What we found — and what we didn't

Before building and screening a random mutant library, we first asked whether growth on cellobiose as the sole carbon source could select for improved β-glucosidase-like activity directly — pool a library, grow it on cellobiose over repeated rounds, and let only the active variants multiply, without screening any colonies by hand. E. coli expressing wild-type CxnA was compared against the same strain expressing a secreted β-glucosidase as a positive control, each grown in parallel on glucose (a fitness control) and on cellobiose (the actual test).

(A) OD₆₀₀ at 24 h. The two strains are indistinguishable on glucose (1.82 vs 1.87), ruling out a general fitness difference. On cellobiose, the β-glucosidase control reaches 0.53 against 0.23 for CxnA — the latter close to the casamino-acid-only background, confirming CxnA itself doesn't support growth on cellobiose. (B) Viable counts by serial dilution plating tell the same story independently. (C) Cumulative enrichment over repeated rounds of liquid selection, against the roughly 10,000-fold enrichment a rare variant would need to become recoverable.

CxnA was carried on a pSB1C3 plasmid backbone with a lac promoter and ribosome binding site ahead of the insert. Transforming E. coli with this construct retained both exo- and endo-glucanase activity, confirming the starting enzyme was functional before any mutagenesis began.

Plasmid map of the CxnA construct. CxnA itself is only two domains — Cex (exoglucanase) and CenA (endoglucanase) — joined through a carbohydrate-binding module (CBM). The duplicated chloramphenicol-resistance region on the right was an unplanned discovery (below).

Error-prone PCR (epPCR) was applied only to the CxnA insert, using manganese chloride to deliberately lower the copying polymerase's accuracy — the backbone, promoter and ribosome binding site were amplified separately with a high-fidelity polymerase so mutations wouldn't land somewhere that would confound the activity readout. We ran this on two versions of the gene in parallel: the native sequence, and a codon-optimised version ordered as synthetic DNA blocks (to work around the gene's unusually high, amplification-hostile GC content — 71%).

24 codon-optimised and 15 native-sequence colonies were recovered. Fifteen of the 24 codon-optimised colonies were picked forward, giving the 30 mutants screened below.

Congo red dye binds intact cellulose (here, the related polymer CMC) but washes off wherever an active endoglucanase has chewed it up — so a colony with working CenA activity shows up as a clear halo on a red background. This was run first, as a cheap way to immediately discard any mutant that had simply broken the protein.

Congo red screen. (A) All 15 codon-optimised mutants, numbered. (B) All 15 native-sequence mutants (only 15 total colonies were available in that library). A clear halo = endoglucanase activity retained.

X-gluc is a colourless compound that turns a colony visibly blue if the cell is producing β-glucosidase-like activity. Every mutant from both libraries — 30 colonies total — was screened this way in parallel with Congo red, alongside controls: a known β-glucosidase (positive control), wild-type CxnA, and a non-functional His-tagged CxnA (negative control).

Controls for both screens. Left: Congo red. Right: X-gluc — note that wild-type CxnA (top right spot) already shows a faint blue tint, not just the positive-control β-glucosidase (bottom left).

X-gluc screen on all 30 mutants. (A) codon-optimised. (B) native sequence.

Congo red and X-gluc are useful for sorting mutants into rough buckets, but neither gives a number you can actually compare between colonies. To measure exoglucanase activity precisely, we used MUC — a substrate that releases a fluorescent molecule (4-MU) in proportion to how much active enzyme is present. Getting trustworthy numbers out of this assay took six rounds of calibration (the full story — including a substrate mix-up that briefly looked like a real signal — is in the lab notebook); the validated version is summarised below.

Establishing a trustworthy assay. (A) Fluorescence is only proportional to 4-MU concentration below 6.25 µM — above that, the reading undershoots for optical reasons, not biology, so all rates were fit only within the reliable range. (B) Adding more lysate gave progressively less signal per microlitre — a volume artefact, not reduced activity. (C) Fixed by titrating active lysate against inactive (frameshift) lysate instead, which restored a straight-line relationship at two different temperatures/settings. (D) No signal from substrate, medium, or host alone — the frameshift control sits at just 0.7% of induced CxnA.

With the assay validated, we quantified exoglucanase activity for the seven mutants flagged by the earlier screens, each compared against a wild-type control grown and lysed alongside it on the same day. Seven variants were carried forward to the quantitative assay, limited by the capacity of a single 96-well plate once controls, blanks and standards were accommodated. The selection deliberately spanned both extremes of the Congo red screen — variants with the largest halos and variants with no clearing at all — rather than sampling only the apparent improvers. Native colony 12, which also showed no clearing, was not carried forward for this reason; T14 and T15 represent that class.

Activity of all seven flagged mutants, relative to a same-day wild-type control (bars = median of 3 technical replicates; whiskers = range). The shaded band marks how much two colonies of the same, unmutated genotype normally differ from each other just from day-to-day prep variation — only a bar that clears this band by itself is a plausible real effect.

Eight samples were sent for Sanger sequencing (MRC PPU DNA Sequencing and Services): the seven variants carried through the quantitative screen, plus the unmutated codon-optimised construct as a parental reference. Plasmid DNA was isolated by miniprep (Qiagen) and quantified on a NanoDrop 2000 (Table 2).

All four codon-optimised variants — CO2, CO4, CO7 and CO12 — returned the parent sequence with no mutations detected. Sequencing failed for T14 and T15, so neither of the two dead variants can be characterised genetically. Of the seven variants, only T4 returned usable evidence of a mutation: a single point substitution in each domain, changing residue 14 from GGC to GAC (glycine → aspartic acid) in the exoglucanase domain, and residue 614 from aspartic acid to glycine in the endoglucanase domain. Changes flagged elsewhere in the chromatograms were consistent with signal decay rather than real substitutions.

Read-length limits left an unsequenced gap in the middle of every read. The gap spans the carbohydrate-binding module in all samples and extends into the Cex or CenA coding sequence in some, and its extent differs from sample to sample — so coverage is not uniform, and a mutation falling inside the gap would not have been detected.

No mutant demonstrated a confirmed improvement in activity. CO12, the only variant to read above wild type, returned the parent sequence — placing its 1.35× average inside the range spanned by genetically identical isolates, so it is not a candidate for a further round. The screen's useful output is a quantified noise floor rather than a hit: variation between preparations is wide enough that a real improvement would have to be considerably larger than 1.4× before a single-colony measurement could resolve it. Two variants, T14 and T15, lost function entirely; sequencing failed for both, so the explanation rests on phenotype — simultaneous loss of two activities whose active sites lie several hundred residues apart is more easily explained by loss of the full-length protein than by two independent active-site mutations.

What did work: we built a complete, repeatable pipeline — random mutagenesis, two fast qualitative screens, a validated quantitative fluorescence assay, and sequencing to close the loop — that took a library from 30 random colonies to a quantified detection limit and one genetically characterised loss-of-function variant, in a single round. The library was small and the GC-rich gene was a genuine technical obstacle throughout; a larger library run through the same pipeline is the obvious next step, not a different approach.

Two observations point the same way about the codon-optimised library. All four of its variants that were sequenced returned the parent sequence, and all 15 of its colonies retained a Congo red halo — against 7 of 15 for the native-sequence library. The simplest reading is that error-prone PCR introduced few or no mutations into the codon-optimised construct, consistent with the gBlock-and-Gibson route reconstituting parental sequence, and that the native library is where this project's real variants are: T4, with a substitution in each domain, and T14 and T15, both of which lost function entirely. Only four of the 15 codon-optimised colonies were sequenced, so this is the most parsimonious explanation rather than an established one.