Paperonium

← Back to the plain-language case

Figure 6D calls a TRPV4 knockout effect significant at p = 0.06706, above the 0.05 threshold

J Neurosci (2018) · doi:10.1523/JNEUROSCI.0897-18.2018 · PMC6181316

Summary

The Figure 6D legend says that conventional TRPV4 knockout (TRPV4KO) mice "showed significantly less photoreceptor cell death than WT mice" 24 h after retinal detachment. It marks the comparison with an asterisk and prints p = 0.06706. The Methods state one decision rule for these TUNEL comparisons: "p < 0.05 was considered significant". Since 0.06706 > 0.05, the comparison is not significant under the paper's own rule. Nothing in the paper (a one-sided test, a different threshold, a correction notice) reconciles the two. Of the 16 exact p-values printed in the paper, this is the only one above 0.05. The same comparison is the first result stated in the Abstract (a ~50% reduction in apoptotic photoreceptors). The paper's main conclusion still has other support, including the Müller-glia conditional knockout in the same legend (p = 0.020863, which is significant and consistent).

What the paper reports

Methods, "Statistical analysis":

The Mann–Whitney U test was used to compare unpaired values of TUNEL-positive cell density and proinflammatory cytokine concentration. The data analyses were performed using Excel 2016 (Microsoft) with add-in software Statce14@. p < 0.05 was considered significant.

Figure 6 legend, panel D:

Conventional TRPV4KO mice showed significantly less photoreceptor cell death than WT mice 24 h after retinal detachments (*p = 0.06706, n = 7 or 8 eyes). The TRPV4CKO mice showed significantly less photoreceptor cell death than WT mice (**p = 0.020863, n = 6 eyes). The TRPV4CKO results were perfectly matched with those in TRPV4KO, indicating that Müller glial TRPV4 activation accelerates photoreceptor cell death in RD.

Abstract:

After RD, the number of apoptotic photoreceptors was reduced by ∼50% in TRPV4KO mice relative to wild-type mice, indicating the possible involvement of TRPV4 activation in RD-induced photoreceptor cell death.

Results, TUNEL paragraph. This passage gives the size of the effect but does not call the KO result significant:

The number of dead photoreceptors in TRPV4KO mice was 50% less than that for WT mice (Fig. 6D).

Consistent with those previous reports and our data (Figs. 1–5), the number of dead photoreceptors in TRPV4CKO mice was 50% less than that for WT mice, and the results were perfectly matched with the result in the conventional TRPV4KO (Fig. 6D).

Discussion, second paragraph:

Indeed, our Müller glia-specific conditional KO mice can significantly reduce the RD-induced photoreceptor cell death (Fig. 6), and the reduction ratio of dead photoreceptors was perfectly the same as the conventional TRPV4KO mice (Fig. 6).

What the paper's own numbers show

Method. The paper gives no per-eye TUNEL counts, group summaries or U statistics, so the test cannot be rerun. Two checks are possible. First, compare each printed p with the threshold stated in the Methods. Second, ask whether 0.06706 is a plausible two-sided Mann–Whitney p for the stated sample sizes, and so whether it could be a misprint of a smaller value. For the second check I enumerated the exact null distribution of U for n = 7 vs 8 (6435 rank arrangements, assuming no ties), and for 7 vs 7 and 8 vs 8 because the legend says "7 or 8 eyes". I also computed the normal approximation with and without continuity correction. For context I collected every exact p-value printed in the paper. All numbers below are printed by recompute.py.txt.

Quantity Location Reported Recomputed
Significance threshold Methods, Statistical analysis "p < 0.05 was considered significant" α = 0.05
WT vs conventional TRPV4KO, TUNEL+ photoreceptors at 24 h Fig. 6D legend "significantly less", *p = 0.06706, n = 7 or 8 eyes 0.06706 > 0.05 by 0.01706 (1.34 × the threshold); not significant under the stated rule. Equivalent |z| = 1.831, against 1.960 needed
WT vs TRPV4CKO Fig. 6D legend "significantly less", **p = 0.020863, n = 6 eyes 0.020863 < 0.05; significant, consistent
Other exact p-values in the paper legends of Figs. 4, 5, 7, 8, 9, 10 14 values, from 0.000116 to 0.012851 all < 0.05. Of the 16 printed exact p-values, 0.06706 is the only one ≥ 0.05; the next largest is 0.020863
Hypothetical one-sided reading not stated in the paper — 0.06706 / 2 = 0.03353 (would be < 0.05). If 0.06706 were already one-sided, the two-sided p would be 0.13412
Exact two-sided Mann–Whitney p near the printed value, n = 7 vs 8, no ties recomputation 0.06706 U = 11: 0.05408; U = 12: 0.07211. The printed value lies between them, and both are above 0.05
Normal approximations closest to 0.06706, n = 7 vs 8 recomputation 0.06706 U = 12: 0.06408 (no continuity correction), 0.07285 (with); U = 11 with correction: 0.05620
Same check for n = 7 vs 7 and 8 vs 8 recomputation 0.06706 7 vs 7: exact 0.05303 / 0.07284 (U = 9 / 10); 8 vs 8: exact 0.06496 / 0.08298 (U = 14 / 15)
Exact p near a misprint value of 0.006706, n = 7 vs 8 recomputation — U = 5: 0.00591; U = 6: 0.00932. A value near 0.0067 is also within the attainable range, so a misprint cannot be excluded

