Machine-learning algorithm incorporating capacitated sperm intracellular pH predicts conventional in vitro fertilization success in normospermic patients

Abstract

Objective:

To measure human sperm intracellular pH (pH_i) and develop a machine-learning algorithm to predict successful conventional in vitro fertilization (IVF) in normospermic patients.

Design:

Spermatozoa from 76 IVF patients were capacitated in vitro. Flow cytometry was used to measure sperm pH_i, and computer-assisted semen analysis was used to measure hyperactivated motility. A gradient-boosted machine-learning algorithm was trained on clinical data and sperm pH_i and membrane potential from 58 patients to predict successful conventional IVF, defined as a fertilization ratio (number of fertilized oocytes [2 pronuclei]/number of mature oocytes) greater than 0.66. The algorithm was validated on an independent set of data from 18 patients.

Setting:

Academic medical center.

Patient(s):

Normospermic men undergoing IVF. Patients were excluded if they used frozen sperm, had known male factor infertility, or used intracytoplasmic sperm injection only.

Intervention(s):

None.

Main Outcome Measure(s):

Successful conventional IVF.

Result(s):

Sperm pH_i positively correlated with hyperactivated motility and with conventional IVF ratio (n = 76) but not with intracytoplasmic sperm injection fertilization ratio (n = 38). In receiver operating curve analysis of data from the test set (n = 58), the machine-learning algorithm predicted successful conventional IVF with a mean accuracy of 0.72 (n = 18), a mean area under the curve of 0.81, a mean sensitivity of 0.65, and a mean specificity of 0.80.

Conclusion(s):

Sperm pH_i correlates with conventional fertilization outcomes in normospermic patients undergoing IVF. A machine-learning algorithm can use clinical parameters and markers of capacitation to accurately predict successful fertilization in normospermic men undergoing conventional IVF.

One in eight couples worldwide are infertile (1), and in approximately half of cases, a male factor contributes to or is the primary cause of infertility (2). To help such couples conceive, assisted reproductive technology providers perform either conventional in vitro fertilization (IVF) or intracytoplasmic sperm injection (ICSI). In conventional IVF (also known as conventional insemination), spermatozoa from men with normal semen parameters (by World Health Organization [WHO] criteria regarding number, morphology, and motility) are incubated overnight with oocytes in conditions that mimic the female reproductive tract (3, 4). These conditions cause spermatozoa to capacitate, after which they exhibit hyperactivated motility and can undergo the acrosome reaction, both of which are essential for spermatozoa to penetrate the zona pellucida and fertilize an oocyte (4). Despite normal semen parameters, some semen samples fail to fertilize via conventional insemination (5). One explanation for this phenomenon is that these samples are not able to capacitate.

If men do not have normal semen parameters, the American Society for Reproductive Medicine guidelines state that ICSI is a safe and effective treatment (6). In ICSI, a single morphologically normal, noncapacitated sperm is injected into an oocyte. Although ICSI was developed to allow couples with a severe male factor to use autologous sperm to conceive (4), its use has increased in the United States over the past two decades. However, ICSI use varies by region, and its increased use has not correlated with an increase in medical indications or effectiveness (7). Additionally, some studies have reported that implantation rates and clinical pregnancy rates are higher with conventional IVF than with ICSI in the setting of non–male factor infertility (8). These findings suggest that ICSI is overused (9) and that clinics choose ICSI over conventional IVF for many normospermic patients (6). According to a 2020 statement from the American Society for Reproductive Medicine, ICSI is recommended in cases of non–male factor infertility when there is a history of previous conventional insemination failure; preimplantation genetic testing is being used; oocytes matured in vitro; and cryopreserved oocytes are used (6). Although ICSI is associated with a lower rate of total failed fertilization (0%–2% vs. 4%–16.7%) (10) and often higher fertilization rates (8) than conventional IVF, ICSI is invasive, more expensive, and more time-consuming than conventional IVF.

One tool designed to predict fertilization capacity and thus choose between conventional IVF and ICSI is the Cap-Score (11), which assesses capacitation by measuring localization of the ganglioside GM1 in human sperm. Although this tool can predict clinical pregnancy, it has been difficult to implement clinically because the sperm assessed cannot be used for treatment, so patients must provide multiple semen samples (11). Other methods of predicting fertilization ability, such as exposing sperm to animal oocytes (12) or human oocytes that failed to fertilize (13), are technically challenging. Thus, new methods are needed to choose between conventional IVF and ICSI for cases in which the male is normospermic.

