Choosing a Protein Assay: Bradford, Lowry, BCA and A280 Compared Across Sample Matrices

Choosing a Protein Assay is an experimental M.Sc biochemistry dissertation that answers a question every working lab faces and most student reports ignore: when the same protein sample is measured by the Bradford, Lowry, BCA and A280 methods, the four answers disagree — so which method do you trust, for which matrix, and by how much?

The four assays rest on different chemistry. Bradford binds Coomassie Brilliant Blue G-250 to protein with a colour shift read near 595 nm; Lowry and BCA are copper-based reactions read at 750/562 nm respectively; A280 exploits the intrinsic absorbance of aromatic residues. Each has a documented Achilles' heel — detergents wreck Bradford, reducing agents and chelators wreck the copper assays, and A280 depends on amino-acid composition. Textbooks list the interfering substances, but rarely quantify the damage.

This study does the quantification. A bovine serum albumin standard system anchors the calibration; test matrices are everyday Indian lab materials — chickpea (bengal gram) flour extract, pasteurised toned milk and spinach leaf extract — plus buffer spiked with the classic interferents (Triton X-100, DTT, EDTA). For every method-by-matrix cell the protocol measures linearity, spike recovery, precision and the limit of detection, then compares methods pairwise with Passing-Bablok regression and Bland–Altman plots rather than the misleading correlation coefficient. Pre-declared acceptance criteria are fixed before the first pipetting session. The result is a decision table — the best assay per matrix with the measured bias to expect — that an examiner can audit line by line.

Every protein-based result in a biochemistry lab — an enzyme activity normalised to protein, a purification table, an expression yield — inherits the accuracy of the protein assay behind it. Yet assay choice is rarely justified: students default to Bradford because it is quick, or Lowry because the manual says so, without asking whether that assay's known interferents are sitting in their own extraction buffer. When two methods disagree on the same sample, there is usually no data to say which is right or by how much, and correlation coefficients get quoted as if agreement were proven.

The consequence is real: protein concentrations misreported by tens of percent, specific activities that cannot be compared across reports, and purification tables whose fold-change is an artefact of assay interference. There is a need for a designed, statistically honest method-comparison study on the actual matrices a departmental lab uses — flour extracts, milk, leaf lysates, detergent- and reductant-spiked buffers — that measures linearity, recovery, precision and pairwise bias per assay per matrix under pre-declared acceptance criteria, and states its conclusions as a selection table a working lab can act on.

Abstract

Protein quantitation is the first step of most biochemical work, yet the choice of assay is usually made by habit. Because Bradford, Lowry, BCA and A280 rest on different chemistry, they respond differently to the buffers, detergents and extracts that surround a target protein. This dissertation performs a designed method-comparison study across four assays and five matrices, quantifying sensitivity, interference, precision and pairwise bias, and converts the findings into a defensible selection table for routine lab use.

Introduction

A department lab that extracts protein from chickpea flour for a protease study, or measures casein in milk samples, or prepares spinach leaf lysates for chloroplast work, will meet all four assays at some point. The manufacturers' inserts compare the assays on buffer standards; the interference literature reports individual substances in isolation. What a working student needs is a single, statistically honest comparison on the actual matrices of this lab, with the bias expressed in the units that matter — the percentage error in a reported concentration.

Existing practice and the gap

Typical PG reports validate one assay against a BSA curve and stop. Correlation coefficients are quoted to imply agreement, which is statistically wrong: two methods can correlate at r = 0.99 while one reads forty percent higher. Interference studies present thresholds without recovery measurements. Published comparison papers rarely include plant-flour or dairy matrices, which behave very differently from serum. The gap: a matrix-resolved, acceptance-criteria-driven comparison with method-comparison statistics done correctly, reproducible on a departmental spectrophotometer with reagents already stocked.

Proposed study

• Design: 4 assays (Bradford, Lowry, BCA, A280) × 5 matrices (buffer control, chickpea-flour extract, pasteurised toned milk, spinach leaf extract, buffer + interferents) in a full factorial layout.
• Anchoring: matrix-matched BSA calibration, with standard addition where the matrix suppresses signal; calibration ranges fixed from preliminary scans and stated before measurement.
• Per cell: linearity (regression with residual inspection), accuracy by spike recovery at three levels, precision (n = 6 intra-day across three days), LOD and LOQ from blank replicates.
• Comparison: Passing-Bablok regression and Bland–Altman plots pairwise between assays per matrix, with the recovery bias reported as percent difference, plus two-way ANOVA on recovery with assay and matrix as factors.
• Deliverable: a selection table — recommended assay per matrix with expected bias, plus an interference profile ranked by severity.

