norwegian hydro-dominated, the Polish among the most carbon-intensive in the EU. There is no single context that can claim to be representative. Reporting impacts under the actual operating mix therefore makes explicit a property that LCA results carry by design: environmental performance is location-dependent, and conclusions cannot be automatically transposed across geographies without recalculation.

To illustrate the magnitude of this dependency, a sensitivity analysis was performed by substituting the French electricity dataset with a European-scale dataset. We selected the market group for electricity, low voltage (RER) from Ecoinvent, which provides a consumption-weighted average of the national low-voltage markets composing the European region (348 vs 131 g CO₂eq/kWh for France). This dataset is the appropriate construct in Ecoinvent to represent a generic European operating context. Climate change values are presented in Figure S1. Under the European average mix, climate change increases by + 9.0% for UPLC-PDA/HRMS (4.30 to 4.69 kg CO₂eq), + 13.9% for GC–MS (3.82 to 4.35), + 3.6% for HPLC–UV (6.85 to 7.09), and + 15.2% for UPLC-UV (3.78 to 4.35). The relative ranking of the four workflows is preserved: UPLC-UV remains the lowest-impact phenoxyethanol quantification method and HPLC–UV the highest. The comparative interpretation of the manuscript is therefore robust to the choice of electricity context.

Across the other EF 3.1 categories, the substitution rebalances rather than uniformly worsens the impact profile. Categories driven by fossil combustion (acidification, eutrophication, photochemical ozone formation) increase, while ionizing radiation decreases by 43 to 57%, reflecting the high nuclear share of the French mix. Fossil resource use also decreases by 6 to 17% under the European mix. This category quantifies the depletion of fossil reserves consumed along the entire upstream supply chain (fuel extraction, transformation, infrastructure…) and is distinct from direct CO₂ emissions tracked by the climate change category. The French mix hence relies on substantial upstream fossil inputs, notably for the uranium fuel cycle and back-up generation. This is a key methodological point: while the French context is genuinely low-carbon at the climate change indicator, this advantage does not extend to fossil resource use, which is higher in France than under the European average. Such a conclusion can only be drawn by considering the full life cycle and the full set of impact categories, not a single indicator. It reinforces the central message of the study: environmental hotspots are workflow- and context-specific and cannot be inferred from generic assumptions or from any single metric taken in isolation.

Eco-design recommendations

Building on the workflow-specific hotspots identified, four targeted eco-design strategies emerge.

• (i)Optimization of data-processing workflows: Two days of computational processing per sequence drive a substantial fraction of the electricity-related impacts in untargeted UPLC-PDA/HRMS; integration of automated retention-index checking with manufacturer workflows is projected to halve the data treatment phase, simultaneously reducing operator workload, error risk, and environmental footprint.
• (ii)Equipment lifetime extension and mutualization: Device manufacturing contributes up to 58% of human toxicity (cancer) in the UPLC-UV workflow; instrument refurbishment, sourcing of reconditioned equipment, and shared-use platforms allow manufacturing impacts to be amortized over a larger number of analyses.
• (iii)Substitution of single-use consumables: Borosilicate vials, glass inserts, PTFE/silicone septa, and PTFE syringe filters were modeled as single-use in line with current SOP. As previously demonstrated in LCA studies dedicated to sample preparation [2], these single-use items often dominate the environmental profile over solvent use. Therefore, where contamination-control requirements allow, transition to reusable formats constitutes a direct reduction lever for HPLC–UV and UPLC-UV. The feasibility of such a transition depends on the conditions of reuse, namely the cleaning protocols (solvent rinses, washing cycles, drying) required to meet traceability and contamination-control standards, whose own environmental impacts (water, energy, cleaning agents) must be integrated into the assessment to confirm a net benefit over the single-use baseline.
• (iv)Method redesign to avoid dichloromethane: 64% climate change contribution of dichloromethane in GC–MS is unlikely to be addressed by a simple solvent substitution within the same method and likely calls instead for a redesign of the analytical method itself (e.g., alternative extraction or chromatographic strategies) so that the use of this solvent can be avoided altogether.

A substantial share of the environmental impacts arises upstream of laboratory activities, in equipment manufacturing, electricity generation, IT infrastructure, and the supply of high-purity reagents and gases. This does not diminish the role of the analytical chemist, but it extends that role from bench operation to workflow design.

Four life-cycle stages of the analytical method emerge as actionable levers. Method development determines solvent, gradient and replicate choices, and thereby a large share of downstream impacts. Equipment investment governs device manufacturing impacts through specifications, supplier selection, and the new versus refurbished trade-off. Data-processing workflow design controls the electricity demand of computation through automation and a validation strategy. Routine operation, through duty cycles, consumable reuse, and waste segregation, drives operational impacts.