To address this challenge, we recently developed a flow cytometric technique that can measure human sperm membrane potential (14). In many species, sperm membrane potential becomes hyperpolarized (increased net negative charge inside the membrane) during capacitation (reviewed in (15)). Moreover, hyperpolarization is required for human sperm to fertilize an oocyte (14–16). We found that sperm membrane potential correlated with hyperactivated motility, ability to undergo the acrosome reaction, and conventional IVF success in normospermic patients (14). Here we focused on another physiological component of sperm capacitation, intracellular pH (pH_i). As spermatozoa travel through the female genital tract, they are exposed to fluid that contains bicarbonate and thus an alkaline pH. As a result of this environment and the action of signaling molecules and ion channels, sperm pH_i increases (17–19).

However, whether this pH_i increase is required for human sperm to fertilize an oocyte is unknown. Additionally, spectrophotometric methods to measure human sperm pH_i (18, 20, 21) require a large number of spermatozoa and long periods of irradiation, leading to accumulation of reactive oxygen species, cell death, and loss of motility (22). To overcome these challenges, we developed a new flow cytometric method to measure absolute value of human sperm pH_i. The purpose of our study was to use this method to measure human sperm pH_i and develop a machine-learning algorithm to predict successful conventional IVF in normospermic patients.

MATERIALS AND METHODS

Human Samples and Ethical Statement

This study was approved by the Washington University in St. Louis Institutional Review Board. Men provided semen samples via masturbation after 2–5 days of abstinence in private collection rooms at the Washington University Fertility and Reproductive Medicine Center. Only samples that met the WHO 2010 criteria for normal semen parameters (ejaculated volume ≥ 1.5 mL, sperm concentration ≥ 15 million/mL, motility ≥ 40%, progressive motility ≥32%, normal morphology ≥ 4%) (3) were included in this study.

Couples undergoing IVF were included if their oocytes were subjected to split insemination (half ICSI, half conventional IVF) or conventional IVF only. Couples were excluded if they had known male factor infertility as indicated by previous semen analysis, if they used frozen sperm, if they used ICSI only, or if semen parameters measured on the day of egg retrieval did not meet WHO 2010 parameters for normospermia described above. Of a total of 615 fresh cycles undertaken during the study period, 274 cases (45%) used conventional IVF. Of the 80 men contacted during the study period, 76 consented for the study. On the day of oocyte retrieval, male partners gave written informed consent. We also obtained deidentified sperm samples from 10 men who had previously fathered a child and were undergoing semen analysis.

Human Sperm Capacitation

Ejaculated semen from IVF patients was liquified for 30 minutes at room temperature. Spermatozoa were washed twice, and the clean sperm pellet was overlayed with 200–500 μL of Quinn’s Advantage capacitating and fertilization media (CooperSurgical) supplemented with 3 mg/mL human serum albumin (CooperSurgical). The spermatozoa were allowed to swim up for 1 hour at 37°C and 6% CO₂. A portion of the highly motile sperm recovered after swim-up was used to inseminate oocytes. The leftover sperm samples were capacitated overnight (18 hours) at 37°C and 5% CO₂ and subsequently used for analysis.

Conventional IVF and ICSI

For conventional IVF, ~100,000 spermatozoa/mL were incubated overnight (18 hours) with oocytes with intact zona pellucida and cumulus cells. The fertilization ratio was calculated as the number of normally fertilized mature oocytes with two pronuclei divided by the total number of mature oocytes. For ICSI, oocytes were cultured for 2–3 hours, and then the cumulus cells were removed by treatment with recombinant human hyaluronidase (Cumulase, CooperSurgical). Only mature (metaphase II) oocytes were injected with sperm.

Determination of Sperm pH_i by Flow Cytometry

High potassium buffered solution contained 1.2 mM MgSO₄, 1.6 mM CaCl₂, 23.8 mM HEPES, 2.78 mM glucose, 3.38 mM sodium pyruvate, and 120 mM KCl. The KCl concentration was chosen to reflect the reported intracellular K⁺ concentration (23, 24). This solution was divided into aliquots, and the pH was adjusted by adding NaOH. The pH of the high potassium buffered solution was measured with each experiment at room temperature with a Denver Instrument ultrabasic pH meter.

