Paperonium

← Back to the plain-language case

IL-6 × leptin feeding interaction: reported F(1,11) = 0.563 gives p = 0.47, not 0.05

Cell Rep (2019) · doi:10.1016/j.celrep.2019.02.044 · PMC6418345

Summary

The Results report a two-way repeated-measures ANOVA on 6-h chow intake after co-injection of subthreshold leptin and IL-6 into the lateral parabrachial nucleus (lPBN; Figure 3J, n = 12 rats). They give the IL-6 × leptin interaction as "F(1, 11) = 0.563, p = 0.05". An F of 0.563 on (1, 11) degrees of freedom has an upper-tail p of 0.469. A p of 0.05 would take F = 4.84, about 8.6 times larger. The reported F and p therefore cannot both be correct. Taken at face value, the F shows no statistically significant interaction. This interaction is the only test of interaction reported for the paper's claim, repeated in the abstract, highlights, Discussion and conclusion, that IL-6 and leptin cooperate in the lPBN to reduce feeding. Every other F/p pair in the Results (15 of 16) is consistent with its recomputed p.

What the paper reports

Results, "Cooperative Interaction of Leptin and IL-6 on Food Intake Control":

Results indicated that IL-6 and leptin indeed interact to suppress food intake (two-way ANOVA: IL-6 alone: F(1, 11) = 0.21, p > 0.05; leptin alone F(1, 11) = 0.88, p > 0.05; interaction IL-6 × leptin F(1, 11) = 0.563, p = 0.05; Figure 3J).

Same paragraph, two sentences later:

BAT region temperature was unaffected (IL-6 alone: F(1, 11) = 4.524, p > 0.05; leptin alone F(1, 11) = 7.652, p < 0.05; interaction IL-6 × leptin F(1, 11) = 1.528, p > 0.05; Figure 3K).

Same paragraph, interpretation:

These results suggest that leptin and IL-6 can acutely act in concert, possibly by sharing a common downstream target, ultimately amplifying each other’s reduction of food intake.

Abstract (Summary):

Parabrachial IL-6 interacted with leptin to reduce feeding.

Highlights:

IL-6 interacts with leptin at the level of the lPBN to reduce food intake

Discussion, third paragraph:

In line with our hypothesis, injection of subthreshold doses of both ligands produced a clear reduction in feeding when co-applied to the lPBN indicating that IL-6 can amplify leptin’s anorexic effects at the level of the lPBN.

Discussion, concluding paragraph:

The latter point, along with the identified here cooperative interaction of exogenously applied IL-6 with leptin in the lPBN, highlights the emerging potential importance of the local circuit identified here for development of obesity therapeutics.

Figure 3 legend (panels J and K):

In a new cohort of 11-week-old male rats (n = 12) chow intake and BAT temperature were measured after intra-lPBN delivery of sub-threshold doses of IL-6 (0.1 μg) or leptin (0.1 μg) alone, or delivery of the two substances together into the lPBN.

two-way ANOVA (J and K) and one-way ANOVA (B–E, G, and I) were used for comparisons with Sidak’s post hoc test.

STAR Methods, microinjections, and Quantification and Statistical Analysis:

All injections were performed in a counterbalanced Latin square design, where each animal received each treatment.

p < 0.05 was considered statistically significant.

What the paper's own numbers show

Method. For every F statistic in the Results, which reports them in three sections (the lPBN leptin dose-response, Figure 3B–3I; the leptin/IL-6 interaction experiment, Figure 3J/3K; and the lPBN IL-6 knockdown, Figure 4G/4H), I computed the ANOVA p-value as the upper-tail probability P(Fdf1, df2 ≥ F) with scipy.stats.f.sf, using the F and degrees of freedom exactly as printed. With 12 rats each receiving all four treatments, F(1, 11) is the expected test for each effect in a 2 × 2 within-subject ANOVA. For df1 = 1, F equals t², so I also report the equivalent t and its one-sided p. All numbers below are printed by recompute.py.txt.