Feasibility

• Technical: all four assays are routine on a double-beam spectrophotometer; the statistics run in R's mcr package and Excel; the design is sized to one bench, one analyst and departmental reagent stocks.
• Economic: reagents (Bradford dye, Folin–Ciocalteu, BCA kit, BSA) are standard departmental purchases; matrices are market-available chickpea flour, milk and spinach — the whole study costs less than a single commercial kit's price.
• Operational: the 20-week schedule interleaves wet-lab weeks with analysis weeks; the main risk (reagent lot change mid-study) is controlled by purchasing enough of one lot up front and logging lot numbers for every session.

In scope

• Four colorimetric/spectrophotometric protein assays — Bradford (595 nm), Lowry (750 nm), BCA (562 nm) and A280 — compared on five defined matrices.
• Matrices: aqueous buffer (control), chickpea (bengal gram) flour extract, commercial pasteurised toned milk, spinach leaf extract, and buffer spiked with Triton X-100, DTT and EDTA at fixed concentrations.
• BSA as the calibration protein throughout; percent-recovery and percent-bias as the reporting currencies.
• Wet-lab work sized for one analyst on one departmental spectrophotometer over a 20-week project slot.

Out of scope

• Amino-acid analysis or Kjeldahl nitrogen as a gold-standard reference — beyond the department's instrumentation; the study therefore reports relative bias between methods, honestly framed as such.
• Fluorescence-based or UV-microvolume instruments, mass-spectrometric quantitation and label-free quantification.
• Native protein purification beyond preparing the extracts; enzymatic activity assays are not part of the design.
• Interferents beyond the three named ones (urea, guanidine, non-ionic detergents other than Triton are noted as future extensions).

Research design: a factorial, experimental method-comparison study with pre-declared acceptance criteria — four assay methods × five matrices, each cell measured in triplicate wells and repeated on three independent days (a nested repeated-measures layout).

Instruments and materials: double-beam UV-Vis spectrophotometer (or microplate reader with path-length correction), 0.1 mg analytical balance, calibrated variable micropipettes, Class-A volumetric glassware; BSA fraction V, Coomassie reagent, Lowry reagents (or a kit), BCA working reagent, Triton X-100, DTT, EDTA; matrices from local market sources, prepared under a written SOP.

Analysis plan (pre-declared): linearity accepted if residuals show no systematic curvature; recovery window 80–120%; precision target CV ≤ 10%; LOD = 3.3 σ_blank/slope and LOQ = 10 σ_blank/slope; Passing-Bablok fitted with 95% CIs on slope and intercept; bias quoted as mean percent difference with its CI. Two-way ANOVA with assay and matrix factors on percent recovery; post-hoc Tukey test if the interaction is significant. All data entry is double-checked against the bound raw notebook before analysis.

The study is organised as a measurement pipeline: preparation produces defined matrices, a calibration layer anchors each assay to BSA, the measurement layer fills the 4 × 5 factorial grid, and an analysis layer converts raw absorbances into defensible statements about bias.

flowchart TD
  A["Matrix preparation: buffer, chickpea extract, milk, leaf lysate, interferent spikes"] --> B["Working dilutions under written SOPs"]
  C["BSA fraction V stock, gravimetrically prepared"] --> D["Matrix-matched calibration curves per assay"]
  B --> E["4 x 5 factorial measurement grid: Bradford, Lowry, BCA, A280"]
  D --> E
  E --> F["Raw absorbances logged with date, analyst, reagent lot"]
  F --> G["Calibration regression + residual diagnostics"]
  F --> H["Spike recovery at three levels per cell"]
  F --> I["Precision: intra-day n=6, three inter-day repeats"]
  G --> J["Passing-Bablok regression + Bland-Altman plots per matrix"]
  H --> K["Two-way ANOVA: assay x matrix on recovery"]
  I --> L["LOD / LOQ from blank statistics"]
  J --> M["Selection table: best assay per matrix, expected bias"]
  K --> M
  L --> M

Three design decisions deserve notice. First, calibration is matrix-matched, not buffer-only: standards are prepared in each matrix's diluent so that suppression or enhancement is measured rather than assumed, with standard addition as the arbiter where the two disagree. Second, every absorbance row carries its provenance — date, analyst, reagent lot, matrix batch — because lot-to-lot drift is a confounder this design must be able to see. Third, the analysis layer deliberately replaces correlation with method-comparison statistics: Passing-Bablok is robust to outlier pairs, and Bland–Altman plots expose proportional bias that an r-value conceals, which is the statistical heart of the dissertation.

• Add a reference-grade anchor (Kjeldahl nitrogen or amino-acid analysis through a collaborating lab) to convert relative bias into absolute accuracy.
• Widen the interferent panel: urea, guanidine hydrochloride, Tween-20, imidazole and reducing-agent cocktails at gradient concentrations.
• Test microplate-reader versus cuvette implementations of the same assay, since path-length differences shift both range and bias.
• Extend to a real application: re-quantify a departmental enzyme-preparation series with the winning assay and recompute the purification table.
• Package the R analysis into a small Shiny app so junior students can upload absorbances and get the bias plots automatically.