After 18 hours of incubation in Quinn’s Advantage capacitating media (the same incubation time that was used in the clinic for incubating sperm in the presence of the oocytes), sperm samples were centrifuged at 400 g for 5 minutes and resuspended in 0.1 μM BCECF-AM (Sigma). The samples were protected from light, incubated at 37°C for 10 minutes, and divided into aliquots. Aliquots were washed, centrifuged at 400 g for 5 minutes, and resuspended in noncapacitating Human Tubal Fluid (HTF) media with 25 mM NaHCO₃ (tested sample) or high potassium buffered solutions at pH 6.3, 6.5, 7.0, 7.4, or 8.0. Before recording, 5 μM nigericin (Cayman Chemical) was added to the aliquots suspended in known high potassium buffered solutions to equilibrate intracellular and extracellular pH, and these samples were used to create a pH calibration curve (24).

Data were recorded as individual cellular events on a FACSCanto II TM cytometer (Becton Dickinson). Forward scatter area and side scatter area fluorescence data were collected from 20,000 events per sample. The BCECF-AM signal was detected by using the filter for fluorescein isothio-cyanate (533/530 nm) and excited with a blue laser (488 nm). BCECF-AM is a vital dye that is only incorporated in cells with active esterases and is present in propidium iodide–negative cells and thus indicates viability (25). Only data from single BCECF-positive spermatozoa were analyzed with FACS Diva and FlowJo X 10.0.7 software. Sperm pH_i was calculated by linearly interpolating the median of the histogram of BCECF fluorescence of the unknown sample to the calibration curve. Depending on the amount of sample received, pH_i was calculated in up to three replicates per sample. The average standard deviation between triplicates of 24 samples was 0.05 units of pH.

Determination of Sperm Membrane Potential by Flow Cytometry

Membrane potential was measured with the voltage-sensitive dye DiSC₃(5) as previously described (14). Briefly, after BCECF-AM incubation, 63 sample aliquots (not the same aliquots as those in which pH_i was measured) were resuspended in noncapacitating HTF media with 25 mM NaHCO₃. Then 5 nM DiSC₃(5) was added, and data were recorded as individual cellular events on a FACSCanto II TM cytometer. Forward scatter area and side scatter area data were collected from 20,000 events per recording. Only data from single, viable (BCECF-positive) spermatozoa were collected. DiSC₃(5) fluorescence was detected with the filter for allophycocyanin (585/40 nm). Recordings were initiated after reaching steady-state fluorescence (1–3 minutes).

Computer-Assisted Sperm Analysis

Aliquots (3 μL) of sperm suspension were placed into a 20 μm Leja standard count four-chamber slide, prewarmed at 37°C, and analyzed on a Hamilton-Thorne digital image analyzer (HTR-CEROS II v.1.7; Hamilton-Thorne Research). Phase alignment was checked, and the settings were as described (14). The criteria to define hyperactivated spermatozoa were curvilinear velocity > 150 μm/second, lateral head displacement > 7.0 μm, and linearity coefficient < 50% (26). Percentage of hyperactivated motility was defined relative to total motility.

Statistical Analysis

Data are expressed as mean ± standard error of the mean. P < .05 was considered statistically significant. To achieve 80% power with a significance level of .05, a sample size of 58 was needed to detect a significant correlation between pH_i and conventional IVF ratio and between pH_i and hyperactivated motility. Statistical analysis was performed with GraphPad Prism version 6.01 (GraphPad Software). The D’Agostino-Pearson test was used to determine whether values were distributed in a Gaussian fashion. Parametric or nonparametric comparisons were performed as dictated by data distribution.

Machine-Learning Algorithm

A gradient-boosted machine-learning algorithm was trained to identify patients who would have successful conventional IVF, defined as a fertilization ratio greater than 2/3 or 0.66 (number of fertilized oocytes [2 pronuclei]/number of mature oocytes). This cutoff was chosen in accordance with the Vienna Consensus, which states, “The fertilization ratio with conventional IVF is expected to be around 67%, with a range of reported values of 53–81%” (27).