quantity location reported recomputed
IL-6 × leptin interaction, 6-h chow intake, F(1, 11) = 0.563 Results, Fig 3J p = 0.05 p = 0.469
F(1, 11) needed for p = 0.05 derived (implied by p = 0.05) 4.844 (8.6 × the reported F)
F(1, 11) range whose p rounds to 0.05 (0.045 ≤ p < 0.055) derived n/a 4.61 to 5.11
Equivalent t(11) = √0.563; one-sided p derived n/a t = 0.750; p = 0.234
p for F = 0.563 with df1 = 1 and any df2 from 1 to 106 derived n/a 0.45 to 0.59
Smallest F (df1 = 1) that can reach p < 0.05 for any df2 derived n/a 3.841
Leptin dose, 6-h chow intake, F(2, 10) = 5.496 (printed "F (2, 10)") Results, Fig 3B p < 0.05 p = 0.0245 (consistent)
Leptin dose, 12-h chow intake, F(2, 10) = 6.685 Results, Fig 3C p < 0.05 p = 0.0143 (consistent)
Leptin dose, 24-h chow intake, F(2, 10) = 7.464 Results, Fig 3D p < 0.05 p = 0.0104 (consistent)
Leptin dose, 24-h body weight, F(2, 12) = 7.55 Results, Fig 3E p < 0.01 p = 0.00754 (consistent)
Leptin dose, core body temperature, F(2, 34) = 1.765 Results, Fig 3G p > 0.05 p = 0.187 (consistent)
Leptin dose, locomotor activity, F(2, 34) = 3.493 Results, Fig 3I p < 0.05 p = 0.0417 (consistent)
IL-6 main effect, chow intake, F(1, 11) = 0.21 Results, Fig 3J p > 0.05 p = 0.656 (consistent)
Leptin main effect, chow intake, F(1, 11) = 0.88 Results, Fig 3J p > 0.05 p = 0.368 (consistent)
IL-6 main effect, BAT temperature, F(1, 11) = 4.524 Results, Fig 3K p > 0.05 p = 0.0569 (consistent)
Leptin main effect, BAT temperature, F(1, 11) = 7.652 Results, Fig 3K p < 0.05 p = 0.0184 (consistent)
IL-6 × leptin interaction, BAT temperature, F(1, 11) = 1.528 Results, Fig 3K p > 0.05 p = 0.242 (consistent)
Knockdown, interaction, F(17, 544) = 8.345 Results, Fig 4G p < 0.01 p = 5.05 × 10-19 (consistent)
Knockdown, treatment, F(1, 32) = 7.548 Results, Fig 4G p < 0.05 p = 0.00978 (consistent)
Knockdown, interaction, F(17, 544) = 4.944 Results, Fig 4H p < 0.01 p = 5.28 × 10-10 (consistent)
Knockdown, treatment, F(1, 32) = 3.462 Results, Fig 4H p > 0.05 p = 0.072 (consistent)

Fifteen of the sixteen F/p pairs in the Results agree with the recomputation. The one that does not is the interaction on which the cooperative-effect claim rests. Its recomputed p is 9.4 times the printed value.

Figure

Reported and recomputed p-values for the six F tests of the leptin/IL-6 interaction experiment (Figure 3J and 3K), on one log-scaled p axis

Figure 1. Each row is one F test from the Results paragraph on the leptin/IL-6 interaction experiment (Figure 3J, 6-h chow intake; Figure 3K, BAT temperature). Orange shows what the paper reports: a bound ("p > 0.05" or "p < 0.05") is drawn as the range it allows, and the one exact value (p = 0.05) as a dot. Blue dots are the upper-tail p-values recomputed from the printed F and degrees of freedom. The thin vertical line marks p = 0.05; the paper's own criterion is p < 0.05. Five recomputed values fall inside the range the paper states. For the IL-6 × leptin interaction on chow intake, F(1, 11) = 0.563 gives p = 0.469, not 0.05.

The F(1,11) distribution with the reported F = 0.563, the F of 4.84 that p = 0.05 would require, and the decimal-shift reading F = 5.63

Figure 2. The density of the F distribution with (1, 11) degrees of freedom. The p-value of an ANOVA F test is the area to the right of the observed F. For the reported F = 0.563 (blue line) that area, shaded light blue, is 0.469. The reported p = 0.05 corresponds to the area to the right of F = 4.84 (orange line); that tail is too thin to see at this scale, and the arrow marks where it lies. The dashed line marks F = 5.63, a possible decimal-shift reading of the printed value, which would give p = 0.037. Nothing in the published material confirms that reading.

Why it matters

Explanations considered

Status

Suggested correction

The authors should consult the original ANOVA output for the Figure 3J experiment and correct the interaction statistic in the Results sentence

interaction IL-6 × leptin F(1, 11) = 0.563, p = 0.05

so that F and p agree. If F = 0.563 is correct, the interaction p is 0.47. The statements that IL-6 and leptin interact or cooperate to reduce feeding (abstract, highlights, Results, Figure 3 title, Discussion and conclusion) would then need to be revised or withdrawn, since neither main effect nor the interaction is significant. If the F value is the error (for example 5.63, for which p = 0.037), the corrected F and its exact p should be printed. In either case, as printed, p = 0.05 does not satisfy the paper's p < 0.05 criterion. If p = 0.05 is a Sidak post hoc comparison rather than the interaction test, it should be labelled as such and reported separately from the interaction F.