Plant vs animal protein: 18 long-term trials, no difference

Protein source is the oldest argument in sports nutrition, and almost all of it has been settled over eight to twelve weeks. A systematic review published in Frontiers in Nutrition in 2026 asked a narrower question: once a trial runs six months or longer, does the source still matter? It pooled 18 randomised controlled trials and 1,893 adults, with interventions of 6 to 31 months, and found nothing. Not a small effect — no effect, on any of the 18 outcomes examined.

A null result is a weak basis for a marketing line and a useful basis for a specification. If the source does not move the outcome over the long run, what a brand buys from an ingredient supplier is quantity, composition, and the method that proves both. That is a documentation question, and it is where this paper earns its keep for anyone writing a purchase spec for a multi-source protein system.

What the review compared

PubMed, Scopus and Web of Science were searched to 25 February 2025: 4,229 records, 3,379 screened after de-duplication, 43 read in full, 18 included. Trials had to be randomised, in adults, with a plant protein concentrate or isolate against an animal one, for at least six months. Protocol registered on PROSPERO (CRD42024604240).

The included trials were published between 1998 and 2021, with 10 to 158 participants each, aged 23 to 74, all parallel and double-blind. Seventeen of the 18 intervention arms used soy protein and one used rice. Controls used milk protein (9), whey (7), casein (1) and casein plus milk (1). Daily doses ran from 8 to 45 g, with 13 trials below 30 g. Fourteen trials enrolled women only, mostly peri- or postmenopausal. Three included exercise.

Eighteen outcomes, no separation

Pooled standardised mean differences were null across the board. The figure carries the 15 outcomes with usable intervals. Lean body mass came out at 0.26 (95% CI −0.58 to 0.90), fat mass 0.16 (−0.58 to 0.90), total body mass 0.37 (−0.33 to 1.07). Three outcomes sit outside the chart: fasting insulin 0.12 (−0.16 to 0.40) and HOMA-IR 0.18 (−0.10 to 0.46), from two trials each, and total cholesterol, printed as −0.38 with an interval of −0.55 to 0.13 — not symmetric around its own estimate. We reproduce it as published.

The two panels do not say the same thing. The cardiometabolic estimates sit close to zero with narrow intervals: a real absence of effect. The muscle and function estimates are null because they are imprecise. Lower-body strength pools three effect sizes with I² of 98.6% and spans −1.07 to 6.14 — an interval that wide excludes nothing. Panel A does not say the sources are the same. It says these trials cannot tell them apart.

Subgroups by age, dose, study quality, duration and exercise moved nothing. Two signals appeared only in secondary analysis: a triglyceride reduction confined to participants aged 60 and over, and small blood-pressure reductions that surface when single trials are removed. The authors call both exploratory. So do we.

Why shorter reviews disagree

Shorter syntheses of the same question have reported an advantage for dairy. The difference is follow-up length and comparator, not study quality.

Different estimators: Yimam and Piri Damaghi compare sources head to head, Li compares each source against placebo in a network. They are not interchangeable.

The 2026 review explains itself, and it is the most useful sentence in the paper. Among trials reporting baseline intake, habitual protein was already around 1.27 g/kg/day (range 0.89 to 1.85), and only three trials added a training stimulus. People who already eat enough protein and do not train have little headroom to show a source effect. The equivalence is conditional on that context.

What it means for a formulator

1. When the source stops being the variable, the assay becomes the product. If two sources behave alike at adequate intake, what a buyer pays for is the declared quantity — and that is only as good as the method behind it. Kjeldahl and Dumas measure nitrogen, not protein. EU labelling defines protein as total Kjeldahl nitrogen × 6.25 (Annex I, Reg. (EU) 1169/2011), while the dairy-specific factor is 6.38, so a multi-source blend specified against one factor will not declare what it contains. Nitrogen is also cheap to buy: a spec that stops at “≥80% protein” cannot separate protein nitrogen from free amino acid nitrogen. What makes that testable on goods-in is an amino acid profile by HPLC with an acceptance criterion per amino acid, free amino acids declared separately from total, and a stated deviation limit against the expected profile of the source mix.

Label figures are claims, not measurements. A 2025 validation study measuring creatine and taurine in commercial sports supplements by qNMR found deviations from label claim of up to +65.97% for creatine and +141.52% for taurine, while batch-to-batch variability within a product stayed at or below 8.49%. Different category, same lesson: consistency between batches says nothing about whether the batches are right.

2. The equivalence is dosed, and the dose was low. Thirteen of the 18 trials used under 30 g/day, the range started at 8 g, and almost none involved training. That supports substituting a source in a maintenance context. It does not support the claim that a 25 g sports serving built on soy isolate performs like one built on whey. For a training or clinical serving, the design target is leucine per serving — a composition parameter, not a percentage. Two blends declaring identical protein can deliver different leucine. Set the WPC / MPI / soy isolate ratio against an amino acid target and verify it per lot.

3. Two Annex II allergens arrive inside one purchased ingredient. A blend of whey concentrate, milk protein isolate and soy isolate carries both milk and soy, two of the 14 substances in Annex II of Reg. (EU) 1169/2011. Article 21 requires each to be named and typographically emphasised in the finished product's ingredient list; the trade name of a protein system does not discharge that, and the allergen declaration is a separate document from the certificate of analysis. One more parameter hides here: the authors attribute part of the exploratory blood-pressure signal to soy isoflavones, and isoflavone content is controlled by no protein-content specification. If a formulation depends on it, specify and measure it on its own. These are the documents to ask for when you request a specification.

Where this evidence is weak

The title says plant versus animal; the data say soy versus dairy. Seventeen of 18 intervention arms used soy and every control arm used a dairy fraction, so the review says almost nothing about pea, rice, potato or fava protein. One rice trial is not a basis for generalising.

Precision is the larger problem. I² exceeded 85% for nine outcomes and reached 98.6% for lower-body strength. Under GRADE, all five strength and physical-performance outcomes were rated very low certainty, body composition and most lipid outcomes low, and only HDL and blood pressure moderate. Risk of bias was low in 8 trials, of some concern in 8, high in 2. The LFK index flagged small-study effects for most outcomes, though estimates adjusted by Rücker's limit method stayed null — the reassuring part of that finding.

Two further caveats. The trials skew heavily towards peri- and postmenopausal women, so the population is not the one most sports formulations are sold to. And seven of the eight authors list an affiliation with a food company research centre. The conclusion is a null result, the least convenient direction for a commercial interest to push, but the affiliation belongs on the record.

One reporting inconsistency is worth flagging: the lean body mass interval is given as −0.58 to 0.90 in the text and −0.68 to 1.21 in the forest plot caption. We used the text figure. Either way, an absence of detected difference at this width is not evidence of equivalence. The question is still open — which is the argument for choosing a protein system on specification, analytical method and documentation rather than on whichever source the literature currently flatters.