Of the 76 total couples, the data were randomly split into training (75%, n = 58) and testing (25%, n = 18) data. To determine which clinical variables were most important in the model, feature selection was performed by using recursive feature elimination (rfe package in R 3.3.0). After an initial optimization pass was made, models were bootstrapped 500 times to assess reproducibility in their predictive power. All data in Table 1 except semen parameters (semen volume, semen sperm concentration, semen sperm motility, and semen sperm progressive motility) were included in the algorithm.

Table 1.

Demographics.

Characteristic	Unsuccessful IVF (fertilization rate <0.66) n = 31	Successful IVF (fertilization rate >0.66) n = 45	P value
Female partner age	32.7 ± 4.1	34.7 ± 4.0	.04
Female partner race			.60
Asian	1 (3.33)	0
Black	1 (3.33)	2 (4.88)
Hispanic	0	1 (2.44)
Southeast Asian	1 (3.33)	4(9.76)
White	27 (90.00)	34 (82.93)
Female partner BMI, kg/m2	26.2 ± 4.7	27.2 ± 6.0	.46
Gravida	0.5(0, 1)	1 (0, 2)	.07
Parity term	0 (0, 0)	0 (0,1)	.12
Parity preterm	0 (0, 0)	0 (0, 0)	.64
Parity abortions	0 (0, 1)	0 (0, 1)	.29
Parity living	0 (0, 0)	0 (0, 1)	.01
Peak estradiol, pg/mL	2,156.33 ± 1,195.97	2,279.90 ± 1,335.91	.68
Male partner age	34.1 ± 5.3	35.3 ± 4.3	.26
Male partner race			.45
Black	0 (0)	2 (4.44)
Hispanic	0	1 (2.22)
Southeast Asian	1 (3.23)	4 (8.89)
White	30 (96.77)	38 (84.44)
Male partner BMI, kg/m2	29.4 ± 6.5	28.8 ± 3.9	.66
Male partner tobacco use	7 (22.58)	5(11.90)	.22
Male partner marijuana use	5(16.13)	5 (12.50)	.74
Fathered pregnancy with current partner	9(29.03)	18(40.91)	.29
Fathered pregnancy with different partner	4 (12.90)	5(11.36)	1.00
No. of mature oocytes retrieved	8 (5, 12)	11 (7, 16)	.15
Conventional IVF rate	0.43 ± 0.22	0.86 ± 0.11	< .01
Semen volume, mL	2.736 ± 0.1952	2.596 ± 0.183	.6071
Semen sperm concentration, M/mL	53.34 ± 4.791	55.73 ± 5.747	.8009
Semen sperm motility, %	67.94 ± 2.903	67.97 ± 2.665	.9892
Semen sperm progressive motility, %	63.19 ± 2.754	61.29 ± 2.705	.6293
Sperm concentration, 18 h, M/mL	13.93 ± 9.01	12.97 ± 10.83	.68
Sperm motility, 18 h, %	68.35 ± 25.37	59.69 ± 27.33	.17
Sperm progressive motility, 18 h, %	46.0 ± 24.42	40.24 ± 25.91	.34
Sperm hyperactivated motility, %	4.72 ± 4.46	5.33 ± 4.57	.57
Sperm linearity coefficient, %	51.57 ± 11.02	51.92 ± 10.80	.89
Sperm lateral head displacement, μm	4.52 ± 1.31	4.38 ± 1.03	.59
Sperm curvilinear velocity, μm/second	88.36 ± 30.33	85.81 ± 26.67	.70
Sperm pHi	6.80 ± 0.16	6.91 ± 0.16	< .01
Sperm membrane potential, mV (n = 63)	−43.46 ± 12.91 (n = 25)	−56.77 ± 16.48 (n = 38)	< .01

Note: BMI = body mass index; IVF = in vitro fertilization.

RESULTS

Absolute Values of Human Sperm pH_i Can Be Determined by Flow Cytometry

This study included 76 patients with idiopathic normospermic infertility who met our inclusion and exclusion criteria. Of these, 31 had unsuccessful conventional IVF (IVF rate < 0.66) and 45 had successful IVF (IVF rate > 0.66). The patients and partners in the two groups were demographically similar except that female partners were older and the couples had more living children in the successful IVF group than in the unsuccessful IVF group (Table 1). Male participants were between 26 and 48 years of age.