The printed 0.06706 is not itself an exact no-ties value for any of these sample sizes. Ties in the TUNEL densities, or a normal approximation as statistics add-ins commonly use, would give values between the exact steps. As printed, 0.06706 falls where an ordinary two-sided Mann–Whitney result for about 7–8 eyes per group would fall when the rank separation is just short of significance (U = 11–12 of a possible 56 for 7 vs 8).

Figure

Reported Fig. 6D p-values against the paper's stated threshold and the neighbouring exact Mann-Whitney values

On a linear p-value axis starting at 0, the two orange dots are the p-values printed in the Figure 6D legend. Both comparisons are described there as "significantly less". The vertical line is the threshold stated in the Methods (p < 0.05), and the shaded band is the region the paper counts as significant. WT vs TRPV4CKO (p = 0.020863) lies inside it. WT vs conventional TRPV4KO (p = 0.06706) lies outside it. The blue open squares are recomputed: the two exact two-sided Mann–Whitney p-values on either side of the printed value for n = 7 vs 8 eyes with no ties (U = 11: 0.0541; U = 12: 0.0721). Both are above 0.05. Taken as printed, the knockout result is a two-sided value just short of significance.

All 16 exact p-values printed in the paper, on a log axis, with the 0.05 threshold

All 16 exact p-values printed in the paper's figure legends, sorted and shown on a log scale. Every open dot is a comparison the legends call significant, and each lies below 0.05. The filled dot (Fig. 6D, WT vs TRPV4KO, p = 0.06706) is the only printed value above the threshold, and the legend calls it "significantly less" all the same. Values are copied from the legends; panel letters follow the legends' own labels.

Why it matters

The WT vs conventional TRPV4KO comparison is the first result in the Abstract ("reduced by ∼50% in TRPV4KO mice relative to wild-type mice"). It is also one half of an argument made twice in the paper: the CKO effect is "perfectly matched" with, or "perfectly the same as", the conventional KO effect, and this match is used to assign the effect to Müller glial TRPV4. Under the paper's criterion, the KO side of that comparison is a non-significant trend (p = 0.067, about 7–8 eyes per group), not an established reduction. The direction of the effect does not change. The 50% figure cannot be checked because no group values are given.

The paper's overall conclusion, that Müller glial TRPV4 contributes to photoreceptor death after retinal detachment, is weakened but not overturned. In the same legend, the Müller-glia/astrocyte conditional knockout reaches significance (p = 0.020863, n = 6). The agonist GSK increases photoreceptor death (Fig. 7A, p = 0.002622). The antagonist HC reduces it (Fig. 7B, p = 0.009823). The MCP-1 and macrophage experiments (Figs. 8–10) point the same way. Readers who rely on the global-knockout result as an independent replication, or quote the Abstract's ~50% reduction as statistically established, are relying on a comparison that the paper's own numbers do not support at α = 0.05.

Explanations considered

Status

As of 2026-10-23:

Suggested correction

The Figure 6D legend should be made consistent with the Methods. There are two possibilities:

  1. If p = 0.06706 is correct: change the legend to describe the KO vs WT difference as a non-significant trend, for example "TRPV4KO mice showed a trend toward less photoreceptor cell death than WT mice (p = 0.067, n = 7 or 8 eyes)". Remove the asterisk from the figure. Qualify the Abstract's "reduced by ∼50%" sentence and the "perfectly matched" / "perfectly the same" statements to match.
  2. If the printed value is a misprint: give the correct p-value.

In either case, it would help readers if the authors stated the group sizes for each genotype (which group had 7 eyes and which had 8), the test used for Fig. 6D, whether it was one- or two-sided and exact or approximate, and the group summaries (median and range, or the per-eye values) behind the "50% less" figure.