The final dataset consisted of 38 studies reporting on HIFA mRNA abundance in 25 species of ray-finned fishes exposed to laboratory hypoxia (Table 1). Across all studies, there were 413 discrete sampling intervals, representing different fish species, HIFA genes, tissues sampled, and severity of hypoxia (DO level and duration). HIFA mRNA abundance was significantly higher in tissues of hypoxic fish compared to normoxic controls in 182 cases (44%), unchanged in 214 cases (52%), and significantly lower in 17 cases (4%).

Thirty-four studies described the effects of hypoxic exposure on HIF1A mRNA levels, 13 described changes in HIF2A mRNA, and five described changes in HIF3A (Supplementary Figure S1A). All but two studies were on species from lineages arising after the teleost-specific genome duplication (Hughes et al., 2018), meaning that certain species were predicted to have teleost-specific “a” and “b” duplicates (Rytkonen et al., 2013; Townley et al., 2022). The exceptions were Wang X. et al. (2021), who reported on HIF1A, HIF2A, and HIF3A in Siberian sturgeon (Acipenser baerii), and Rimoldi et al. (2016), who reported on HIF2A from spotted gar (Lepiosteus oculatus).

Apart from these reports on basal taxa, 21 studies examined changes in HIF1Aa in Neoteleostei (more derived ray-finned fishes) and 13 examined changes in HIF1Ab in Otocephala. For HIF2A, all studies reported data for HIF2Aa, even though HIF2Ab is found in most fish lineages (Rytkonen et al., 2013; Townley et al., 2022). The lack of data on HIF2Ab is a critical gap in knowledge, considering that recent work suggests that sequence variation in this duplicate is associated with variation in hypoxia tolerance among ray-finned fishes (Babin et al., 2024). For HIF3A, only a single teleost-specific duplicate has been retained in ray-finned fishes (Townley et al., 2022), and we did not distinguish between “a” and “b” duplicates. No studies were found on HIF4A.

Prior to assessing differences in HIF1A, HIF2A, and HIF3A, we compared the frequency of reported hypoxia-dependent increases in mRNA abundance for HIF1Aa and HIF1Ab (excluding data for ancestral HIF1A from Siberian sturgeon). No significant difference was found in the frequency of reported increases in the levels of HIF1Aa and HIF1Ab mRNA (p = 0.160) (Supplementary Table S1). Thus, data were combined for teleost-specific duplicates, along with ancestral forms, of each gene. Hereafter, HIF1A, HIF2A, and HIF3A refer to data from all forms of each gene.

The effect of HIFA gene on the frequency of reported hypoxia-induced increases in mRNA abundance approached statistical significance (Supplementary Table S2). The frequency of increased HIF1A mRNA tended to be higher than that for HIF2A (p = 0.0525), but lower than that for HIF3A (p = 0.0566) (Figure 1A). Law et al. (2006) used Northern blot analysis and found higher HIF3A mRNA levels in several tissues of the grass carp (Ctenopharyngodon idella) under hypoxia when HIF1A mRNA levels were largely unaffected in the same tissues. These observations, combined with the widespread expression of HIF3A in tissues of normoxic fish (Townley et al., 2022), suggest that more attention should be given to this understudied gene (Duan, 2016).

Measurements of mRNA abundance were not uniformly distributed among tissues (Supplementary Figure S1B). Across all HIFA genes, the most well-documented tissue was liver, followed by gill, brain, skeletal muscle, heart, kidney, ovary, intestine, spleen, testes, and swim bladder. For HIF1A, the frequency of hypoxia-induced increases in mRNA was highly variable and not significantly different across tissues (Figure 1B; Supplementary Table S3), although there tended to be fewer reports of increased HIF1A mRNA in skeletal muscle (p = 0.113). For HIF2A, there was a trend toward a higher frequency of reported effects of hypoxia in gill (p = 0.0782; Supplementary Table S4). Interestingly, HIF2A is more highly expressed in gill than in other tissues even under normoxia (Townley et al., 2022) and it may play a role in oxygen-sensing in fishes (Pan et al., 2022). For HIF3A, only one tissue (liver) was measured in at least three studies, precluding analysis of the effects of tissue on mRNA changes during hypoxia.