Flow cytometry has been used to measure the pH_i of many cell types and has been used clinically to sort spermatozoa on the presence or absence of a Y chromosome (28, 29). To determine whether flow cytometry could be used to measure the absolute value of sperm pH_i, we incubated capacitated sperm from normospermic patients undergoing IVF with the pH-sensitive fluorescent probe BCECF-AM, which readily passes through the plasma membrane and fluoresces after it is cleaved by esterases in live cells (24). Figure 1A–1C shows gating of single, live human sperm. We created a pH calibration curve by incubating sperm in high potassium solutions at known pH and used nigericin, a proton ionophore, to equilibrate extracellular and intracellular pH (Fig. 1D). To determine the pH_i of a sperm sample, we performed a linear regression of the calibration curve (Fig. 1E) and interpolated the median BCECF-AM fluorescence from the calibration curve (Fig. 1D, red histogram, and Fig. 1E, red cross). Using this method, we obtained pH_i values between 6.54 and 7.25.

FIGURE 1.

Determination of sperm intracellular pH (pH_i) by flow cytometry. (A) Forward scatter area (FSC-A) and side scatter area (SSC-A) were used to gate human sperm. (B) SSC-A and side scatter height (SSC-H) were used to exclude non-single cells. (C) BCECF-AM fluorescence was used to gate on viable cells. (D) Histograms of BCECF-AM fluorescence from sperm incubated in known high potassium pH buffered solutions and nigericin to equilibrate intracellular and extracellular pH. An example of an unknown pH_i value is shown in red. (E) Mean BCECF-AM fluorescence values from the histograms in panel D were plotted, and a linear regression of the calibration curve was calculated. An example interpolated unknown pH_i value corresponding to the red histogram in panel D is indicated with a red X; r² = 0.9908. (F) After addition of 20 mM NH₄Cl, average pH_i increased by 0.59 (6.78 ± 0.15 vs. 7.37 ± 0.15, n = 5, P = .0008 by paired t test).

We performed two experiments to validate our method. First, we added 20 mM NH₄Cl, which increases spermatozoa pH_i by ~0.50 units (21). In our assay, addition of 20 mM NH₄Cl significantly increased the pH_i of five samples by ~0.59 units (Fig. 1F). Second, we measured the fluorescence of sperm incubated in pH buffered solutions before and after adding nigericin. As expected, addition of nigericin decreased the pH_i of sperm resuspended at pH 6.3 and increased the pH_i of sperm resuspended at pH 8.0 (Supplemental Fig. 1A and 1B). We concluded that this flow cytometry method accurately measured sperm pH_i.

Sperm pH_i Correlates with Hyperactivated Motility and Conventional IVF Success

We used computer-assisted sperm analysis to measure hyperactivated motility and used our flow cytometry assay to measure pH_i in the same samples. The percentage of sperm that underwent hyperactivated motility positively correlated with pH_i values (Fig. 2A). Hyperactivated motility was defined by curvilinear velocity >150 μm/second, linearity coefficient <50%, and lateral head displacement >7.0 μm on each sperm (26). Curvilinear velocity (Fig. 2B) and linearity coefficient (Fig. 2D), but not lateral head displacement (Fig. 2C), correlated with pH_i.

FIGURE 2.

Sperm intracellular pH (pH_i) correlates with sperm hyperactivated motility in normospermic in vitro fertilization patients. Plots show correlations between sperm pH_i and (A) hyperactivated motility (HA%, Spearman r² = 0.3027, P = .0079), (B) curvilinear velocity (VCL [μm/second], Pearson r² = 0.2770, P = .0154), (C) lateral head displacement (ALH [μm], Pearson r² = 0.1899, P = .1003, and (D) linearity coefficient (LIN%, Spearman r² = 0.3012, P = .0052). In all panels, n = 76.

In analysis of data from all 76 patients, pH_i significantly differed between those with unsuccessful and successful IVF (6.80 ± 0.16, n = 31 vs. 6.91 ± 0.16, respectively, n = 45; P < .01, Table 1). Additionally, pH_i correlated with conventional IVF ratio (Spearman r² = 0.2691, P = .0188, Fig. 3A). Given that ICSI bypasses capacitation, we were not surprised to find that sperm pH_i did not correlate with fertilization success in ICSI (n = 38, Spearman r² = −0.2776, P = .0915; Fig. 3B). We also performed our assay on sperm from normospermic fertile men that had previously fathered a pregnancy and were undergoing semen analysis. In this cohort, the average pH_i of capacitated sperm was 6.89 (range, 6.70–7.44, n = 10), consistent with the value we found in the successful IVF group.