Data-processing workflow design offers a twofold benefit. It halves the electricity-related impact of an indicator that becomes dominant on carbon-intensive grids (“Sensitivity to the electricity mix”), and it reorients the chemist from sequential validation of outputs toward interpretation, anomaly detection, and assessment of method robustness.

Solvent substitution, although the most visible intervention, would on its own leave the dominant contributors, device manufacturing, data processing, and single-use consumables, unaddressed in three of the four workflows examined. Integrating LCA-derived criteria across method development, procurement, computation, and routine operation defines a hybrid profile, the analytical chemist with applied LCA literacy, as a natural evolution of the discipline rather than a parallel specialty.

Limitations of LCA for analytical chemistry

Despite its decision-support value, applying LCA to analytical workflows faces several limitations. It should be noted that several of these limitations are not specific to LCA, as they reflect broader data and methodological gaps in general environmental assessment of analytical chemistry and also affect score-based metrics such as AGREE in comparable or greater measure.

• (i)Assessment is time- and expertise-intensive (typically days to weeks per workflow), which restricts its use to strategic decisions rather than fast routine screening. This is a genuine difference with AGREE, which can be computed in minutes.
• (ii)Current background databases (Ecoinvent and others) lack specific datasets for analytical-grade chemicals and for specialized laboratory equipment.
• (iii)Results for amortized devices are sensitive to assumed lifetime and number of sequences run, parameters that can shift per-analysis contributions substantially. Score-based metrics bypass this sensitivity only because they do not account for device manufacturing at all.
• (iv)Characterization factors for toxicity-related categories (human toxicity, freshwater ecotoxicity), based on USETox, carry high and well-documented uncertainties, which limit their interpretation in absolute terms; AGREE addresses toxicity through hazard pictograms and operator-exposure proxies, which avoid USETox uncertainties but provide no quantitative impact estimate and only focus on operator exposure.
• (v)Service-related flows (proprietary software, manufacturer maintenance, instrument calibration) are difficult to integrate into current frameworks and are likewise out of scope for AGREE. Improving the accuracy of these flows will require joint efforts between suppliers, who must provide greater transparency on material composition and manufacturing processes, and the LCA community, which needs to refine modeling approaches to better represent analytical-grade products. These are shared challenges for the field rather than weaknesses specific to LCA.

Conclusions

The two LCAs performed in this study demonstrate that analytical workflows show substantial variability in their environmental profiles, even when addressing similar analytical purposes. No universal hotspot can be inferred across the four workflows examined, confirming that assumptions commonly made in analytical chemistry, such as solvent dominance or the primacy of sample preparation, do not hold when assessed through quantified inventories. Environmental priorities instead depend on workflow configuration, sequence structure, equipment characteristics, and data-processing requirements. The results also show that the four workflows show distinct climate change hotspots even when having the same purpose (quantifying phenoxyethanol in our case).

LCA can directly help analytical optimization. In the natural-extracts workflow, LCA led to a revision of the chemoinformatics processing strategy, reducing the duration of the treatment phase and the associated impacts. In the phenoxyethanol comparison, the analysis supported a re-examination of sequence design and analytical quality requirements.

A broader use of LCA in analytical chemistry, however, requires stronger data availability. Foreground information on instrument composition and consumables remains sparse, and background datasets may not adequately represent laboratory-specific inputs. Developing analytical chemistry normalization factors, and potentially company-specific benchmarks, would allow laboratories to position methods relative to sectoral practices, identify whether an analytical development stands above or below typical environmental performance, and derive robust, science-based guidelines. Such a framework would support a more systematic integration of environmental impacts into analytical R&I and feed broader and more consistent sustainability policies at the corporate level.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

P.Riva, B. Raccary, and N.Stojiljkovic conceptualized the study and designed the methodology.

B. Raccary, C.Truan, and C.Pérès performed the Life Cycle Assessment modeling and data curation.

A. Dubuis, J. Héroult, M. In, S. Ramos, F. Leroy, A. Mureau, Chloe Brangeon, and V.Turlier contributed to the analytical data collection, method validation, and resources.

P.Riva, B. Raccary, N. Stojiljkovic, M.In, and A.Dubuis wrote the original draft.

All authors reviewed, edited, and approved the final manuscript.

Funding

This research was supported and funded by L’Oréal Research and Innovation.

Data availability

All datasets generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval

Not applicable. This study does not involve human participants or animal subjects.

Competing interests

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper. Priscilla Riva, Alexis Dubuis, Julien Héroult, Maxime In, Susana Ramos, Florence Leroy, Aurélie Mureau, Vanessa Turlier, and Natali Stojiljkovic are employees of L’Oréal Research and Innovation. Chloe Brangeon is a former employee of L’Oréal. Bastien Raccary, Clément Truan, and Christophe Pérès are employees of Le Labo Durable.

Source of biological material

Not applicable. This research does not involve biological material from human or animal sources.

Statement on animal welfare

Not applicable. This study does not involve animal subjects.