We evaluated the effects of exposure conditions (COD, temperature, and salinity) on the frequency of reported increases in HIF1A mRNA, pooling measurements from all tissues (Supplementary Table S5). Surprisingly, COD, which reflects both the duration and the depth of hypoxia (Nelson and Lipkey, 2015), had no effect on HIF1A abundance (p = 0.918; Figure 1C). The only pattern was that the variability in responses increased as a function of COD. Because we pooled all tissues for this analysis, it is possible that tissue-specific responses were obscured (e.g., increasing mRNA levels over time versus decreasing over time in different tissues). Consequently, we repeated the analysis on the two most-studied tissues, liver and gill; COD was not statistically related to the frequency of increased abundance of liver HIF1A (p = 0.786) or gill HIF1A (p = 0.472). For liver HIF1A, some studies reported higher mRNA levels with increasing time of exposure, as one might expect (Froehlich et al., 2015; Heinrichs-Caldas et al., 2019). However, there were more studies where HIF1A was not different from controls at any time point (Li et al., 2017; Murphy et al., 2023; Wang X. et al., 2021; Yang et al., 2017) or it was higher at all time points (Huang et al., 2022; Luo et al., 2021). Thus, there was no consistent pattern across all studies due to the duration or severity of hypoxia.

We also evaluated temperature and salinity as potential factors related to the increase in HIFA mRNA abundance. We observed a non-significant trend toward a greater frequency of increased HIF1A mRNA at increasing temperature (p = 0.114; Figure 1D). Using Northern blot analyses, Rissanen et al. (2006) presented evidence of an interaction between temperature and hypoxia exposure on levels of HIF1A mRNA in Crucian carp, although this effect differed among tissues. When data from freshwater species were grouped and compared to data pooled from estuarine and marine species, there was no effect of salinity on the frequency of elevated HIF1A mRNA (p = 0.394; Figure 1E). Like the effects of temperature, though, direct tests of the effects of salinity on HIFA mRNA abundance in single study are scarce. COD, temperature, and salinity were unrelated to the frequency of reported increases in HIF2A or HIF3A mRNA (Supplementary Tables S6, S7; Supplementary Figure S2).

We determined whether the number of MIQE criteria deemed essential for ensuring data uniformity, comparability, and reliability (Bustin et al., 2009) reported in each study was related to the frequency of elevated mRNA abundance during hypoxia (pooling tissues and sampling intervals for a given HIFA transcript). For HIF1A, there was a strong, non-linear, negative relationship between the number of essential MIQE criteria and the frequency of reported increases in mRNA abundance (p = 0.0032; Figure 1F; Supplementary Table S8). Studies that reported more details were less likely to find increased HIF1A mRNA compared to those reporting fewer of these details. Similar results were generated for HIF2A (p = 0.127; Supplementary Figure S3A; Supplementary Table S9) and HIF3A (p = 0.0547; Supplementary Figure S3B; Supplementary Table S10), although they were not individually significant due to a much smaller number of observations.

Some of the MIQE criteria that were under-reported in the literature include assessment of RNA integrity prior to reverse transcription, linearity of calibration curves, and justification of reference genes. The last is especially important because many of the “housekeeping” genes used to standardize the levels of a gene of interest can be influenced by hypoxia (e.g., GAPDH). If the abundance of a reference gene is lower during hypoxia, then levels of HIFA mRNA will be overestimated. In addition, given the above-mentioned diversity of HIFA paralogs among fishes, the targets of pPCR must be unambiguously identified. Unfortunately, no consensus exists on the nomenclature of teleost-specific duplicates of HIFA, which has led to the different names for a given gene (e.g., HIF2Aa) or uninformative names (e.g., HIF1-like). We adhered to the nomenclature of Townley et al. (2022) and Babin et al. (2024), which are the most complete phylogenetic surveys of HIFA in ray-finned fishes at the present time.

One limitation of this study is that we cannot exclude the possibility that the response of HIFA mRNA to hypoxia is species-specific. We included study as a random factor in our analyses, which would largely explain the same variation as species because all but one study (O’Brien et al., 2020) measured a single species. Nevertheless, the lack of consistent responses across studies, species, tissues, and experimental conditions suggests that attempts to use HIFA mRNA as a biomarker of hypoxic exposure (e.g., Zhang et al., 2009; Thomas and Rahman, 2009; Froehlich et al., 2015) must be coupled with careful laboratory experiments that adhere to guidelines for conducting and reporting qPCR. Moreover, those results might be useful only for that species. A second limitation is that our analyses considered only author-reported significant differences, rather than the magnitude of the changes in HIFA mRNA abundance. Considering the great variability in the magnitude of those changes across studies, however, it is unlikely that including effect sizes in our analyses would have led to dramatically different conclusions. Finally, we calculated the cumulative oxygen deficit for each sampling interval to represent the degree of hypoxic exposure. Because fish species are known to differ in their hypoxia tolerance (Verberk et al., 2022), relating the DO level of the laboratory exposures to some measure of each species’ hypoxia tolerance, for example, the critical oxygen tension (Ultsch and Regan, 2019), could be a better metric of the stress experienced. Unfortunately, there is not a single measure of hypoxia tolerance that is universally accepted (Wood, 2018) or available for all the species represented in the current analysis.