FIGURE 3.

Capacitated sperm intracellular pH (pH_i) correlates with fertilization ratio in conventional in vitro fertilization (IVF), and a machine-learning algorithm including sperm pH_i predicts conventional IVF success. (A) Correlation between pH_i of capacitated sperm and conventional IVF ratio. Fertilization ratio was calculated as number of zygotes with 2 pronuclei/number of mature oocytes (metaphase II oocytes); n = 76, Spearman r² = 0.2691, P = .0188. (B) Correlation between sperm pH_i and intracytoplasmic sperm injection fertilization ratio from the same patients in panel A undergoing split insemination; n = 38, Spearman r² = −0.2776, P = .0915. (C) The predictor was trained on data from 58 couples and tested on the remaining independent set of data. In receiver operator characteristic analysis of the prediction algorithm, the area under the curve for predicting successful conventional IVF was 0.831. The algorithm had an accuracy of 0.833, sensitivity of 0.857, and specificity of 0.818. The two parameters most predictive of conventional IVF success were sperm membrane potential and pH_i.

A Machine-Learning Algorithm Can Predict Conventional IVF Success

We developed a gradient-boosted machine classifier that used clinical data collected from 58 randomly selected couples during the IVF cycle. The classifier was also provided parameters related to sperm capacitation including hyperactivated motility, pH_i, and membrane potential (Table 1). We trained the algorithm to predict the likelihood of >66% fertilization in conventional IVF. We then tested the algorithm on data from 18 couples. The final algorithm successfully predicted fertilization outcome in the majority of cases. In the receiver operating characteristic analysis, the algorithm performed with a mean accuracy of 0.72 (95% confidence interval [CI], 0.61–0.83), a mean area under the curve (AUC) of 0.81 (95% CI, 0.69–0.93), a mean sensitivity of 0.65 (95% CI, 0.45–0.85), and a mean specificity of 0.80 (95% CI, 0.62–0.98; Fig. 3C). To determine which clinical variables were most important in the model, we used recursive feature elimination. After an initial optimization pass, models were bootstrapped 500 times to assess reproducibility in their predictive power. To select relevant clinical factors and avoid data overfit, variable importance was assessed by using accuracy loss, or the decrease in accuracy over multiple models when each variable was intentionally left out. The most predictive variables in the models were membrane potential and pH, by a significant margin, followed by patient age, sperm hyperactivated motility, and peak estradiol in the female partner.

DISCUSSION

Here we present a new method for predicting conventional IVF success in normospermic men. First, we validated our flow cytometric method for measuring absolute values of sperm pH_i. Second, we showed that sperm pH_i values positively correlated with the percentage of sperm that underwent hyperactivated motility, a hallmark of capacitation, and with conventional IVF ratio. Third, our machine-learning algorithm combining clinical parameters with sperm pH_i, membrane potential, and hyperactivated motility accurately predicted conventional IVF success. These results suggest that the alkalization of sperm is a marker of capacitation and can be used to predict IVF success in normospermic infertile patients.

To our knowledge, this is the first study to use flow cytometry to measure the absolute value of human sperm pH_i. In contrast with other methods that measure the absolute value of human sperm pH_i such as spectrophotometry, our flow cytometry tool uses fewer sperm (5 × 10⁵ vs. 5–15 × 10⁷) and less fluorescent probe (0.1 μM vs. 0.8–4 μM) (18, 20, 21) and avoids long periods of irradiation, which can cause accumulation of reactive oxygen species, cell death, and loss of motility (22). Although the sperm used for analysis in our method cannot be used clinically, given that such a small amount of sperm is needed for analysis, most patients will have adequate sample remaining for conventional IVF, ICSI, or both.

The pH_i values that we obtained were between 6.54 and 7.25, with a median of 6.85, and are in accordance with values obtained via other techniques (18, 21, 30). The pH_i values differed significantly between successful and unsuccessful IVF groups (Table 1). Although appearing small, this 0.11 pH unit difference is considerable, given that in human sperm, pH_i values increase by approximately 0.14–0.11 units (from approximately 6.70 to 6.84) during capacitation (18, 31, 32).

