Experimental CO₂ and temperature treatments: The spawning trial reared wild-caught juveniles to maturity at 20°C under duplicate control (~600 µatm CO₂) and high CO₂ conditions (~2,000 µatm CO₂), whereas the fecundity trial reared newly fertilized embryos to maturity under similar control vs. high CO₂ conditions (~350 µatm, ~2,200 µatm) crossed with two rearing temperatures (17°C, 24°C).

Experimental treatment tanks were continuously bubbled with mixes of air and 100% bone-dry CO₂ controlled by gas proportioners (ColeParmer^®). In control CO₂ treatments, metabolic acidification by the fish was offset by bubbling CO₂-stripped air (~50 µatm CO₂) via diffuser tubes into the rearing tanks (Murray and Baumann, 2020). All treatments maintained near 100% dissolved oxygen saturations (~8.1, 7.6, and 7.1 mg L⁻¹ at 17°C, 20°C, and 24°C, respectively). Salinity (~31 psu) was controlled via refractometer, temperature conditions were maintained by thermostats (Aqualogic^®) connected to submersible heaters or in-line chillers (DeltaStar^®), and target pH conditions were monitored daily using a handheld pH meter (Hach^® Intellical PHC281 pH electrode with HQ11D handheld pH/ORP meter, calibrated bi-weekly using two-point NIST buffers).

Offspring production and laboratory rearing: For the spawning trial, juveniles were caught on 25 September 2015 via beach seine (30 × 2 m) in Mumford Cove; a small seagrass dominated embayment connected to eastern Long Island Sound (41.32°N, 72.02°W). Individuals were transported to the Rankin Seawater Facility at the University of Connecticut, where approximately 100 individuals (mean TL = 4.5 cm) were randomly assigned to each of two 700-L replicate rearing tanks per CO₂ treatment and reared to maturity over the next five months at 20°C.

For the fecundity trial, spawning-ripe silversides were captured in Mumford Cove on 29 May 2018, transported to the laboratory, separated by sex, and held without food in 80-L aerated containers at ambient CO₂ and temperature conditions (17°C). The day after, 20 females and 27 males were strip-spawned together using established protocols (Baumann et al., 2018; Murray and Baumann, 2020). The fecundity trial was terminated before spawning occurred at 317-325 dph, when all fish were euthanized with MS-222 (Tricaine-S).

Trait measurements: For the spawning trial, spawning in each tank began when silversides were provided bundled threads of green yarn suspended mid-tank. Yarns were checked every 48h, deposited embryos counted, and egg diameters measured in a subsample (n = 15) via calibrated pictures (4× magnification, ImagePro Premier, v9.0, Media Cybernetics^®). The spawning trial was terminated after 60 days, which is approximately the extent of silverside spawning season in the wild (Pringle and Baumann, 2019). Surviving fish were euthanized, measured for TL (0.1 mm) and wet weight (wW, 0.01g), and their sex determined by gonad inspection under a stereo microscope (Nikon SMZ-1000). The total number of embryos spawned was divided by the cumulative wet weight of females in each duplicate CO₂ treatment to obtain a mass-specific estimate of embryo production (n_emb g female⁻¹). This necessarily assumed that all females actually participated in spawning.

For the fecundity trial, euthanized fish were measured for TL and wW, followed by dissection for sex identification and gonad removal. Re-weighing yielded the gonad-free body weight (wW_ng), which was subtracted from wW to yield gonad weight (wW_g) and the gonado-somatic index (GSI = 100*wW_g/wW_ng). Male gonads were discarded; female gonads were preserved in 5% buffered formaldehyde/freshwater solution. One ovary lobe per female was saved for histological analyses; the other was used to quantify all secondary growth (i.e., vitellogenic) oocytes. The ovary lobe was divided into anterior, middle, and posterior sections that were each gently teased apart so that oocytes evenly distributed between four rectangular 2 mL wells. Each well was photographed (3× magnification) and analyzed via ImageJ (v1.52a, National Institute of Heath) with the ObjectJ plug-in (v1.04t, University of Amsterdam) and a customized macro to count and measure the diameter of all oocytes. The counted total number of oocytes per lobe was doubled to obtain each female’s potential fecundity (F_Pot, oocytes female⁻¹) and relative fecundity (F_Rel = F_Pot*wW⁻¹, oocytes g female⁻¹). In addition, we averaged the size of the 25 largest oocytes per female (~99% percentile) as a measure of maximum oocyte size.

The other ovary lobe from a subset of 17 females at control CO₂ treatments (n_17°C = 6, n_24°C = 11) and 20 females at high CO₂ treatments (n_17°C = 8, n_24°C = 12) were embedded in paraffin, sectioned, placed on glass slides, and stained with H&E (Horus Scientific, Worcester, MA). Slides were then photographed (4× magnification) and six oocyte developmental stages identified: primary growth (PG), cortical alveolar (CA), early vitellogenesis (V1), late vitellogenesis (V2), nuclear migration (NM), and hydrated (H) oocytes (following Ganias et al., 2004; Hyle et al., 2014; Press et al., 2014). We also checked for any postovulatory follicles that would indicate the beginning of spawning. Up to 56 oocytes per developmental stage and female were measured (mean = 13.5). Histological preparation deformed most oocytes to ellipses; therefore we first calculated the area of each ellipse from its major and minor axis and then inferred the oocyte diameter (=size) from a circle with the same area. For each female, we determined the stage-specific median and maximum oocyte size, which were then compared across treatments.

Sampling and analytical procedures: With the replication unit being each 700-L tank (instead of each female), the spawning trial thus had duplicates for each CO₂ level (single temperature), while the fecundity trial had single replicates for each CO₂ × temperature combination. Although the 24°C treatments ended with two equal density groups per CO₂ level, these were not true replicates, because fish were only separated after 245 dph. Given that both groups were statistically identical with respect to TL and wW (t-tests, p > 0.05), they were pooled for all subsequent analyses. Lacking the 3+ replicates that means-based statistical approaches require (e.g., general linear models), we instead compared trait distributions between CO₂ and temperature treatments using non-parametric Kruskal-Wallis tests. We focused on female traits here, because reproductive investment in fish females is more costly and more tightly linked to body size and fitness than in males (for male distributions, see supplementary material). We examined female TL and wW (spawning + fecundity trials), number and size of spawned embryos (spawning trial), GSI, potential and relative fecundity, maximum oocyte size, as well as stage-specific oocyte sizes (fecundity trial). Three females were removed from analysis, because they had either no discernible or only immature oocytes in their gonads. Statistical analyses were performed in R v3.6.1 (R Core Team, 2019) and SPSS (V20, IBM).