Our data corroborate previous findings in bull, equine, and human sperm that a more alkaline pH_i is associated with hyperactivated motility (18, 33, 34). The development of hyperactivated motility in human, bull, and mouse sperm is likely due to calcium influx through the CatSper channel (35). This channel increases its calcium current threefold in response to a 0.5-unit increase of pH in mice (36). The increase in pH_i during capacitation might activate CatSper channels and facilitate sperm hyperactivation (37–39). Thus, a positive correlation between pH_i and hyperactivated motility makes physiologic sense. Although it has long been known that sperm alkalization occurs during capacitation (40), which is necessary for oocyte fertilization, this is the first study to show a significant correlation between sperm pH_i and conventional IVF rate. As expected, given that capacitation is not required for ICSI, we found no correlation between pH_i and ICSI fertilization rates.

Many factors can affect the absolute value of sperm pH_i during capacitation. In addition to the extracellular pH, the pH_i homeostasis of human sperm is also controlled by the concentration of HCO₃⁻. Although the channel or transporter that is mainly responsible for the increase in pH_i during capacitation is not known for certain, the most likely candidate is the H⁺ channel Hv1, which is encoded by the HVCN1 gene and is activated by the fatty acid anandamide (41) However, to our knowledge, no variants in HVCN1 have been shown to correlate with impaired fertility. Finally, environmental exposures, diet, illness, or lifestyle experienced by a male partner may affect sperm pH_i, but such factors have not been rigorously explored.

We recently reported that the absolute value of sperm membrane potential, determined by flow cytometry, also correlates with fertilization ability (14). Together, these results are consistent with the idea that monitoring human sperm capacitation can be used to assess male factor fertility (42). Additionally, our machine-learning algorithm was able to accurately (mean AUC of 0.81; 95% CI, 0.69–0.93) predict which normospermic patients would be successful with conventional IVF. Thus, pH_i and membrane potential should be considered as markers to aid assisted reproductive technology clinicians in determining whether normospermic couples should undergo conventional IVF or ICSI.

Currently it is difficult to assess whether sperm can undergo capacitation. As noted above, the Cap-Score, animal egg penetration assays, and human zona penetration assays can be used to assess capacitation, but these methods are time-consuming and technically challenging (11–13). Moreover, the sensitivity and specificity of these assays are more variable than our assays of sperm pH_i and membrane potential. For example, the Cap-Score has sensitivity of 91.7% and specificity of 77.8%; the hamster egg penetration assay has sensitivity of 52%–100% and specificity of 0%–100%; and the human zona penetration assay has sensitivity of 75%–100% and specificity of 57%–100% (11, 12). The sensitivity and specificity as well as AUCs of our algorithm are quite good given what is a somewhat small training set and would be expected to improve with an increase in samples. In addition to a relatively high accuracy, the high specificity of the algorithm would allow for safe adoption into clinical practice. Being able to quickly and reliably measure markers of sperm capacitation would allow physicians to guide recommendations at the time of semen analysis. If a patient has an abnormal pH_i or membrane potential, a couple may be guided to pursue ICSI instead of starting with intrauterine insemination. Conversely, if sperm pH_i and membrane potential are within normal ranges, the couple could be guided to less invasive and less expensive options such as intrauterine insemination or conventional IVF.

Our study and method have a few limitations. First, flow cytometry is expensive and not used widely in clinical settings. Second, our findings may have limited generalizability because our patients were recruited from a single institution. Third, the machine-learning algorithm was trained and tested with a limited number of patients. Finally, our study includes couples with a variety of diagnoses. For example, the older women or women with a known oocyte factor (i.e., diminished ovarian reserve) in our study may have had suboptimal oocyte quality, resulting in low conventional fertilization rates. However, the machine-learning algorithm considers multiple clinical parameters including age of the partners, diagnosis, and oocyte yield. Future studies will be directed at developing high-throughput techniques that are less expensive than flow cytometry to measure sperm membrane potential and pH_i and validating our findings in a larger, more diverse population.

Conclusions

Current diagnostic tools for male factor infertility, mainly semen analysis, cannot quantify capacitation and have limited ability to predict success in conventional IVF. A machine-learning algorithm that includes clinical data and parameters related to the capacitating state of sperm could be used to improve and personalize IVF care for hundreds of thousands of couples each year by allowing clinicians to appropriately choose between IVF and ICSI for each couple.